Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Semiautomated pelvic lymph node treatment response evaluation for patients with advanced prostate cancer: based on MET-RADS-P guidelines

Semiautomated pelvic lymph node treatment response evaluation for patients with advanced prostate... Background The evaluation of treatment response according to METastasis Reporting and Data System for Prostate Cancer (MET-RADS-P) criteria is an important but time-consuming task for patients with advanced prostate cancer (APC). A deep learning-based algorithm has the potential to assist with this assessment. Objective To develop and evaluate a deep learning-based algorithm for semiautomated treatment response assess- ment of pelvic lymph nodes. Methods A total of 162 patients who had undergone at least two scans for follow-up assessment after APC metasta- sis treatment were enrolled. A previously reported deep learning model was used to perform automated segmenta- tion of pelvic lymph nodes. The performance of the deep learning algorithm was evaluated using the Dice similarity coefficient (DSC) and volumetric similarity ( VS). The consistency of the short diameter measurement with the radiolo - gist was evaluated using Bland–Altman plotting. Based on the segmentation of lymph nodes, the treatment response was assessed automatically with a rule-based program according to the MET-RADS-P criteria. Kappa statistics were used to assess the accuracy and consistency of the treatment response assessment by the deep learning model and two radiologists [attending radiologist (R1) and fellow radiologist (R2)]. Results The mean DSC and VS of the pelvic lymph node segmentation were 0.82 ± 0.09 and 0.88 ± 0.12, respectively. Bland–Altman plotting showed that most of the lymph node measurements were within the upper and lower limits of agreement (LOA). The accuracies of automated segmentation-based assessment were 0.92 (95% CI: 0.85–0.96), 0.91 (95% CI: 0.86–0.95) and 75% (95% CI: 0.46–0.92) for target lesions, nontarget lesions and nonpathological lesions, respectively. The consistency of treatment response assessment based on automated segmentation and manual segmentation was excellent for target lesions [K value: 0.92 (0.86–0.98)], good for nontarget lesions [0.82 (0.74–0.90)] and moderate for nonpathological lesions [0.71 (0.50–0.92)]. Conclusion The deep learning-based semiautomated algorithm showed high accuracy for the treatment response assessment of pelvic lymph nodes and demonstrated comparable performance with radiologists. Xiang Liu and Zemin Zhu contributed equally to this manuscript. *Correspondence: Xiaoying Wang wangxiaoying@bjmu.edu.cn Full list of author information is available at the end of the article © The Author(s) 2023. 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. Liu et al. Cancer Imaging (2023) 23:7 Page 2 of 9 Keywords Deep learning, MET-RADS-P criteria, Pelvic lymph nodes, Metastases, DWI Background consent was waived due to its retrospective design. Advanced prostate cancer (APC) is characterized by the Two hundred and fifty-nine patients with histologi - recurrence of prostate cancer after definitive treatment or cally confirmed prostate cancer who underwent ini - by metastases without prior therapy [1]. Several therapeu- tial/curative treatment of metastases at our institution tic approaches have been approved for patients with APC. were included in this study between Jan 2017 and Jan Aside from the androgen deprivation and docetaxel treat- 2022. Pelvic MRI scans were performed before and ment, new agents with varying mechanisms of action have after at least one course of treatment (baseline and shown survival benefits in this population [2 , 3]. While the posttreatment). responses of patients with APC to these agents are various According to the MET-RADS-P criteria, lymph and treatment may cause side effects, they may result in nodes with a short diameter < 10  mm were considered the desired outcomes for patients. Therefore, early treat - nonpathological; therefore, only patients with lymph ment response assessment for patients with APC allows nodes ≥ 10  mm at baseline MRI should be included in clinicians to put a timely stop to unbeneficial treatment. the protocols. Hence, 23 of the 259 patients with APC Imagery depicting metastatic state plays a key role in were excluded because of the short diameter of all the patient management [4, 