Get 20M+ Full-Text Papers For Less Than $1.50/day. Subscribe now for You or Your Team.

Learn More →

Development and validation of a radiomics-based model to predict local progression-free survival after chemo-radiotherapy in patients with esophageal squamous cell cancer

Development and validation of a radiomics-based model to predict local progression-free survival... Purpose: To develop a nomogram model for predicting local progress‑free survival (LPFS) in esophageal squamous cell carcinoma (ESCC) patients treated with concurrent chemo‑radiotherapy (CCRT ). Methods: We collected the clinical data of ESCC patients treated with CCRT in our hospital. Eligible patients were randomly divided into training cohort and validation cohort. The least absolute shrinkage and selection operator (LASSO) with COX regression was performed to select optimal radiomic features to calculate Rad‑score for predicting LPFS in the training cohort. The univariate and multivariate analyses were performed to identify the predictive clinical factors for developing a nomogram model. The C‑index was used to assess the performance of the predictive model and calibration curve was used to evaluate the accuracy. Results: A total of 221 ESCC patients were included in our study, with 155 patients in training cohort and 66 patients in validation cohort. Seventeen radiomic features were selected by LASSO COX regression analysis to calculate Rad‑ score for predicting LPFS. The patients with a Rad‑score ≥ 0.1411 had high risk of local recurrence, and those with a Rad‑score < 0.1411 had low risk of local recurrence. Multivariate analysis showed that N stage, CR status and Rad‑score were independent predictive factors for LPFS. A nomogram model was built based on the result of multivariate analy‑ sis. The C‑index of the nomogram was 0.745 (95% CI 0.7700–0.790) in training cohort and 0.723(95% CI 0.654–0.791) in validation cohort. The 3‑ year LPFS rate predicted by the nomogram model was highly consistent with the actual 3‑ year LPFS rate both in the training cohort and the validation cohort. Conclusion: We developed and validated a prediction model based on radiomic features and clinical factors, which can be used to predict LPFS of patients after CCRT. This model is conducive to identifying the patients with ESCC benefited more from CCRT. Keywords: Chemo‑radiotherapy, Esophageal squamous cell cancer, Radiomics, LPFS, Nomogram Introduction Esophageal cancer (EC) is the sixth common malignant *Correspondence: luohesan@163.com Department of Radiation Oncology, Shantou Central Hospital, tumors in China with an estimated 477.9 thousand new Shantou 515000, Guangdong, China cases, accounting for half of the new esophageal cancer Full list of author information is available at the end of the article © The Author(s) 2021. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/. The Creative Commons Public Domain Dedication waiver (http:// creat iveco mmons. org/ publi cdoma in/ zero/1. 0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. Luo et al. Radiat Oncol (2021) 16:201 Page 2 of 11 worldwide [1, 2]. In China, approximately 90% of the Patients and methods patients with esophageal cancer are histologically diag- Patients’ cohort nosed as esophageal squamous cell carcinomas (ESCC) We collected the clinical data of patients diagnosed as which is different from esophageal adenocarcinoma ESCC and received CCRT in our hospital during the (EAC) in risk factors and prognosis [3]. Most patients period from January 2013 to December 2015. Patients with locally advanced ESCC lost the opportunity for sur- were excluded if they met the exclusion criteria as follows: gery, and concurrent chemo-radiotherapy (CCRT) has (1) patients received esophagectomy and preoperative been recommended as a standard treatment [4]. How- or postoperative adjuvant radiotherapy; (2) patients had ever, more than half of patients treated with standard distant metastatic disease; (3) patients received low-dose dose CCRT eventually developed local recurrence or dis- (< 50  Gy) palliative radiotherapy; (4) clinicopathological tant metastases and succumbed to this disease [5, 6]. A information of the patients was incomplete; (5) patients individual CCRT strategy with escalated radiation dose were diagnosed as esophageal fistula before treatment; based on PET-CT would benefit the patients with high (6) poor visualization quality due to image artifacts or tumor burden and risk of recurrence [7, 8]. To facilitate the tumor was too small to be recognized on CT images; a individual CCRT strategy in an early stage, solid predic- (7) patients had other primary tumor; (8) patients died tive model for local recurrence and prognosis could play within three months after chemoradiotherapy. an important role. After multiple iterations, a total of 221 patients were For patients received CCRT, local and regional recur- randomly divided into two groups, with 155 patients rence is the most common failure pattern and pre-treat- in the training cohort and 66 patients in the validation ment clinical TNM staging is still the most commonly cohort. To improve the generalization property of the used system for prognosis prediction [9]. However, the result, multi-factors stratification was used to keep the currently used clinical TNM staging follows the same characteristics of sub cohort consistent with the whole criteria as pathological staging, which is based on imag- cohort. The process of patients’ enrollment and randomi - ing assessment of tumor size and surrounding invasion, zation were shown in Fig.  1. This study was approved by ignoring the information such as length and volume of the Institutional Committee of our hospital on Human esophageal cancer lesions. Recently, a series of clinico- Rights. Disease of the patients was staged according to pathologic biomarkers have been investigated and veri- the 8th edition of AJCC TNM classification for esopha - fied to be available for prediction of therapeutic response geal cancer [16]. and prognosis [10–12]. Radiomics is a new technique for image quantitative analysis about computed tomography Chemoradiotherapy protocol (CT) images, magnetic resonance (MR) images, posi- Radiotherapy was delivered daily to patients with three- tron emission tomography (PET) images, etc. [13]. Sev- dimensional conformal radiation therapy (3DRT) or eral studies demonstrated that radiomic features could intensity-modulated radiation therapy (IMRT) tech- potentially identify prognostic phenotype in patients nique using a Varian IX or Varian 23EX linear accelera- with EC. Yip et al. [14] suggested that a model combined tor in this study. The gross tumor volume (GTV) includes CT-based texture feature and esophageal maximal wall the esophageal cancer (GTVp) and the positive regional thickness assessment could predict the overall survival lymph nodes (GTVnd). The GTV was delineated on CT in EC patients treated with CCRT. Moreover, the model imaging according to barium esophagogram, endoscopic performed better than treatment response alone. Larue examination or PET imaging. The CTV was defined as et  al. [15] extracted out five radiomic features from CT the GTVp with 0.5–1 cm radial expansion and 2.5–3 cm image before chemoradiotherapy to describe the hetero- axial direction expansion or the GTVnd with 0.5–0.8 cm geneity of tumors and found that these five features could uniform expansion. The planning target volume (PTV) predict the 3-year survival rate of patients with EC after was defined as CTV with a 1  cm uniform expansion. A neo-chemoradiotherapy. However, most radiomic studies total prescribed dose of 50–72  Gy (median, 64  Gy) in included a small number of patients with EAC and ESCC. conventional fractionation was delivered to the patients. In this study, we explored the prognostic value of 3D Two cycles of platinum-based chemotherapy were radiomic features from pretreatment CT images of administered concurrently with radiotherapy. Sixty-one esophageal cancer patients and developed a model com- patients received TP (paclitaxel + cisplatin) chemo- bined radiomic features and clinical information to pre- therapy every three weeks, which consists of cisplatin 2 2 dict LPFS in patients with ESCC after CCRT. To evaluate (60  mg/m on Day 1) plus paclitaxel (135–180  mg/m the performance of the model, a validation cohort of on Days 1). One hundred and sixty patients received the patients were employed for validation. PF (cisplatin + fluorouracil) regimen every four weeks, L uo et al. Radiat Oncol (2021) 16:201 Page 3 of 11 Fig. 1 Flow chart of patients’ screening and allocation Luo et al. Radiat Oncol (2021) 16:201 Page 4 of 11 which consists of cisplatin (60 mg/m on Day 1) and fluo - dependence matrix, GLDM). The wavelet filter was used rouracil (750 mg/m /24 h on Days 1–4). in image pre-processing for texture features extraction. In all, for each VOI, 107 original features (Additional file  1: Table  S1) and 744 wavelet features (Additional Response evaluation file  1: Table  S1) were collected. Among the 107 original The response to chemo-radiotherapy was evaluated one features, there were 18 first order statistics features, 14 month after CCRT according to the criteria of short-term shape-based histogram features, 24 GLCM features, 14 response evaluation standard on esophageal cancer using GLDM features, 16 GLRLM features、16 GLSZM fea- CT images and barium esophagogram. According to the tures and 5 NGTDM features. Mathematical definitions response evaluation criteria, clinical response was classi- of these radiomic features have previously been described fied as complete response (CR), partial response (PR), no [18] and available at https:// pyrad iomics. readt hedocs. io/ response (NR), or progressive disease (PD). Patients who en/ latest/ featu res. html. were classified as CR by barium esophagogram and had the maximal esophageal wall thickness of ≤ 1.2  cm and Statistical analysis the volumes of residual lymph nodes of ≤ 1.0 cm on CT At the first, statistical analyses were performed with Chi- were finally defined as CR [17]. squared test or Fisher’s to assess the difference of the clinical characteristics between training cohort and vali- Radiomic feature extraction dation cohort. A p-value of < 0.05 was considered statisti- All patients were scanned using GE Lightspeed 64-slice cally significant. spiral CT (GE Medical systems, Milwaukee, Wis) before In the pre-processing of radiomic features, all the val- radiotherapy. CT image acquisition was performed ues of radiomic features were normalized using Z-score according to the following acquisition protocol: The CT normalization, which made features values lying within tube voltage was 120  kV and the tube current was 120 similar ranges and reduced the influence of large dis - mAs. Rack rotation time: 0.6  s; Detector collimation crete values. The intra-class correlation coefficient (ICC) parameters: 64 × 0.625  mm; field of view (FOV): 400- analysis was performed to evaluate the reproducibility of 500  mm; Matrix: 512 × 512; Layer thickness is 5  mm, each radiomic feature. Only the features with ICCs val- layer spacing is 5  mm. Contrast medium was injected ues ≥ 0.900 were selected for further analysis. Then, the with a high-pressure syringe at a flow rate of 3.0  ml/s least absolute shrinkage and selection operator (LASSO) (1–1.5  ml/kg, ioproxamine injection 300), followed by with COX regression was performed using R software 30 to 40 ml of normal saline for flushing, and late arterial version 3.6.2 (R Foundation for Statistical Computing, CT images were collected with a delay of 30 s. To reduce Vienna, Austria) to identify the features associated with the variability between images from different patients, all LPFS in the training cohort. The optimal parameter images were resampled to voxel of 1*1*1mm . lambda (λ) was chosen from the LASSO model using ten- 3D Slicer (version, 4.10.2, Stable Release) with radiom- fold cross-validation with the minimum partial likelihood ics extension was used for image segmentation to obtain deviance. Radiomic feature score (Rad score) for each volume of interest (VOIs). The primary tumor volume patient was built based on the LASSO COX regression (GTV) delineated by radiation oncologists for radiother- model in the training cohort. The LASSO COX regres - apy treatment planning design was defined as VOI for sion formula: radiomic features extraction. Any pixel with an attenua- tion of less than − 50 HU was excluded to avoid adjacent Rad score = β1X1 + β2X2 + β3X3 + ··· + βnXn air, fat, blood vessels and surrounding organs. Image seg- mentation was performed independently by a radiation In the above formula, X1, X2 … Xn are the different oncologist and another radiologist. To assess the repro- radiomic features identified by the LASSO COX regres - ducibility of the radiomic features extraction, tumor seg- sion model, and β1, β2 … βn are the regression coef- mentation was performed again two months later by the ficients of the corresponding features in the regression same radiologist in 30 randomly chosen patients. model. Pyradiomics V3.6.2 was used to extract radiomic fea- Univariate analysis was performed to identify the tures from delineated VOIs. Several categories of features potential prognostic factors associated with LPFS. Multi- were extracted from VOIs, including first order statistics variable COX regression analysis was performed to iden- features (IH, intensity histogram), shape-based histogram tify the independently predictors for LPFS. A nomogram features, and texture features (gray-level co-occurrence model combined Rad-score and clinical factors for pre- matrix, GLCM; gray-level size-zone matrix, GLSZM; dicting LPFS was developed and validated based on the gray-level run-length matrix, GLRLM; neighboring results of multivariable COX regression analysis using gray-tone difference matrix, NGTDM; and gray-level rms package and foreign package in R software. The L uo et al. Radiat Oncol (2021) 16:201 Page 5 of 11 predictive accuracy of the nomogram model was assessed of 1-year, 2-year and 3-year LPFS were 56.1%, 37.4% and using Calibration curve validation in both training cohort 32.1%, respectively (Fig. 2). and validation cohort. All the analyses were performed In order to develop and validate a radiomics-based with R software version 3.6.2. model for predicting LPFS of the patients, they were ran- domly divided into training cohort and validation cohort. There were 155 patients in the training cohort and 66 Results patients in the validation cohort. No significant differ - Patients’ characteristics ences (All p > 0.05) were found between the distribution A total of 221 ESCC patients who received chemoradio- of baseline characteristics in two cohorts, such as age, therapy in our hospital were eligible for further analysis gender, tumor location, T stage, N stage, clinical staging, in this study. Patients’ characteristics were summarized lactate dehydrogenase (LDH), neutrophil to 1ymphocyte in Table  1. The median follow-up time was 18.6 months. ratio (NLR), platelet to lymphocyte ratio (PLR) and CR By the end of the last follow-up, 153 patients developed ratio (33.5% in the training cohort vs 39.4% in the valida- local regional disease progression or died. The median tion cohort). Therefore, the two cohorts of patients were LPFS in the whole group was 13.7 months, and the rates comparable. Table 1 Comparison of patients’ characteristics between training cohort and validation cohort Variables Training cohort (n = 155) Validation cohort (n = 66) χ /t p Age (years), Mean ± SD 65.7147 ± 9.74 64.73 ± 10.16 0.678 0.499 Gender 0.342 0.559 Male 116 (74.8) 45 (68.2) Female 39 (25.2) 21 (31.8) Tumor location 5.814 0.121 Cervical 6 (3.9) 8 (12.1) Upper thoracic 34 (21.9) 16 (24.2) Middle thoracic 91 (58.7) 33 (50.00) Lower thoracic 24 (15.5) 9 (13.6) T stage 3.193 0.363 T1 2 (1.3) 0 (0) T2 11 (7.1) 9 (13.6) T3 66 (42.6) 27 (40.9) T4 76 (49.0) 30 (45.5) N stage 1.856 0.603 N0 20 (12.9) 13 (19.7) N1 70 (45.2) 28 (42.4) N2 55 (35.5) 22 (33.3) N3 10 (6.5) 3 (4.5) Clinical stage 3.152 0.369 I 2 (1.3) 0 (0) II 15 (9.7) 11 (16.7) III 88 (56.8) 37 (56.1) Iva 50 (32.3) 18 (27.3) Radiation dose, Median (range) 64 (60–66) 64 (60–66) − 0.920 0.358 LDH group 1.282 0.258 High 88 (56.8) 32 (48.5) Normal 67 (43.2) 34 (51.5) NLR, Median (range) 2.73 (1.96–3.71) 2.76 (2.00–3.63) − 0.448 0.654 PLR, Median (range) 137.78 (100.56–181.43) 138.87 (101.31–182.26) − 0.344 0.731 CR ratio 52 (33.5) 26 (39.4) 0.693 0.405 Rad‑score, Mean ± SD − 0.0289 ± 0.35 − 0.058 ± 0.538 0.474 0.636 Luo et al. Radiat Oncol (2021) 16:201 Page 6 of 11 The same result was found in the validation cohort (Fig. 4B, HR 1.997, 95% CI 1.070–3.728, p = 0.026). Development and validation of a predictive nomogram based on Rad‑score In order to develop a model to predict LRFS based on multiple factors, we performed univariate and multi- variate analyses to identify predictive factors for LPFS. Univariate analysis showed that the T stage, N stage, clinical stage and CR status were significantly associated with LPFS both in training cohort and validation cohort (Table  3). Multivariate analysis showed that N stage, CR status and Rad-score were independent predictive fac- tors for LPFS in ESCC patients after chemoradiotherapy Fig. 2 Kaplan–Meier curve of local‑progression free survival for all (Table  4). A nomogram model for predicting LPFS was patients built based on the result of multivariate analysis (Fig. 5A). As shown in Fig.  5 1-year, 2-year and 3-year LPFS prob- ability of every patient could be predicted based on the independent clinical characteristics and Rad-score. The Rad‑score building based on radiomic features C-index of the nomogram was 0.745 (95% CI 0.7700– LASSO-COX regression was used to screen out the 0.790) in training cohort and 0.723(95% CI 0.654–0.791) optimal radiomic features associated with LPFS of in validation cohort. the patients in the training cohort (Fig.  3A, B). As a Finally, we performed calibration curve to evaluate the result, seventeen radiomic features were screened accuracy of the nomogram model. As shown in Fig.  6, out (The features and their coefficients were listed in the 3-year LPFS rate predicted by the nomogram model the Table  2). The Rad-score was calculated as follows: based on Rad-score was highly consistent with the actual Rad-score = -0.104667846*or ig inal_f irstorder_Ske w- 3-year LPFS rate both in the training cohort and the vali- ness + 0.001161134*or ig in_g l sz m_Si zeZ oneNonUni- dation cohort. formityNormalized + 0.034339901*origin_glszm_Size- ZoneNonUniformity-0.017089976*origin_glszm_Low- GrayLevelZoneEmphasis + 0.062595767*wavelet-HLL_ Discussion glcm_Idn + 0.026703955*wavelet-HLL_firstorder_Maxi - Concurrent chemoradiotherapy (CCRT) is a radical m u m + 0 . 