5]. There is a growing body of lesions < 10  mm. In addition, the time interval between research demonstrating how whole-body magnetic reso- baseline pelvic MRI and treatment initiation was sug- nance imaging can be used to diagnose and evaluate APC gested to be within 4  weeks; therefore, 45 patients were tumors and determine the efficacy of treatment [6 , 7]. excluded due to an interval of more than 4 weeks. Twelve The METastasis Reporting and Data System for Prostate patients were excluded because of the unqualified scan - Cancer (MET-RADS-P) guidelines aim to reduce vari- ning range on baseline and follow-up MRI. Fifteen ability in the acquisition, interpretation, and reporting of patients were excluded for inadequate image quality. metastatic cancer by promoting standardization of prac- Finally, 162 patients who had undergone at least two tices [8]. As recommended by the Prostate Cancer Clini- scans for follow-up assessment after APC metastasis cal Trials Working Group (PCWG), MET-RADS-P allows treatment were analyzed (Fig.  1). Clinical and radiologi- the subclassification of patients based on their metastatic cal features of the enrolled patients were acquired from spread pattern (bone, nodal, visceral, or local) [5]. the electronic information system, including age, pros- Diffusion-weighted imaging (DWI) has been shown tate-specific antigen (PSA) level, PI-RADS v2.1 scores to successfully reflect tumor response and discriminate and TNM staging. between future responders and nonresponders, which could be valuable in adapting future management [9]. MRI acquisition Manual segmentation and measurement of DWI lesions Three 3.0 T scanners were used (Achieva, Philips Health - based on MET-RADS-P require a high level of exper- care; Discovery MR750, GE Healthcare; Intera, Philips tise, are time-consuming, and are subject to operator Healthcare) to perform pelvic MRI scans. The pelvic error [10, 11]. Deep learning technologies have extended MRI protocol performed in our institution included this quantitative approach with promising preliminary T2-weighted imaging (T2WI), T1WI, DWI with appar- results in the assessment of tumor response in the liver ent diffusion coefficient (ADC) maps and dynamic gad - [12, 13]. In this study, we hypothesized that the deep olinium-DTPA (Gd-DTPA)-enhanced (DCE) sequences. learning model could also be trained to estimate the The detailed scanning parameters of DWI are listed in treatment response of APC according to MET-RADS-P Table 1. guidelines. This study aimed to investigate the feasibility of deep learning-based treatment response evaluation of patients with APC, and for proof-of-concept, we focused Pelvic lymph nodes segmentation on the assessment in the pelvic lymph nodes. A previously trained 3D U-Net segmentation model developed by the same authors in this study based on deep learning was used to automatically segment the visi- Materials and methods ble pelvic lymph nodes on DWI images [14]. The training Patient enrollment data used for the model development were different from This study was approved by the local institutional the data included here. All visible lymph nodes included review board, and the requirement for informed target lesions (short diameter ≥ 15  mm), nontarget Liu  et al. Cancer Imaging (2023) 23:7 Page 3 of 9 Fig. 1 The workflow of patient enrollment Table 1 The detailed imaging parameters for diffusion-weighted made by a radiologist expert (with more than 20 years of imaging reading experience) were considered the reference stand- ard for segmentation evaluation. Parameters 3.0 T Discovery 3.0 T Intera 3.0 T Achieva B value (s/mm ) 0, 800 0, 800 0, 800 Imaging matrix 256 × 256 224 × 224 156 × 180 Treatment response assessment Echo time (ms) 60 56 54 Based on the MET-RADS-P criteria, treatment response Repetition time (ms) 4000 6628 3300 assessments of lymph nodes were conducted [15], includ- Field of view (mm ) 450 × 366 662 × 400 512 × 356 ing complete response (CR), partial response (PR), stable Section thickness (mm) 4 4 4 disease (SD), and progressive disease (PD). Number of slices 25 20 24 The radiologists who corrected the lymph nodes manually provided the reference standard for treat- ment response assessment. An algorithm for semiau- tomatic response assessment was developed using the lesions (10  mm ≤ short diameter < 15  mm) and non- MET-RADS-P criteria by automatically calculating the