0 4 2 9 5 7 1 4 3 * w a v e l e t- H L L _ g l s z m _ S i z e Z o - treatment for patients with inoperable esophageal can- neNonUniformityNormalized + 0.017543973*wavelet- cer or refused surgery [4]. Many studies have shown that LHL_firstorder_TotalEnerg y + 0.003781538*w avele t- dose-escalation radiotherapy properly can improve the L H L_f irst or der_Ma xim um-0.007364328*w a ve le t- local control and survival of patients with ESCC [19–22]. LLH_gldm_SmallD ep endenceLowGrayLe velEmpha- Nevertheless, 30–50% of patients have local recurrence sis + 0.157807433*wavelet-LLH_glcm_DifferenceVari- within 3  years [23–25]. In our present study, we con- ance + 0.042028490*wavelet-LLH_glrlm_ShortRunHigh- structed a prediction model combined the clinical char- GrayLevelEmphasis-0.101981005*wavelet-LLH_ngtdm_ acteristics and CT radiomic features which can predict Coarseness-0.073958943*wavelet-HLH_gldm_SmallDe- the LPFS of patients after CCRT. With the help of this pendenceLowGrayLevelEmphasis + 0.051287394*wave- model, we can preliminarily judge the probability of LPFS let-HLH_firstorder_Maximum-0.055239045*wavelet- of patients and identify the patients benefit more from HHH_glcm_MaximumProbability-0.028958889*wavelet- CCRT. LLL_glcm_Imc2. Radiomics studies in esophageal cancer started rela- There was an optimal cutoff value of Rad score to tively late, and there are still few data about applying radi- divide the patients into two groups with different risk of omics analysis to evaluate the prognosis of esophageal local recurrence. As shown in Fig.  3C, the patients with cancer. Ganeshan et  al. [26] first analyzed the radiomic a Rad-score ≥ 0.1411 had high risk of local recurrence, features of CT before treatment in esophageal cancer and those with a Rad-score < 0.1411 had low risk of local patients and found that the radiomic features represent- recurrence. In the training cohort, the patients in the ing uniformity parameters were significantly different group with high risk of local recurrence had significantly between stage I/II and stage III/IV disease, which were shorter time of LPFS than those with risk of local recur- independent predictors of patients’ prognosis. Subse- rence (Fig. 4A, HR 2.882, 95% CI 1.926–4.313, p < 0.001). quently, Yip et al. [14] found that the tumor heterogeneity L uo et al. Radiat Oncol (2021) 16:201 Page 7 of 11 Fig. 3 Selection of radiomic features associated with LPFS using the LASSO COX regression model. A Coefficients profiles of radiomic features. The horizontal axis value is logλ, and the vertical axis value represent the coefficients of radiomic features. B The cross‑ validation curve. The horizontal axis value is logλ, and the vertical axis value is partial likelihood deviance. C The optimal cutoff of Rad‑score. Red lines or red dots represent patients at high risk of local recurrence and green lines or green dots represent patients at low risk of local recurrence. The optimal cutoff value is 0.1411, as shown in the vertical line in the figure could be represented by the change of CT radiomic fea- chemoradiotherapy, which was also supported in our tures before and after neoadjuvant treatment, which was study.- related to the prognosis and survival of patients. Larue Clinical TNM staging before treatment is still the most et al. [15] also found that five radiomic features extracted commonly used prediction system of prognosis for ESCC from CT before chemoradiotherapy could be used to patients treated with chemoradiotherapy. Combination describe the tumor heterogeneity and predict the 3-year of TNM staging and other prognostic factors can predict survival rate of patients after neoadjuvant chemoradio- the prognosis of patients more individually and accu- therapy and surgery with AUCs (AUC, area under the rately [27, 28]. Some studies have shown that the progno- receiver) of 0.69 in the training group and 0.61 in the vali- sis of patients who achieved CR after chemoradiotherapy dation group. All these studies suggested that radiomic was better than that of patients not CR [29, 30]. There - features played an important role in evaluating the prog- fore, CR after CCRT had become another important nosis of esophageal cancer and could be used to predict predictor for the prognosis of patients besides clinical the long-term survival of esophageal cancer patients after stages. In the present study, univariate analysis showed Luo et al. Radiat Oncol (2021) 16:201 Page 8 of 11 Table 2 Radiomics feature associated with LPFS selected by LASSO COX analysis Radiomics features Coefficients original_firstorder_Skewness − 0.104667846 origin_glszm_SizeZoneNonUniformityNormalized 0.001161134 origin_glszm_SizeZoneNonUniformity 0.034339901 origin_glszm_LowGrayLevelZoneEmphasis − 0.017089976 wavelet‑HLL_glcm_Idn 0.062595767 wavelet‑HLL_firstorder_Maximum 0.026703955 wavelet‑HLL_glszm_SizeZoneNonUniformityNormalized 0.042957143 wavelet‑LHL_firstorder_TotalEnergy 0.017543973 wavelet‑LHL_firstorder_Maximum 0.003781538 wavelet‑LLH_gldm_SmallDependenceLowGrayLevelEmphasis − 0.007364328 wavelet‑LLH_glcm_DifferenceVariance 0.157807433 wavelet‑LLH_glrlm_ShortRunHighGrayLevelEmphasis 0.042028490 wavelet‑LLH_ngtdm_Coarseness − 0.101981005 wavelet‑HLH_gldm_SmallDependenceLowGrayLevelEmphasis − 0.073958943 wavelet‑HLH_firstorder_Maximum 0.051287394 wavelet‑HHH_glcm MaximumProbability − 0.055239045 wavelet‑LLL_glcm_Imc2 − 0.028958889 Table 4 Multivariate analysis of prognostic factors associated with LPFS for patients with ESCC treated with chemoradiotherapy Variables Multivariate analysis p HR 95% CI p T stage 0.858 0.526–1.400 0.540 N stage 1.892 1.122–3.190 0.017 Clinical stage 0.627 0.313–1.258 0.189 Rad‑score 4.423 1.993–9.814 0.000 CR status 0.154 0.080–0.297 0.000 related to LPFS after CCRT. Moreover, Rad-score based on 17 radiomic features extracted from CT images before chemoradiotherapy was significantly related to LPFS of patients after chemoradiotherapy. Further multivariate analysis showed that Rad-score, N stage and CR after radiotherapy were independent predictors of patients’ Fig. 4 Kaplan–Meier survival curve of patients with high and low LPFS, while T stage and clinical stage had no statistical recurrence risk based on Rad‑score. A LPFS survival curve of patients significance in this multivariate analysis model, prob - in the training cohort: green line represents patients with low risk ably because Rad-score derived from the primary tumor of local recurrence and red represents patients with high risk of focus and had interactive effects with T stage and clinical local recurrence. The difference is significant between two groups, stage. We developed and validated a nomogram model p < 0.001. B LPFS survival curve of patients in the validation cohort: green line represents patients with low risk of local recurrence and based on the results of multivariate analysis. C-index red line represents patients with high risk of local recurrence. The and calibration curve were used to evaluate the perfor- difference is significant between two groups, p = 0.026 mance and prediction accuracy of the nomogram model. The C-index of the model was 0.745(95% CI 0.700–0.790) in the training cohort and 0.723(95% CI 0.654–0.791) that pre-treatment clinical T stage, N stage, clinical stage in the validation cohort, indicating high prediction and CR after radiotherapy were the prognostic factors L uo et al. Radiat Oncol (2021) 16:201 Page 9 of 11 Table 3 Univariate analysis of prognostic factors associated with LPFS in patients with ESCC treated with chemoradiotherapy Variables Training cohort p Validation cohort p HR 95% CI p HR 95% CI p Age 0.987 0.967–1.007 0.202 0.984 0.957–1.011 0.252 Gender 1.639 1.014–2.650 0.044 1.198 0.652–2.202 0.560 Tumor location 1.120 0.851–1.473 0.419 1.289 0.917–1.812 0.143 T stage 2.015 1.453–2.793 < 0.001 1.943 1.227–3.077 0.005 N stage 1.867 1.446–2.410 < 0.001 1.765 1.215–2.563 0.003 Clinical stage 2.194 1.581–3.044 < 0.001 2.309 1.440–3.704 0.001 Radiation dose 0.965 0.923–1.009 0.118 0.996 0.921–1.078 0.927 LDH 1.641 1.116–2.414 0.012 1.369 0.778–2.408 0.276 NLR 1.069 0.965–1.184 0.199 1.015 0.901–1.142 0.810 PLR 1.001 0.999–1.003 0.177 1.001 0.999–1.004 0.377 CR status 0.128 0.072–0.228 < 0.001 0.295 0.157–0.556 < 0.001 performance. The calibration curve also showed a high prediction accuracy. Therefore, we believe that this pre - diction model based on Rad-score can provide a more accurate tool to predict LPFS, which was a convenient and economical means. Although a prognosis prediction model was established and validated, there are some challenges for interpreta- tion of the results. Due to the fact that the machine and scanning parameters of CT in other centers are usually different and not standardized, the utility of the results or the radiomic features in other study was full of uncer- Fig. 5 Nomogram model for predicting LPFS based on Rad‑score. tainty. Moreover, radiomic-biology correlations have not Rad.score refers to Rad‑score. 0, 1, 2, and 3 refers to N0, N1, N2 and yet to be identified in published literature and clinical N3 in N stage line respectively. CR represents complete response, the experience, so there is no concrete interpretation about value of 0 and 1 refer to non‑ CR and CR status respectively the features or the feature sets. On the other hand, differ - ent methodologies for feature selection and the focus on Fig. 6 Calibration curve validation for Nomogram model in training cohort (A) and validation cohort (B). The horizontal axis represents the predicted 3‑ year LPFS and the vertical axis represents the actual 3‑ year LPFS. The blue diagonal dot line represents the ideal nomogram, and the red line represents the observed nomogram. The closer the calibration curve is to the diagonal line, the higher the consistency between the predicted results and the actual situation Luo et al. Radiat Oncol (2021) 16:201 Page 10 of 11 Ren‑Liang Xue, and Shao ‑Fu Huang collected the follow‑up data. Sheng‑ Xi different feature sets could have led to different results. Wu, Ze‑Sen Du and Xu‑ Yuan Li made statistical analysis. All authors reviewed These issues have also been addressed in other studies the manuscript and approved the final manuscript. [31]. Funding Our study provides a good enlightenment to the com- None. ing studies to prospectively establish patient cohorts. However, there are some defects worth noting. First Declarations of all, this study is a retrospective study. Because of the long-time span of CT images used in image data acquisi- Ethics approval and consent to participate Not applicable. tion, there were inevitably some problems that the image quality and scanning parameters were hard to be exactly Consent for publication the same, especially the development time and dosage Not applicable. of enhancer, so this study only collects the information Conflict of interest of plain CT images. Secondly, as patients’ response to Authors declare that there is no conflict of interest. chemoradiotherapy could not be evaluated pathologi- Availability of data and materials cally, clinical CR used in our study can’t represent patho- The datasets used and/or analyzed during the current study are available from logical CR completely truly. Fortunately, a considerable the corresponding author upon reasonable request. number of patients achieved CR had been confirmed by Author details gastroscopy pathology. Being limited by the nature of Department of Radiation Oncology, Shantou Central Hospital, Shan‑ a single-center retrospective study, the results may be tou 515000, Guangdong, China. Department of Medical Oncology, Huizhou biased to some extent, and its reliability and universality Municipal Central Hospital of Guangdong Province, Huizhou, China. Depar t‑ ment of Surgical Oncology, Shantou Central Hospital, Shantou, Guangdong, still need different centers to further carry out large sam - China. Department of Medical Oncology, Shantou Central Hospital, Shantou, ple size research verification. Guangdong, China. Received: 2 June 2021 Accepted: 24 September 2021 Conclusion In a word, this study established and validated a predic- tion model based on radiomic features and clinical fac- tors, which can be used to predict LPFS of patients after References CCRT. As an intuitive and convenient prediction method, 1. Chen W, Zheng R, Baade PD, et al. Cancer statistics in China, 2015. CA this model is conducive to identifying the patients with Cancer J Clin. 2016;66:115–32. 2. Malhotra GK, Yanala U, Ravipati A, et al. Global trends in esophageal ESCC benefited more from CCRT. cancer. J Surg Oncol. 2017;115:564–79. 3. Zeng H, Zheng R, Zhang S, et al. Esophageal cancer statistics in China, 2011: estimates based on 177 cancer registries. Thorac Cancer. Abbreviations 2016;7:232–7. ESCC: Esophageal squamous cell cancer; CCRT : Concurrent chemo‑radiother ‑ 4. Ajani JA, D’Amico TA, Bentrem DJ, et al. Esophageal and esophagogastric apy; CR: Complete response; LPFS: Local progress‑free survival; LASSO: Least junction cancers, Version 2.2019, NCCN clinical practice guidelines in absolute shrinkage and selection operator; AUC : Area under the receiver; EC: oncology. J Natl Compr Cancer Netw. 2019;17:855–83. Esophageal cancer; CT: Computed tomography; MR: Magnetic resonance; 5. Minsky BD, Pajak TF, Ginsberg RJ, et al. INT 0123 (Radiation Therapy 3DRT: Three‑ dimensional conformal radiation therapy; GTV: Gross tumor Oncology Group 94–05) phase III trial of combined‑modality therapy for volume; GTVnd: Nodal gross tumor volume; CTV: Clinical target volume; CTVt: esophageal cancer: high‑ dose versus standard‑ dose radiation therapy. J Tumor clinical target volume; CTVnd: Nodal clinical target volume; PTV: Plan‑ Clin Oncol. 2002;20:1167–74. ning target volume; VOI: Volume of interest; GLCM: Gray‑level co ‑ occurrence 6. Welsh J, Settle SH, Amini A, et al. Failure patterns in patients with matrix; GLSZM: Gray‑level size ‑zone matrix; GLRLM: Gray‑level run‑length esophageal cancer treated with definitive chemoradiation. Cancer. matrix; NGTDM: Neighboring gray‑tone difference matrix; GLDM: Gray‑level 2012;118:2632–40. dependence matrix. 7. Li Y, Zschaeck S, Lin Q, et al. Metabolic parameters of sequential 18F‑FDG PET/CT predict overall survival of esophageal cancer patients treated with (chemo‑) radiation. Radiat Oncol. 2019;14:35. Supplementary Information 8. Nkhali L, Thureau S, Edet‑Sanson A, et al. FDG‑PET/CT during concomi‑ The online version contains supplementary material available at https:// doi. tant chemo radiotherapy for esophageal cancer: reducing target volumes org/ 10. 1186/ s13014‑ 021‑ 01925‑z. to deliver higher radiotherapy doses. Acta Oncol. 2015;54:909–15. 9. Wang WP, He SL, Yang YS, Chen LQ. Strategies of nodal staging of the TNM system for esophageal cancer. Ann Transl Med. 2018;6:77. Additional file 1. The list of radiomics features extracted from the deline ‑ 10. Luo HS, Xu HY, Du ZS, et al. Impact of sex on the prognosis of patients ated VOIs. with esophageal squamous cell cancer underwent definitive radiother ‑ apy: a propensity score‑matched analysis. Radiat Oncol. 2019;14:74. 11. Yang Z, He B, Zhuang X, et al. CT‑based radiomic signatures for pre ‑ Acknowledgements diction of pathologic complete response in esophageal squamous None. cell carcinoma after neoadjuvant chemoradiotherapy. J Radiat Res. 2019;60:538–45. Authors’ contributions 12. Hu P, Liu Q, Deng G, et al. Radiosensitivity nomogram based on circulat‑ He‑San Luo designed the study. He ‑San Luo, Wei‑Zhen Huang and Ying‑ ing neutrophils in thoracic cancer. Future Oncol. 2019;15:727–37. Ying Chen prepared figures and wrote the manuscript text. Hong‑ Yao Xu, L uo et al. Radiat Oncol (2021) 16:201 Page 11 of 11 13. Lambin P, Rios‑ Velazquez E, Leijenaar R, et al. Radiomics: extracting more 24. Zhou S, Zhang L, Luo L, et al. Failure pattern of elective nodal irradiation information from medical images using advanced feature analysis. Eur J for esophageal squamous cell cancer treated with neoadjuvant chemora‑ Cancer. 2012;48:441–6. diotherapy. Jpn J Clin Oncol. 2018;48:815–21. 14. Yip C, Landau D, Kozarski R, et al. Primary esophageal cancer: heterogene‑ 25. Zhu SC, Li QF, Zhang XY, et al. Clinical outcomes of different irradiation ity as potential prognostic biomarker in patients treated with definitive ranges in definitive intensity‑modulated radiotherapy for esophageal chemotherapy and radiation therapy. Radiology. 2014;270:141–8. cancer. Zhonghua Zhong Liu Za Zhi. 2020;42:1040–7. 15. Larue R, Klaassen R, Jochems A, et al. Pre‑treatment CT radiomics to pre ‑ 26. Ganeshan B, Skogen K, Pressney I, et al. Tumour heterogeneity in oesoph‑ dict 3‑ year overall survival following chemoradiotherapy of esophageal ageal cancer assessed by CT texture analysis: preliminary evidence of cancer. Acta Oncol. 2018;57:1475–81. an association with tumour metabolism, stage, and survival. Clin Radiol. 16. Rice TW, Gress DM, Patil DT, et al. Cancer of the esophagus and esoph‑ 2012;67:157–64. agogastric junction‑Major changes in the American Joint Committee 27. Zhang X, Wang Y, Qu P, et al. Prognostic value of tumor length for on Cancer eighth edition cancer staging manual. CA Cancer J Clin. cause‑specific death in resectable esophageal cancer. Ann Thorac Surg. 2017;67:304–17. 2018;106:1038–46. 17. Chun H, Xue‑jiao R, Lan W, et al. Evaluating short ‑term radiotherapeutic 28. Zhiguo Z, Xin W, Lan W, et al. Eec ff t of tumor length on clinical stage for effect on esophageal cancer by barium meal combined with CT scans. non‑ operative esophageal squamous cell carcinoma patients—multi‑ Chin J Radiat Oncol. 2013;22:26–9. center retrospective data analysis (3JECROG R‑01D). Chin J Radiat Oncol. 18. van Griethuysen JJM, Fedorov A, Parmar C, et al. Computational 2019;28:490–4. radiomics system to decode the radiographic phenotype. Cancer Res. 29. Liu SL, Xi M, Yang H, et al. Is There a correlation between clinical complete 2017;77:e104–7. response and pathological complete response after neoadjuvant chemo‑ 19. Li C, Ni W, Wang X, et al. A phase I/II radiation dose escalation trial using radiotherapy for esophageal squamous cell cancer? Ann Surg Oncol. simultaneous integrated boost technique with elective nodal irradia‑ 2016;23:273–81. tion and concurrent chemotherapy for unresectable esophageal cancer. 30. Li Z, Shan F, Wang Y, et al. Correlation of pathological complete response Radiat Oncol. 2019;14:48. with survival after neoadjuvant chemotherapy in gastric or gastroesoph‑ 20. Lin FC, Chang WL, Chiang NJ, et al. Radiation dose escalation can improve ageal junction cancer treated with radical surgery: a meta‑analysis. PLoS local disease control and survival among esophageal cancer patients ONE. 2018;13:e0189294. with large primary tumor volume receiving definitive chemoradiother ‑ 31. Lambin P, Leijenaar RTH, Deist TM, et al. Radiomics: the bridge between apy. PLoS ONE. 2020;15:e0237114. medical imaging and personalized medicine. Nat Rev Clin Oncol. 21. Zhang W, Luo Y, Wang X, et al. Dose‑ escalated radiotherapy improved 2017;14:749–62. survival for esophageal cancer patients with a clinical complete response after standard‑ dose radiotherapy with concurrent chemotherapy. Cancer Publisher’s Note Manag Res. 2018;10:2675–82. Springer Nature remains neutral with regard to jurisdictional claims in pub‑ 22. Zhang W, Zhao J, Han W, et al. Dose escalation of 3D radiotherapy is lished maps and institutional affiliations. effective for esophageal squamous cell carcinoma: a multicenter retro ‑ spective analysis (3JECROG R‑03). Ann Transl Med. 2020;8:1140. 23. Oppedijk V, van der Gaast A, van Lanschot JJ, et al. Patterns of recurrence after surgery alone versus preoperative chemoradiotherapy and surgery in the CROSS trials. J Clin Oncol. 2014;32:385–91. Re Read ady y to to submit y submit your our re researc search h ? Choose BMC and benefit fr ? Choose BMC and benefit from om: : fast, convenient online submission thorough peer review by experienced researchers in your field rapid publication on acceptance support for research data, including large and complex data types • gold Open Access which fosters wider collaboration and increased citations maximum visibility for your research: over 100M website views per year At BMC, research is always in progress. Learn more biomedcentral.com/submissions http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Radiation Oncology Springer Journals