pathological lesions (short diameter < 10  mm). Manual diameters of the lymph nodes first and then assessing corrections of the automatically segmented lymph nodes the treatment response by a rule-based program. More Liu et al. Cancer Imaging (2023) 23:7 Page 4 of 9 Fig. 2 The flowchart of treatment response assessment of lymph nodes details about the algorithm development of pelvic PI-RADS scores and T/N/M staging were recorded lymph nodes were shown in our previous study [14]. from the baseline MRI reports, and PI-RADS 5 In addition, an attending radiology radiologist (R1) (74.07%), T4 (30.86%), N1 (56.79%) and M0 (38.89%) and a fellow radiology radiologist (R2), with 8 and accounted for the largest proportion. The Gleason 4  years of pelvic imaging experience, performed the scores were obtained from the pathological report, treatment response assessments on all patients by pri- and Gleason 4 + 5 (37.65%) accounted for the largest mary review of the MRI images. The two radiologists percentage. compared baseline scans before treatment and subse- All patients had received at least one course of post- quent scans after treatment for every patient. The defi- treatment MRI examination, 63 patients had two nition and evaluation rules are shown in Fig. 2. posttreatment examinations, 23 patients had three posttreatment examinations, 8 patients had four post- treatment examinations, 3 patients had five post - Statistical analysis treatment examinations, and 1 patient had seven The “median (interquartile range)” values are used for posttreatment examinations. In the baseline pelvic the description of continuous variables, and descrip- MRI, 112 patients had target lesions, 129 patients had tive statistics of the categorical data are presented with nontarget lesions, and all patients had nonpathological “n (%)”. The segmentation results are quantitatively lymph nodes. evaluated by the overlap-based metric [Dice similar- ity coefficient (DSC)] and the volume-based metric Assessment of automated lymph node segmentation [volumetric similarity (VS)] [16]. The independent One hundred and sixty-two APC patients with 162 base- t-test was applied to determine the difference in the line pelvic MRI scans and 260 posttreatment MRI scans evaluation metrics between the subgroups. We used were used to perform automated lymph node segmen- the Kappa statistic to evaluate the consistency of treat- tation. As shown in Table  3, the mean DSC and VS are ment response. A P value less than 0.05 was treated 0.82 ± 0.09 and 0.88 ± 0.12, respectively. In the subgroup as significant. Statistical analysis was performed with analyses, the DSC and VS values of the target lesions and MedCalc (version 14.8; MedCalc Software, Ostend, nontarget lesions showed no significant difference (DSC: Belgium). 0.85 vs. 0.82, P > 0.05; VS: 0.88 vs. 0.86, P > 0.05) but were significantly higher than those of nonpathological lesions Results (all P values > 0.05). The subgroups of baseline and post - Study population treatment MRI scans showed no significant difference In this study, 162 eligible APC patients with metasta- (all P values > 0.05). The explementary segmentation of ses were included. The baseline characteristics of the lymph nodes is shown in Fig. 3. enrolled patients are shown in Table  2. The median T-PSA level in this population was 35.39  ng/ml. The Liu  et al. Cancer Imaging (2023) 23:7 Page 5 of 9 Table 2 Main baseline demographics and clinical characteristics (interquartile range, 15.93—26.77  mm), respectively of patients (P = 0.231). The mean short diameters of automatically segmented and manually segmented nontarget lesions Characteristics Value were 11.91  mm (interquartile range, 10.85—13.14  mm) Age(y) 69 (64, 75) and 12.33  mm (interquartile range, 11.07—13.59  mm), PSA (ng/ml) respectively (P = 0.082). The agreement between the T-PSA 35.39 (14.2, 70) automatically segmented and manually segmented target F-PSA 4.32 (1.53, 9.00) lesions and nontarget lesions in terms of short diameter is F/T-PSA 0.11 (0.07, 0.17) shown in Fig. 4. The Bland–Altman analysis showed good PI-RADS scores (n%) consistency between the automated segmentation and 3 7 (4.32%) manual segmentation, and most values were within the 4 35 (21.60%) upper and lower limits of agreement (LOA). 5 120 (74.07%) T staging (n%) Accuracy of the treatment response assessment T2 42 (25.93%) In this population, 75 APC patients with 112 pairs of T3a 21 (12.96%) pelvic MRI performed the target lesion evaluation; 129 T3b 49 (30.25%) APC patients with 209 pairs of pelvic MRI performed the T4 50 (30.86%) nontarget lesion evaluation, and 162 APC patients with N staging (n%) 260 pairs of pelvic MRI performed the nonpathological X 52 (32.10%) lesion evaluation. As shown in Fig.  5, the accuracies of 0 18 (11.11%) the automated segmentation-based response assessment 1 92 (56.79%) were 0.92 (95% CI: 0.85–0.96), 0.91 (95% CI: 0.86–0.95) M staging (n%) and 75% (95% CI: 0.46–0.92) for target lesions, nontarget X 56 (34.57%) lesions and nonpathological lesions, respectively. 0 63 (38.89%) 1a 3 (1.85%) Consistency of the treatment response assessment 1b 40 (24.69%) As shown in Table  4, the agreement of treatment Gleason score (n%) response assessment based on automated segmentation 3 + 3 10 (6.17%) and manual correction was excellent for target lesions [K 3 + 4 10 (6.17%) value: 0.92 (0.86–0.98)], good for nontarget lesions [0.82 3 + 5 6 (3.70%) (0.74–0.90)] and moderate for nonpathological lesions 4 + 3 24 (14.81%) [0.71 (0.50–0.92)], which were approximately equal to the 4 + 4 25 (15.43%) agreement between R1 and manual correction [0.89, 0.81 4 + 5 61 (37.65%) and 0.68 for target lesions, nontarget lesions and non- 5 + 4 23 (14.20%) pathological lesions, respectively] but slightly higher than 5 + 5 3 (1.85%) the agreement between R2 and the reference standard [0.86, 0.82 and 0.60 for target lesions, nontarget lesions and nonpathological lesions, respectively]. Quantitative measurement of the lymph node segmentation Discussion The mean short diameters of the automatically segmented MET-RADS-P is a guideline for the treatment response and manually segmented target lesions were 23.53  mm evaluation of systemic metastases of patients with APC, (interquartile range, 17.61- 26.55  mm) and 27.94  mm which involves the evaluation of primary focus, bone Table 3 Segmentation results of pelvic lymph nodes Metrics All Subgroup analysis Target lesions Nontarget lesions Nonpathological Baseline lesions Post- lesions treatment lesions DSC 0.82 ± 0.09 0.85 ± 0.09 0.82 ± 0.09 0.78 ± 0.09 0.81 ± 0.10 0.82 ± 0.09 VS 0.88 ± 0.12 0.88 ± 0.09 0.86 ± 0.08 0.80 ± 0.08 0.87 ± 0.09 0.88 ± 0.08 DSC Dice similarity coefficient, VS volumetric similarity. The DSC and VS values were used to evaluate the performance of the automated lymph node segmentation by comparison with the manual annotation Liu et al. Cancer Imaging (2023) 23:7 Page 6 of 9 Fig. 3 Explementary results of lymph node segmentation and correction. Light green: target lesion; light blue: nontarget lesion; light yellow: nonpathological lesion Fig. 4 Agreement between the automatically segmented and manually segmented lymph nodes. a target lesions; b nontarget lesions Fig. 5 Confusion matrix of treatment response assessment Liu  et al. Cancer Imaging (2023) 23:7 Page 7 of 9 Table 4 The consistency of treatment response assessment Comparison Target lesions Nontarget lesions Nonpathological lesions Automated segmentation vs. Manual correction 0.92 (0.86–0.98) 0.82 (0.74–0.90) 0.71 (0.50–0.92) R1 vs. Automated segmentation 0.89 (0.81–0.96) 0.81 (0.73–0.89) 0.68 (0.47–0.88) R2 vs. Automated segmentation 0.86 (0.79–0.94) 0.78 (0.69–0.87) 0.60 (0.37–0.82) R1 vs. R2 0.90 (0.85–0.97) 0.95 (0.90–0.99) 0.84 (0.68–0.99) R1 vs. Manual correction 0.96 (0.93–1.00) 0.99 (0.97–1.00) 0.96 (0.88–1.00) R2 vs. Manual correction 0.94 (0.89–0.99) 0.96 (0.92–0.99) 0.88 (0.73–1.00) R1 an attending radiologist with 8 years of reading experience, R2 a fellow radiologist with 4 years of reading experience metastases, lymph node metastases and organ metasta- according to MET-RADS-P criteria included two parts. ses. In this study, we established a semiautomatic pelvic First, a previously established pelvic lymph node seg- lymph node treatment response evaluation process for mentation model was used to perform the automatic patients with APC through lymph node segmentation segmentation of lymph nodes. The model achieved based on deep learning. Our results showed that the good segmentation performance here, which is similar accuracies of automated segmentation-based response to the segmentation results reported in previous litera- assessment were high for all the target lesions, nontarget ture (the DSC and VS values for all visible lymph nodes lesions and nonpathological lesions according to