Development and validation of a radiomics-based model to predict local progression-free survival after chemo-radiotherapy in patients with esophageal squamous cell cancer

Loading next page...
 
/lp/springer-journals/development-and-validation-of-a-radiomics-based-model-to-predict-local-4p0lt8rPRB

References (35)

Publisher
Springer Journals
Copyright
Copyright © The Author(s) 2021
eISSN
1748-717X
DOI
10.1186/s13014-021-01925-z
Publisher site
See Article on Publisher Site

Abstract

Purpose: To develop a nomogram model for predicting local progress‑free survival (LPFS) in esophageal squamous cell carcinoma (ESCC) patients treated with concurrent chemo‑radiotherapy (CCRT ). Methods: We collected the clinical data of ESCC patients treated with CCRT in our hospital. Eligible patients were randomly divided into training cohort and validation cohort. The least absolute shrinkage and selection operator (LASSO) with COX regression was performed to select optimal radiomic features to calculate Rad‑score for predicting LPFS in the training cohort. The univariate and multivariate analyses were performed to identify the predictive clinical factors for developing a nomogram model. The C‑index was used to assess the performance of the predictive model and calibration curve was used to evaluate the accuracy. Results: A total of 221 ESCC patients were included in our study, with 155 patients in training cohort and 66 patients in validation cohort. Seventeen radiomic features were selected by LASSO COX regression analysis to calculate Rad‑ score for predicting LPFS. The patients with a Rad‑score ≥ 0.1411 had high risk of local recurrence, and those with a Rad‑score < 0.1411 had low risk of local recurrence. Multivariate analysis showed that N stage, CR status and Rad‑score were independent predictive factors for LPFS. A nomogram model was built based on the result of multivariate analy‑ sis. The C‑index of the nomogram was 0.745 (95% CI 0.7700–0.790) in training cohort and 0.723(95% CI 0.654–0.791) in validation cohort. The 3‑ year LPFS rate predicted by the nomogram model was highly consistent with the actual 3‑ year LPFS rate both in the training cohort and the validation cohort. Conclusion: We developed and validated a prediction model based on radiomic features and clinical factors, which can be used to predict LPFS of patients after CCRT. This model is conducive to identifying the patients with ESCC benefited more from CCRT. Keywords: Chemo‑radiotherapy, Esophageal squamous cell cancer, Radiomics, LPFS, Nomogram Introduction Esophageal cancer (EC) is the sixth common malignant *Correspondence: luohesan@163.com Department of Radiation Oncology, Shantou Central Hospital, tumors in China with an estimated 477.9 thousand new Shantou 515000, Guangdong, China cases, accounting for half of the new esophageal cancer Full list of author information is available at the end of the article © The Author(s) 2021. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/. The Creative Commons Public Domain Dedication waiver (http:// creat iveco mmons. org/ publi cdoma in/ zero/1. 0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. Luo et al. Radiat Oncol (2021) 16:201 Page 2 of 11 worldwide [1, 2]. In China, approximately 90% of the Patients and methods patients with esophageal cancer are histologically diag- Patients’ cohort nosed as esophageal squamous cell carcinomas (ESCC) We collected the clinical data of patients diagnosed as which is different from esophageal adenocarcinoma ESCC and received CCRT in our hospital during the (EAC) in risk factors and prognosis [3]. Most patients period from January 2013 to December 2015. Patients with locally advanced ESCC lost the opportunity for sur- were excluded if they met the exclusion criteria as follows: gery, and concurrent chemo-radiotherapy (CCRT) has (1) patients received esophagectomy and preoperative been recommended as a standard treatment [4]. How- or postoperative adjuvant radiotherapy; (2) patients had ever, more than half of patients treated with standard distant metastatic disease; (3) patients received low-dose dose CCRT eventually developed local recurrence or dis- (< 50  Gy) palliative radiotherapy; (4) clinicopathological tant metastases and succumbed to this disease [5, 6]. A information of the patients was incomplete; (5) patients individual CCRT strategy with escalated radiation dose were diagnosed as esophageal fistula before treatment; based on PET-CT would benefit the patients with high (6) poor visualization quality due to image artifacts or tumor burden and risk of recurrence [7, 8]. To facilitate the tumor was too small to be recognized on CT images; a individual CCRT strategy in an early stage, solid predic- (7) patients had other primary tumor; (8) patients died tive model for local recurrence and prognosis could play within three months after chemoradiotherapy. an important role. After multiple iterations, a total of 221 patients were For patients received CCRT, local and regional recur- randomly divided into two groups, with 155 patients rence is the most common failure pattern and pre-treat- in the training cohort and 66 patients in the validation ment clinical TNM staging is still the most commonly cohort. To improve the generalization property of the used system for prognosis prediction [9]. However, the result, multi-factors stratification was used to keep the currently used clinical TNM staging follows the same characteristics of sub cohort consistent with the whole criteria as pathological staging, which is based on imag- cohort. The process of patients’ enrollment and randomi - ing assessment of tumor size and surrounding invasion, zation were shown in Fig.  1. This study was approved by ignoring the information such as length and volume of the Institutional Committee of our hospital on Human esophageal cancer lesions. Recently, a series of clinico- Rights. Disease of the patients was staged according to pathologic biomarkers have been investigated and veri- the 8th edition of AJCC TNM classification for esopha - fied to be available for prediction of therapeutic response geal cancer [16]. and prognosis [10–12]. Radiomics is a new technique for image quantitative analysis about computed tomography Chemoradiotherapy protocol (CT) images, magnetic resonance (MR) images, posi- Radiotherapy was delivered daily to patients with three- tron emission tomography (PET) images, etc. [13]. Sev- dimensional conformal radiation therapy (3DRT) or eral studies demonstrated that radiomic features could intensity-modulated radiation therapy (IMRT) tech- potentially identify prognostic phenotype in patients nique using a Varian IX or Varian 23EX linear accelera- with EC. Yip et al. [14] suggested that a model combined tor in this study. The gross tumor volume (GTV) includes CT-based texture feature and esophageal maximal wall the esophageal cancer (GTVp) and the positive regional thickness assessment could predict the overall survival lymph nodes (GTVnd). The GTV was delineated on CT in EC patients treated with CCRT. Moreover, the model imaging according to barium esophagogram, endoscopic performed better than treatment response alone. Larue examination or PET imaging. The CTV was defined as et  al. [15] extracted out five radiomic features from CT the GTVp with 0.5–1 cm radial expansion and 2.5–3 cm image before chemoradiotherapy to describe the hetero- axial direction expansion or the GTVnd with 0.5–0.8 cm geneity of tumors and found that these five features could uniform expansion. The planning target volume (PTV) predict the 3-year survival rate of patients with EC after was defined as CTV with a 1  cm uniform expansion. A neo-chemoradiotherapy. However, most radiomic studies total prescribed dose of 50–72  Gy (median, 64  Gy) in included a small number of patients with EAC and ESCC. conventional fractionation was delivered to the patients. In this study, we explored the prognostic value of 3D Two cycles of platinum-based chemotherapy were radiomic features from pretreatment CT images of administered concurrently with radiotherapy. Sixty-one esophageal cancer patients and developed a model com- patients received TP (paclitaxel + cisplatin) chemo- bined radiomic features and clinical information to pre- therapy every three weeks, which consists of cisplatin 2 2 dict LPFS in patients with ESCC after CCRT. To evaluate (60  mg/m on Day 1) plus paclitaxel (135–180  mg/m the performance of the model, a validation cohort of on Days 1). One hundred and sixty patients received the patients were employed for validation. PF (cisplatin + fluorouracil) regimen every four weeks, L uo et al. Radiat Oncol (2021) 16:201 Page 3 of 11 Fig. 1 Flow chart of patients’ screening and allocation Luo et al. Radiat Oncol (2021) 16:201 Page 4 of 11 which consists of cisplatin (60 mg/m on Day 1) and fluo - dependence matrix, GLDM). The wavelet filter was used rouracil (750 mg/m /24 h on Days 1–4). in image pre-processing for texture features extraction. In all, for each VOI, 107 original features (Additional file  1: Table  S1) and 744 wavelet features (Additional Response evaluation file  1: Table  S1) were collected. Among the 107 original The response to chemo-radiotherapy was evaluated one features, there were 18 first order statistics features, 14 month after CCRT according to the criteria of short-term shape-based histogram features, 24 GLCM features, 14 response evaluation standard on esophageal cancer using GLDM features, 16 GLRLM features、16 GLSZM fea- CT images and barium esophagogram. According to the tures and 5 NGTDM features. Mathematical definitions response evaluation criteria, clinical response was classi- of these radiomic features have previously been described fied as complete response (CR), partial response (PR), no [18] and available at https:// pyrad iomics. readt hedocs. io/ response (NR), or progressive disease (PD). Patients who en/ latest/ featu res. html. were classified as CR by barium esophagogram and had the maximal esophageal wall thickness of ≤ 1.2  cm and Statistical analysis the volumes of residual lymph nodes of ≤ 1.0 cm on CT At the first, statistical analyses were performed with Chi- were finally defined as CR [17]. squared test or Fisher’s to assess the difference of the clinical characteristics between training cohort and vali- Radiomic feature extraction dation cohort. A p-value of < 0.05 was considered statisti- All patients were scanned using GE Lightspeed 64-slice cally significant. spiral CT (GE Medical systems, Milwaukee, Wis) before In the pre-processing of radiomic features, all the val- radiotherapy. CT image acquisition was performed ues of radiomic features were normalized using Z-score according to the following acquisition protocol: The CT normalization, which made features values lying within tube voltage was 120  kV and the tube current was 120 similar ranges and reduced the influence of large dis - mAs. Rack rotation time: 0.6  s; Detector collimation crete values. The intra-class correlation coefficient (ICC) parameters: 64 × 0.625  mm; field of view (FOV): 400- analysis was performed to evaluate the reproducibility of 500  mm; Matrix: 512 × 512; Layer thickness is 5  mm, each radiomic feature. Only the features with ICCs val- layer spacing is 5  mm. Contrast medium was injected ues ≥ 0.900 were selected for further analysis. Then, the with a high-pressure syringe at a flow rate of 3.0  ml/s least absolute shrinkage and selection operator (LASSO) (1–1.5  ml/kg, ioproxamine injection 300), followed by with COX regression was performed using R software 30 to 40 ml of normal saline for flushing, and late arterial version 3.6.2 (R Foundation for Statistical Computing, CT images were collected with a delay of 30 s. To reduce Vienna, Austria) to identify the features associated with the variability between images from different patients, all LPFS in the training cohort. The optimal parameter images were resampled to voxel of 1*1*1mm . lambda (λ) was chosen from the LASSO model using ten- 3D Slicer (version, 4.10.2, Stable Release) with radiom- fold cross-validation with the minimum partial likelihood ics extension was used for image segmentation to obtain deviance. Radiomic feature score (Rad score) for each volume of interest (VOIs). The primary tumor volume patient was built based on the LASSO COX regression (GTV) delineated by radiation oncologists for radiother- model in the training cohort. The LASSO COX regres - apy treatment planning design was defined as VOI for sion formula: radiomic features extraction. Any pixel with an attenua- tion of less than − 50 HU was excluded to avoid adjacent Rad score = β1X1 + β2X2 + β3X3 + ··· + βnXn air, fat, blood vessels and surrounding organs. Image seg- mentation was performed independently by a radiation In the above formula, X1, X2 … Xn are the different oncologist and another radiologist. To assess the repro- radiomic features identified by the LASSO COX regres - ducibility of the radiomic features extraction, tumor seg- sion model, and β1, β2 … βn are the regression coef- mentation was performed again two months later by the ficients of the corresponding features in the regression same radiologist in 30 randomly chosen patients. model. Pyradiomics V3.6.2 was used to extract radiomic fea- Univariate analysis was performed to identify the tures from delineated VOIs. Several categories of features potential prognostic factors associated with LPFS. Multi- were extracted from VOIs, including first order statistics variable COX regression analysis was performed to iden- features (IH, intensity histogram), shape-based histogram tify the independently predictors for LPFS. A nomogram features, and texture features (gray-level co-occurrence model combined Rad-score and clinical factors for pre- matrix, GLCM; gray-level size-zone matrix, GLSZM; dicting LPFS was developed and validated based on the gray-level run-length matrix, GLRLM; neighboring results of multivariable COX regression analysis using gray-tone difference matrix, NGTDM; and gray-level rms package and foreign package in R software. The L uo et al. Radiat Oncol (2021) 16:201 Page 5 of 11 predictive accuracy of the nomogram model was assessed of 1-year, 2-year and 3-year LPFS were 56.1%, 37.4% and using Calibration curve validation in both training cohort 32.1%, respectively (Fig. 2). and validation cohort. All the analyses were performed In order to develop and validate a radiomics-based with R software version 3.6.2. model for predicting LPFS of the patients, they were ran- domly divided into training cohort and validation cohort. There were 155 patients in the training cohort and 66 Results patients in the validation cohort. No significant differ - Patients’ characteristics ences (All p > 0.05) were found between the distribution A total of 221 ESCC patients who received chemoradio- of baseline characteristics in two cohorts, such as age, therapy in our hospital were eligible for further analysis gender, tumor location, T stage, N stage, clinical staging, in this study. Patients’ characteristics were summarized lactate dehydrogenase (LDH), neutrophil to 1ymphocyte in Table  1. The median follow-up time was 18.6 months. ratio (NLR), platelet to lymphocyte ratio (PLR) and CR By the end of the last follow-up, 153 patients developed ratio (33.5% in the training cohort vs 39.4% in the valida- local regional disease progression or died. The median tion cohort). Therefore, the two cohorts of patients were LPFS in the whole group was 13.7 months, and the rates comparable. Table 1 Comparison of patients’ characteristics between training cohort and validation cohort Variables Training cohort (n = 155) Validation cohort (n = 66) χ /t p Age (years), Mean ± SD 65.7147 ± 9.74 64.73 ± 10.16 0.678 0.499 Gender 0.342 0.559 Male 116 (74.8) 45 (68.2) Female 39 (25.2) 21 (31.8) Tumor location 5.814 0.121 Cervical 6 (3.9) 8 (12.1) Upper thoracic 34 (21.9) 16 (24.2) Middle thoracic 91 (58.7) 33 (50.00) Lower thoracic 24 (15.5) 9 (13.6) T stage 3.193 0.363 T1 2 (1.3) 0 (0) T2 11 (7.1) 9 (13.6) T3 66 (42.6) 27 (40.9) T4 76 (49.0) 30 (45.5) N stage 1.856 0.603 N0 20 (12.9) 13 (19.7) N1 70 (45.2) 28 (42.4) N2 55 (35.5) 22 (33.3) N3 10 (6.5) 3 (4.5) Clinical stage 3.152 0.369 I 2 (1.3) 0 (0) II 15 (9.7) 11 (16.7) III 88 (56.8) 37 (56.1) Iva 50 (32.3) 18 (27.3) Radiation dose, Median (range) 64 (60–66) 64 (60–66) − 0.920 0.358 LDH group 1.282 0.258 High 88 (56.8) 32 (48.5) Normal 67 (43.2) 34 (51.5) NLR, Median (range) 2.73 (1.96–3.71) 2.76 (2.00–3.63) − 0.448 0.654 PLR, Median (range) 137.78 (100.56–181.43) 138.87 (101.31–182.26) − 0.344 0.731 CR ratio 52 (33.5) 26 (39.4) 0.693 0.405 Rad‑score, Mean ± SD − 0.0289 ± 0.35 − 0.058 ± 0.538 0.474 0.636 Luo et al. Radiat Oncol (2021) 16:201 Page 6 of 11 The same result was found in the validation cohort (Fig. 4B, HR 1.997, 95% CI 1.070–3.728, p = 0.026). Development and validation of a predictive nomogram based on Rad‑score In order to develop a model to predict LRFS based on multiple factors, we performed univariate and multi- variate analyses to identify predictive factors for LPFS. Univariate analysis showed that the T stage, N stage, clinical stage and CR status were significantly associated with LPFS both in training cohort and validation cohort (Table  3). Multivariate analysis showed that N stage, CR status and Rad-score were independent predictive fac- tors for LPFS in ESCC patients after chemoradiotherapy Fig. 2 Kaplan–Meier curve of local‑progression free survival for all (Table  4). A nomogram model for predicting LPFS was patients built based on the result of multivariate analysis (Fig. 5A). As shown in Fig.  5 1-year, 2-year and 3-year LPFS prob- ability of every patient could be predicted based on the independent clinical characteristics and Rad-score. The Rad‑score building based on radiomic features C-index of the nomogram was 0.745 (95% CI 0.7700– LASSO-COX regression was used to screen out the 0.790) in training cohort and 0.723(95% CI 0.654–0.791) optimal radiomic features associated with LPFS of in validation cohort. the patients in the training cohort (Fig.  3A, B). As a Finally, we performed calibration curve to evaluate the result, seventeen radiomic features were screened accuracy of the nomogram model. As shown in Fig.  6, out (The features and their coefficients were listed in the 3-year LPFS rate predicted by the nomogram model the Table  2). The Rad-score was calculated as follows: based on Rad-score was highly consistent with the actual Rad-score = -0.104667846*or ig inal_f irstorder_Ske w- 3-year LPFS rate both in the training cohort and the vali- ness + 0.001161134*or ig in_g l sz m_Si zeZ oneNonUni- dation cohort. formityNormalized + 0.034339901*origin_glszm_Size- ZoneNonUniformity-0.017089976*origin_glszm_Low- GrayLevelZoneEmphasis + 0.062595767*wavelet-HLL_ Discussion glcm_Idn + 0.026703955*wavelet-HLL_firstorder_Maxi - Concurrent chemoradiotherapy (CCRT) is a radical m u m + 0 . 