MET- were 0.76 ± 0.15 and 0.82 ± 0.14, respectively) [14], espe- RADS-P criteria and achieved good consistency with the cially the target lesions, further highlighting its potential attending radiologist and fellow radiologist. usefulness. Based on the morphology and signal characteristics of Second, based on the quantitative measurements all acquired images, the MET-RADS-P system mapped obtained from the automated segmentation, we can unequivocal diseases to 14 predefined body regions [8, directly evaluate the treatment response according to 15]. Analysis of lymph node metastases in the pelvis is MET-RADS-P criteria, which can be more practical in crucial for clinical practice and drug studies in patients clinical settings. A clinical radiology report provides a with APC, which is the most common metastatic site qualitative narrative, but does not provide standardized, [17]. A lymph node’s size is highly correlated with sur- quantitative information about the patient’s progress or vival time, a measurement that radiologists and clini- response to treatment [21]. Natural language processing cians perform to monitor disease progression or assess and deep learning models have been employed in previous therapeutic options, due to the fact that many malig- studies to estimate responses from clinical text [22, 23]. nancies can enlarge lymph nodes [18]. According to These approaches can be feasible for quantitative assess - the Response Evaluation Criteria in Solid Tumors 1.1 ment related to MET-RADS-P criteria but can be indirect. (RECIST 1.1) Guidelines, lymph nodes with a short-axis Our proposed semiautomated algorithm achieved high diameter of at least 10 mm are considered to be enlarged Kappa values in terms of treatment response assessment lymph nodes and are clinically significant [19]. The size with attending and fellow radiologists when measuring standard of pathological lymph nodes defined by MET- the same set of target and nontarget lesions. The con - RADS-P based on MRI was similar to RECIST 1.1, while sistency of nonpathological lesions was lower, which MET-RADS-P provides a more complete assessment of may be due to the relatively poor segmentation perfor- nodal metastases response including the nontarget nodes mance. Tang et  al. [24] proposed a deep learning-based and nonpathologic nodes, which was usually qualitatively method for semiautomated RECISTS measurement and assessed by RECIST 1.1 criteria. assessed using a mean difference between the deep learn - According to the MET-RADS-P criteria, the core whole ing algorithm and manual measurement in the unit of body MRI protocol designed for bone and lymph node pixels. Scores using pixel difference, however, may not metastasis detection included T1WI (GRE Dixon tech- be reliable, as scores are largely determined by data com- nique) and axial DWI [8]. DWI is a well-recognized and position. In this study, we used Bland–Altman plotting used sequence for pelvic lymph node imaging, that is able based on percent measurement difference to address the to offer qualitative and quantitative assessments for dis - issue as suggested by Woo et  al. [25]. As demonstrated, ease characterizations [14, 20]. Therefore, in this study, the Bland–Altman analysis indicated good consistency we performed the treatment response assessment only between the automated segmentation and manual seg- on DWI images. mentation, and most values were within the upper and In this study, the established semiautomatic pelvic lower LOA. lymph node treatment response evaluation process Liu et al. Cancer Imaging (2023) 23:7 Page 8 of 9 was written by Xiang Liu and Zemin Zhu, and all authors commented on There are some limitations that need to be addressed. previous versions of the manuscript. All authors read and approved the final First, in this study, the deep learning-based treatment manuscript. response assessment was only focused on the pelvic Funding lymph node, and other regions of the body according to This work was supported by the Capital’s Funds for Health Improvement and the MET-RADS-P guideline need to be investigated in Research (2020–2-40710) and Innovation Fund for Outstanding Doctoral Can- the future. Second, we acknowledge that there remain didates of Peking University Health Science Centre (BMU2022BSS001). opportunities for further model refinement, including Availability of data and materials the achievement of lymph node