0 4 2 9 5 7 1 4 3 * w a v e l e t- H L L _ g l s z m _ S i z e Z o - treatment for patients with inoperable esophageal can- neNonUniformityNormalized + 0.017543973*wavelet- cer or refused surgery [4]. Many studies have shown that LHL_firstorder_TotalEnerg y + 0.003781538*w avele t- dose-escalation radiotherapy properly can improve the L H L_f irst or der_Ma xim um-0.007364328*w a ve le t- local control and survival of patients with ESCC [19–22]. LLH_gldm_SmallD ep endenceLowGrayLe velEmpha- Nevertheless, 30–50% of patients have local recurrence sis + 0.157807433*wavelet-LLH_glcm_DifferenceVari- within 3  years [23–25]. In our present study, we con- ance + 0.042028490*wavelet-LLH_glrlm_ShortRunHigh- structed a prediction model combined the clinical char- GrayLevelEmphasis-0.101981005*wavelet-LLH_ngtdm_ acteristics and CT radiomic features which can predict Coarseness-0.073958943*wavelet-HLH_gldm_SmallDe- the LPFS of patients after CCRT. With the help of this pendenceLowGrayLevelEmphasis + 0.051287394*wave- model, we can preliminarily judge the probability of LPFS let-HLH_firstorder_Maximum-0.055239045*wavelet- of patients and identify the patients benefit more from HHH_glcm_MaximumProbability-0.028958889*wavelet- CCRT. LLL_glcm_Imc2. Radiomics studies in esophageal cancer started rela- There was an optimal cutoff value of Rad score to tively late, and there are still few data about applying radi- divide the patients into two groups with different risk of omics analysis to evaluate the prognosis of esophageal local recurrence. As shown in Fig.  3C, the patients with cancer. Ganeshan et  al. [26] first analyzed the radiomic a Rad-score ≥ 0.1411 had high risk of local recurrence, features of CT before treatment in esophageal cancer and those with a Rad-score < 0.1411 had low risk of local patients and found that the radiomic features represent- recurrence. In the training cohort, the patients in the ing uniformity parameters were significantly different group with high risk of local recurrence had significantly between stage I/II and stage III/IV disease, which were shorter time of LPFS than those with risk of local recur- independent predictors of patients’ prognosis. Subse- rence (Fig. 4A, HR 2.882, 95% CI 1.926–4.313, p < 0.001). quently, Yip et al. [14] found that the tumor heterogeneity L uo et al. Radiat Oncol (2021) 16:201 Page 7 of 11 Fig. 3 Selection of radiomic features associated with LPFS using the LASSO COX regression model. A Coefficients profiles of radiomic features. The horizontal axis value is logλ, and the vertical axis value represent the coefficients of radiomic features. B The cross‑ validation curve. The horizontal axis value is logλ, and the vertical axis value is partial likelihood deviance. C The optimal cutoff of Rad‑score. Red lines or red dots represent patients at high risk of local recurrence and green lines or green dots represent patients at low risk of local recurrence. The optimal cutoff value is 0.1411, as shown in the vertical line in the figure could be represented by the change of CT radiomic fea- chemoradiotherapy, which was also supported in our tures before and after neoadjuvant treatment, which was study.- related to the prognosis and survival of patients. Larue Clinical TNM staging before treatment is still the most et al. [15] also found that five radiomic features extracted commonly used prediction system of prognosis for ESCC from CT before chemoradiotherapy could be used to patients treated with chemoradiotherapy. Combination describe the tumor heterogeneity and predict the 3-year of TNM staging and other prognostic factors can predict survival rate of patients after neoadjuvant chemoradio- the prognosis of patients more individually and accu- therapy and surgery with AUCs (AUC, area under the rately [27, 28]. Some studies have shown that the progno- receiver) of 0.69 in the training group and 0.61 in the vali- sis of patients who achieved CR after chemoradiotherapy dation group. All these studies suggested that radiomic was better than that of patients not CR [29, 30]. There - features played an important role in evaluating the prog- fore, CR after CCRT had become another important nosis of esophageal cancer and could be used to predict predictor for the prognosis of patients besides clinical the long-term survival of esophageal cancer patients after stages. In the present study, univariate analysis showed Luo et al. Radiat Oncol (2021) 16:201 Page 8 of 11 Table 2 Radiomics feature associated with LPFS selected by LASSO COX analysis Radiomics features Coefficients original_firstorder_Skewness − 0.104667846 origin_glszm_SizeZoneNonUniformityNormalized 0.001161134 origin_glszm_SizeZoneNonUniformity 0.034339901 origin_glszm_LowGrayLevelZoneEmphasis − 0.017089976 wavelet‑HLL_glcm_Idn 0.062595767 wavelet‑HLL_firstorder_Maximum 0.026703955 wavelet‑HLL_glszm_SizeZoneNonUniformityNormalized 0.042957143 wavelet‑LHL_firstorder_TotalEnergy 0.017543973 wavelet‑LHL_firstorder_Maximum 0.003781538 wavelet‑LLH_gldm_SmallDependenceLowGrayLevelEmphasis − 0.007364328 wavelet‑LLH_glcm_DifferenceVariance 0.157807433 wavelet‑LLH_glrlm_ShortRunHighGrayLevelEmphasis 0.042028490 wavelet‑LLH_ngtdm_Coarseness − 0.101981005 wavelet‑HLH_gldm_SmallDependenceLowGrayLevelEmphasis − 0.073958943 wavelet‑HLH_firstorder_Maximum 0.051287394 wavelet‑HHH_glcm MaximumProbability − 0.055239045 wavelet‑LLL_glcm_Imc2 − 0.028958889 Table 4 Multivariate analysis of prognostic factors associated with LPFS for patients with ESCC treated with chemoradiotherapy Variables Multivariate analysis p HR 95% CI p T stage 0.858 0.526–1.400 0.540 N stage 1.892 1.122–3.190 0.017 Clinical stage 0.627 0.313–1.258 0.189 Rad‑score 4.423 1.993–9.814 0.000 CR status 0.154 0.080–0.297 0.000 related to LPFS after CCRT. Moreover, Rad-score based on 17 radiomic features extracted from CT images before chemoradiotherapy was significantly related to LPFS of patients after chemoradiotherapy. Further multivariate analysis showed that Rad-score, N stage and CR after radiotherapy were independent predictors of patients’ Fig. 4 Kaplan–Meier survival curve of patients with high and low LPFS, while T stage and clinical stage had no statistical recurrence risk based on Rad‑score. A LPFS survival curve of patients significance in this multivariate analysis model, prob - in the training cohort: green line represents patients with low risk ably because Rad-score derived from the primary tumor of local recurrence and red represents patients with high risk of focus and had interactive effects with T stage and clinical local recurrence. The difference is significant between two groups, stage. We developed and validated a nomogram model p < 0.001. B LPFS survival curve of patients in the validation cohort: green line represents patients with low risk of local recurrence and based on the results of multivariate analysis. C-index red line represents patients with high risk of local recurrence. The and calibration curve were used to evaluate the perfor- difference is significant between two groups, p = 0.026 mance and prediction accuracy of the nomogram model. The C-index of the model was 0.745(95% CI 0.700–0.790) in the training cohort and 0.723(95% CI 0.654–0.791) that pre-treatment clinical T stage, N stage, clinical stage in the validation cohort, indicating high prediction and CR after radiotherapy were the prognostic factors L uo et al. Radiat Oncol (2021) 16:201 Page 9 of 11 Table 3 Univariate analysis of prognostic factors associated with LPFS in patients with ESCC treated with chemoradiotherapy Variables Training cohort p Validation cohort p HR 95% CI p HR 95% CI p Age 0.987 0.967–1.007 0.202 0.984 0.957–1.011 0.252 Gender 1.639 1.014–2.650 0.044 1.198 0.652–2.202 0.560 Tumor location 1.120 0.851–1.473 0.419 1.289 0.917–1.812 0.143 T stage 2.015 1.453–2.793 < 0.001 1.943 1.227–3.077 0.005 N stage 1.867 1.446–2.410 < 0.001 1.765 1.215–2.563 0.003 Clinical stage 2.194 1.581–3.044 < 0.001 2.309 1.440–3.704 0.001 Radiation dose 0.965 0.923–1.009 0.118 0.996 0.921–1.078 0.927 LDH 1.641 1.116–2.414 0.012 1.369 0.778–2.408 0.276 NLR 1.069 0.965–1.184 0.199 1.015 0.901–1.142 0.810 PLR 1.001 0.999–1.003 0.177 1.001 0.999–1.004 0.377 CR status 0.128 0.072–0.228 < 0.001 0.295 0.157–0.556 < 0.001 performance. The calibration curve also showed a high prediction accuracy. Therefore, we believe that this pre - diction model based on Rad-score can provide a more accurate tool to predict LPFS, which was a convenient and economical means. Although a prognosis prediction model was established and validated, there are some challenges for interpreta- tion of the results. Due to the fact that the machine and scanning parameters of CT in other centers are usually different and not standardized, the utility of the results or the radiomic features in other study was full of uncer- Fig. 5 Nomogram model for predicting LPFS based on Rad‑score. tainty. Moreover, radiomic-biology correlations have not Rad.score refers to Rad‑score. 0, 1, 2, and 3 refers to N0, N1, N2 and yet to be identified in published literature and clinical N3 in N stage line respectively. CR represents complete response, the experience, so there is no concrete interpretation about value of 0 and 1 refer to non‑ CR and CR status respectively the features or the feature sets. On the other hand, differ - ent methodologies for feature selection and the focus on Fig. 6 Calibration curve validation for Nomogram model in training cohort (A) and validation cohort (B). The horizontal axis represents the predicted 3‑ year LPFS and the vertical axis represents the actual 3‑ year LPFS. The blue diagonal dot line represents the ideal nomogram, and the red line represents the observed nomogram. The closer the calibration curve is to the diagonal line, the higher the consistency between the predicted results and the actual situation Luo et al. Radiat Oncol (2021) 16:201 Page 10 of 11 Ren‑Liang Xue, and Shao ‑Fu Huang collected the follow‑up data. Sheng‑ Xi different feature sets could have led to different results. Wu, Ze‑Sen Du and Xu‑ Yuan Li made statistical analysis. All authors reviewed These issues have also been addressed in other studies the manuscript and approved the final manuscript. [31]. Funding Our study provides a good enlightenment to the com- None. ing studies to prospectively establish patient cohorts. However, there are some defects worth noting. First Declarations of all, this study is a retrospective study. Because of the long-time span of CT images used in image data acquisi- Ethics approval and consent to participate Not applicable. tion, there were inevitably some problems that the image quality and