registration between The datasets used and/or analyzed during the current study are available from baseline and posttreatment images, thus realizing fully the corresponding author on reasonable request. automated lymph node treatment response evaluation. Finally, our results demonstrated that the semiautomated Declarations treatment response assessment can be achieved on the Ethics approval and consent to participate DWI sequence, but the values of other sequences (e.g. This study was performed in accordance with the principles of the Declaration T1WI, DCE or T2WI) on response assessment also need of Helsinki and was approved by the Committee for Medical Ethics, Peking University First Hospital (2021–060). Informed consent was waived according to be investigated in further studies. to its retrospective design. Consent for publication Not applicable. Conclusion In conclusion, we have developed a semiautomated deep Competing interests learning-based model to estimate response assessments Yaofeng Zhang, Jialun Li and Xiangpeng Wang are from a medical technical corporation provided technical support for model development. The authors of pelvic lymph nodes in patients with APC. The accu - declare that they have no competing interests. racy of response assessments based on the automati- cally segmented lymph nodes showed close similarity to Author details Department of Radiology, Peking University First Hospital, No.8 Xishiku the manually segmented lymph nodes and yielded out- Street, Xicheng District, Beijing 100034, China. Department of Hepatobiliary put comparable to the radiologists. These initial results and Pancreatic Surgery, Zhuzhou Central Hospital, Zhuzhou 412000, China. provide a promising way to achieve a fully automated School of Basic Medical Sciences, Capital Medical University, Beijing 100069, China. Beijing Smart Tree Medical Technology Co. Ltd, Beijing 100011, China. treatment response assessment algorithm according to MET-RADS-P criteria. Received: 4 October 2022 Accepted: 5 January 2023 Abbreviations ADC Apparent diffusion coefficient APC Advanced prostate cancer References CR Complete response 1. Teo MY, Rathkopf DE, Kantoff P. Treatment of Advanced Prostate Cancer. DCE Dynamic contrast-enhanced Annu Rev Med. 2019;70:479–99. DSC Dice similarity coefficient 2. Komura K, Sweeney CJ, Inamoto T, Ibuki N, Azuma H, Kantoff PW. DWI Diffusion-weighted imaging Current treatment strategies for advanced prostate cancer. Int J Urol. LOA Limits of agreement 2018;25(3):220–31. MET-RADS-P METastasis Reporting and Data System for Prostate Cancer 3. Swami U, McFarland TR, Nussenzveig R, Agarwal N. Advanced pros- PD Progressive disease tate cancer: treatment advances and future directions. Trends Cancer. PR Partial response 2020;6(8):702–15. PCWG P rostate Cancer Clinical Trials Working Group 4. Halabi S, Kelly WK, Ma H, Zhou H, Solomon NC, Fizazi K, et al. Meta- PSA Prostate-specific antigen analysis evaluating the impact of site of metastasis on overall sur- SD Stable disease vival in men with castration-resistant prostate cancer. J Clin Oncol. T2WI T2-weighted imaging 2016;34(14):1652–9. VS Volumetric similarity 5. Scher HI, Morris MJ, Stadler WM, Higano C, Basch E, Fizazi K, et al. Trial design and objectives for castration-resistant prostate cancer: updated Acknowledgements recommendations from the prostate cancer clinical trials working group Not applicable. 3. J Clin Oncol. 2016;34(12):1402–18. 6. Abern MR, Tsivian M, Polascik TJ. Focal therapy of prostate cancer: Code availability evidence-based analysis for modern selection criteria. Curr Urol Rep. The codes used for the development the algorithm are available from the 2012;13(2):160–9. corresponding author on reasonable request. 7. Lecouvet FE, Talbot JN, Messiou C, Bourguet P, Liu Y, de Souza NM. Monitoring the response of bone metastases to treatment with Magnetic Authors’ contributions Resonance Imaging and nuclear medicine techniques: a review and posi- All authors contributed to the study conception and design. Material prepa- tion statement by the European Organisation for Research and Treatment ration, data collection and analysis were performed by Xiang Liu and Kexin of Cancer imaging group. Eur J Cancer. 