scanning parameters were hard to be exactly Consent for publication the same, especially the development time and dosage Not applicable. of enhancer, so this study only collects the information Conflict of interest of plain CT images. Secondly, as patients’ response to Authors declare that there is no conflict of interest. chemoradiotherapy could not be evaluated pathologi- Availability of data and materials cally, clinical CR used in our study can’t represent patho- The datasets used and/or analyzed during the current study are available from logical CR completely truly. Fortunately, a considerable the corresponding author upon reasonable request. number of patients achieved CR had been confirmed by Author details gastroscopy pathology. Being limited by the nature of Department of Radiation Oncology, Shantou Central Hospital, Shan‑ a single-center retrospective study, the results may be tou 515000, Guangdong, China. Department of Medical Oncology, Huizhou biased to some extent, and its reliability and universality Municipal Central Hospital of Guangdong Province, Huizhou, China. Depar t‑ ment of Surgical Oncology, Shantou Central Hospital, Shantou, Guangdong, still need different centers to further carry out large sam - China. Department of Medical Oncology, Shantou Central Hospital, Shantou, ple size research verification. Guangdong, China. Received: 2 June 2021 Accepted: 24 September 2021 Conclusion In a word, this study established and validated a predic- tion model based on radiomic features and clinical fac- tors, which can be used to predict LPFS of patients after References CCRT. As an intuitive and convenient prediction method, 1. Chen W, Zheng R, Baade PD, et al. Cancer statistics in China, 2015. CA this model is conducive to identifying the patients with Cancer J Clin. 2016;66:115–32. 2. Malhotra GK, Yanala U, Ravipati A, et al. Global trends in esophageal ESCC benefited more from CCRT. cancer. J Surg Oncol. 2017;115:564–79. 3. Zeng H, Zheng R, Zhang S, et al. Esophageal cancer statistics in China, 2011: estimates based on 177 cancer registries. Thorac Cancer. Abbreviations 2016;7:232–7. ESCC: Esophageal squamous cell cancer; CCRT : Concurrent chemo‑radiother ‑ 4. Ajani JA, D’Amico TA, Bentrem DJ, et al. Esophageal and esophagogastric apy; CR: Complete response; LPFS: Local progress‑free survival; LASSO: Least junction cancers, Version 2.2019, NCCN clinical practice guidelines in absolute shrinkage and selection operator; AUC : Area under the receiver; EC: oncology. J Natl Compr Cancer Netw. 2019;17:855–83. Esophageal cancer; CT: Computed tomography; MR: Magnetic resonance; 5. Minsky BD, Pajak TF, Ginsberg RJ, et al. INT 0123 (Radiation Therapy 3DRT: Three‑ dimensional conformal radiation therapy; GTV: Gross tumor Oncology Group 94–05) phase III trial of combined‑modality therapy for volume; GTVnd: Nodal gross tumor volume; CTV: Clinical target volume; CTVt: esophageal cancer: high‑ dose versus standard‑ dose radiation therapy. J Tumor clinical target volume; CTVnd: Nodal clinical target volume; PTV: Plan‑ Clin Oncol. 2002;20:1167–74. ning target volume; VOI: Volume of interest; GLCM: Gray‑level co ‑ occurrence 6. Welsh J, Settle SH, Amini A, et al. Failure patterns in patients with matrix; GLSZM: Gray‑level size ‑zone matrix; GLRLM: Gray‑level run‑length esophageal cancer treated with definitive chemoradiation. Cancer. matrix; NGTDM: Neighboring gray‑tone difference matrix; GLDM: Gray‑level 2012;118:2632–40. dependence matrix. 7. Li Y, Zschaeck S, Lin Q, et al. Metabolic parameters of sequential 18F‑FDG PET/CT predict overall survival of esophageal cancer patients treated with (chemo‑) radiation. Radiat Oncol. 2019;14:35. Supplementary Information 8. Nkhali L, Thureau S, Edet‑Sanson A, et al. FDG‑PET/CT during concomi‑ The online version contains supplementary material available at https:// doi. tant chemo radiotherapy for esophageal cancer: reducing target volumes org/ 10. 1186/ s13014‑ 021‑ 01925‑z. to deliver higher radiotherapy doses. Acta Oncol. 2015;54:909–15. 9. Wang WP, He SL, Yang YS, Chen LQ. Strategies of nodal staging of the TNM system for esophageal cancer. Ann Transl Med. 2018;6:77. Additional file 1. The list of radiomics features extracted from the deline ‑ 10. Luo HS, Xu HY, Du ZS, et al. Impact of sex on the prognosis of patients ated VOIs. with esophageal squamous cell cancer underwent definitive radiother ‑ apy: a propensity score‑matched analysis. Radiat Oncol. 2019;14:74. 11. Yang Z, He B, Zhuang X, et al. CT‑based radiomic signatures for pre ‑ Acknowledgements diction of pathologic complete response in esophageal squamous None. cell carcinoma after neoadjuvant chemoradiotherapy. J Radiat Res. 2019;60:538–45. Authors’ contributions 12. Hu P, Liu Q, Deng G, et al. Radiosensitivity nomogram based on circulat‑ He‑San Luo designed the study. He ‑San Luo, Wei‑Zhen Huang and Ying‑ ing neutrophils in thoracic cancer. Future Oncol. 2019;15:727–37. Ying Chen prepared figures and wrote the manuscript text. Hong‑ Yao Xu, L uo et al. Radiat Oncol (2021) 16:201 Page 11 of 11 13. Lambin P, Rios‑ Velazquez E, Leijenaar R, et al. Radiomics: extracting more 24. Zhou S, Zhang L, Luo L, et al. Failure pattern of elective nodal irradiation information from medical images using advanced feature analysis. Eur J for esophageal squamous cell cancer treated with neoadjuvant chemora‑ Cancer. 2012;48:441–6. diotherapy. Jpn J Clin Oncol. 2018;48:815–21. 14. Yip C, Landau D, Kozarski R, et al. Primary esophageal cancer: heterogene‑ 25. Zhu SC, Li QF, Zhang XY, et al. Clinical outcomes of different irradiation ity as potential prognostic biomarker in patients treated with definitive ranges in definitive intensity‑modulated radiotherapy for esophageal chemotherapy and radiation therapy. Radiology. 2014;270:141–8. cancer. Zhonghua Zhong Liu Za Zhi. 2020;42:1040–7. 15. Larue R, Klaassen R, Jochems A, et al. Pre‑treatment CT radiomics to pre ‑ 26. Ganeshan B, Skogen K, Pressney I, et al. Tumour heterogeneity in oesoph‑ dict 3‑ year overall survival following chemoradiotherapy of esophageal ageal cancer assessed by CT texture analysis: preliminary evidence of cancer. Acta Oncol. 2018;57:1475–81. an association with tumour metabolism, stage, and survival. Clin Radiol. 16. Rice TW, Gress DM, Patil DT, et al. Cancer of the esophagus and esoph‑ 2012;67:157–64. agogastric junction‑Major changes in the American Joint Committee 27. Zhang X, Wang Y, Qu P, et al. Prognostic value of tumor length for on Cancer eighth edition cancer staging manual. CA Cancer J Clin. cause‑specific death in resectable esophageal cancer. Ann Thorac Surg. 2017;67:304–17. 2018;106:1038–46. 17. Chun H, Xue‑jiao R, Lan W, et al. Evaluating short ‑term radiotherapeutic 28. Zhiguo Z, Xin W, Lan W, et al. Eec ff t of tumor length on clinical stage for effect on esophageal cancer by barium meal combined with CT scans. non‑ operative esophageal squamous cell carcinoma patients—multi‑ Chin J Radiat Oncol. 2013;22:26–9. center retrospective data analysis (3JECROG R‑01D). Chin J Radiat Oncol. 18. van Griethuysen JJM, Fedorov A, Parmar C, et al. Computational 2019;28:490–4. radiomics system to decode the radiographic phenotype. Cancer Res. 29. Liu SL, Xi M, Yang H, et al. Is There a correlation between clinical complete 2017;77:e104–7. response and pathological complete response after neoadjuvant chemo‑ 19. Li C, Ni W, Wang X, et al. A phase I/II radiation dose escalation trial using radiotherapy for esophageal squamous cell cancer? Ann Surg Oncol. simultaneous integrated boost technique with elective nodal irradia‑ 2016;23:273–81. tion and concurrent chemotherapy for unresectable esophageal cancer. 30. Li Z, Shan F, Wang Y, et al. Correlation of pathological complete response Radiat Oncol. 2019;14:48. with survival after neoadjuvant chemotherapy in gastric or gastroesoph‑ 20. Lin FC, Chang WL, Chiang NJ, et al. Radiation dose escalation can improve ageal junction cancer treated with radical surgery: a meta‑analysis. PLoS local disease control and survival among esophageal cancer patients ONE. 2018;13:e0189294. with large primary tumor volume receiving definitive chemoradiother ‑ 31. Lambin P, Leijenaar RTH, Deist TM, et al. Radiomics: the bridge between apy. PLoS ONE. 2020;15:e0237114. medical imaging and personalized medicine. Nat Rev Clin Oncol. 21. Zhang W, Luo Y, Wang X, et al. Dose‑ escalated radiotherapy improved 2017;14:749–62. survival for esophageal cancer patients with a clinical complete response after standard‑ dose radiotherapy with concurrent chemotherapy. Cancer Publisher’s Note Manag Res. 2018;10:2675–82. Springer Nature remains neutral with regard to jurisdictional claims in pub‑ 22. Zhang W, Zhao J, Han W, et al. Dose escalation of 3D radiotherapy is lished maps and institutional affiliations. effective for esophageal squamous cell carcinoma: a multicenter retro ‑ spective analysis (3JECROG R‑03). Ann Transl Med. 2020;8:1140. 23. Oppedijk V, van der Gaast A, van Lanschot JJ, et al. Patterns of recurrence after surgery alone versus preoperative chemoradiotherapy and surgery in the CROSS trials. J Clin Oncol. 2014;32:385–91. Re Read ady y to to submit y submit your our re researc search h ? Choose BMC and benefit fr ? Choose BMC and benefit from om: : fast, convenient online submission thorough peer review by experienced researchers in your field rapid publication on acceptance support for research data, including large and complex data types • gold Open Access which fosters wider collaboration and increased citations maximum visibility for your research: over 100M website views per year At BMC, research is always in progress. Learn more biomedcentral.com/submissions

Journal

Radiation OncologySpringer Journals

Published: Oct 12, 2021

Keywords: Chemo-radiotherapy; Esophageal squamous cell cancer; Radiomics; LPFS; Nomogram

There are no references for this article.