2014;50(15):2519–31. Wang. Xiang Liu performed manual segmentation under the supervision 8. Padhani AR, Lecouvet FE, Tunariu N, Koh DM, De Keyzer F, Collins DJ, of Xiaoying Wang. Xiaodong Zhang participated in the image interpreta- et al. METastasis reporting and data system for prostate cancer: practical tion. Zemin Zhu, Jialun Li, Yaofeng Zhang, and Xiangpeng Wang performed guidelines for acquisition, interpretation, and reporting of whole-body data interpretation and statistical analysis. The first draft of the manuscript Liu  et al. Cancer Imaging (2023) 23:7 Page 9 of 9 magnetic resonance imaging-based evaluations of multiorgan involve- ment in advanced prostate cancer. Eur Urol. 2017;71(1):81–92. 9. Cook GJR, Goh V. Molecular Imaging of Bone Metastases and Their Response to Therapy. J Nucl Med. 2020;61(6):799–806. 10. Liu X, Han C, Cui Y, Xie T, Zhang X, Wang X. Detection and segmentation of pelvic bones metastases in MRI images for patients with prostate cancer based on deep learning. Front Oncol. 2021;11:773299. 11. Liu X, Wang X, Zhang Y, Sun Z, Zhang X, Wang X. Preoperative prediction of pelvic lymph nodes metastasis in prostate cancer using an ADC-based radiomics model: comparison with clinical nomograms and PI-RADS assessment. Abdom Radiol (NY ). 2022;47(9):3327–37. 12. Aerts HJ. The Potential of Radiomic-Based Phenotyping in Precision Medicine: A Review. JAMA Oncol. 2016;2(12):1636–42. 13. Xu X, Zhang HL, Liu QP, Sun SW, Zhang J, Zhu FP, et al. Radiomic analysis of contrast-enhanced CT predicts microvascular invasion and outcome in hepatocellular carcinoma. J Hepatol. 2019;70(6):1133–44. 14. Liu X, Sun Z, Han C, Cui Y, Huang J, Wang X, et al. Development and validation of the 3D U-Net algorithm for segmentation of pelvic lymph nodes on diffusion-weighted images. BMC Med Imaging. 2021;21(1):170. 15. Padhani AR, Tunariu N. Metastasis reporting and data system for prostate cancer in practice. Magn Reson Imaging Clin N Am. 2018;26(4):527–42. 16. Taha AA, Hanbury A. Metrics for evaluating 3D medical image segmenta- tion: analysis, selection, and tool. BMC Med Imaging. 2015;15:29. 17. Kim YJ, Song C, Eom KY, Kim IA, Kim JS. Lymph node ratio determines the benefit of adjuvant radiotherapy in pathologically 3 or less lymph node-positive prostate cancer after radical prostatectomy: a popu- lation-based analysis with propensity-score matching. Oncotarget. 2017;8(66):110625–34. 18. Alheejawi S, Xu H, Berendt R, Jha N, Mandal M. Novel lymph node segmentation and proliferation index measurement for skin melanoma biopsy images. Comput Med Imaging Graph. 2019;73:19–29. 19. Eisenhauer EA, Therasse P, Bogaerts J, Schwartz LH, Sargent D, Ford R, et al. New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur J Cancer. 2009;45(2):228–47. 20. Eiber M, Beer AJ, Holzapfel K, Tauber R, Ganter C, Weirich G, et al. Pre- liminary results for characterization of pelvic lymph nodes in patients with prostate cancer by diffusion-weighted MR-imaging. Invest Radiol. 2010;45(1):15–23. 21. Arbour KC, Luu AT, Luo J, Rizvi H, Plodkowski AJ, Sakhi M, et al. Deep learning to estimate RECIST in patients with NSCLC treated with PD-1 blockade. Cancer Discov. 2021;11(1):59–67. 22. Chen MC, Ball RL, Yang L, Moradzadeh N, Chapman BE, Larson DB, et al. Deep learning to classify radiology free-text reports. Radiology. 2018;286(3):845–52. 23. Bozkurt S, Alkim E, Banerjee I, Rubin DL. Automated detection of meas- urements and their descriptors in radiology reports using a hybrid natural language processing algorithm. J Digit Imaging. 2019;32(4):544–53. 24. Chlebus G, Schenk A, Moltz JH, van Ginneken B, Hahn HK, Meine H. Automatic liver tumor segmentation in CT with fully convolutional neural networks and object-based postprocessing. Sci Rep. 2018;8(1):15497. 25. Woo M, Devane AM, Lowe SC, Lowther EL, Gimbel RW. Deep learning for semi-automated unidirectional measurement of lung tumor size in CT. Cancer Imaging. 2021;21(1):43. Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in pub- 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: : lished maps and institutional affiliations. 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 Cancer Imaging Springer Journals

Semiautomated pelvic lymph node treatment response evaluation for patients with advanced prostate cancer: based on MET-RADS-P guidelines

Download PDF
9 pages

Loading next page...
 
/lp/springer-journals/semiautomated-pelvic-lymph-node-treatment-response-evaluation-for-rqfm2BMUfH

References (28)

Publisher
Springer Journals
Copyright
Copyright © The Author(s) 2023
eISSN
1470-7330
DOI
10.1186/s40644-023-00523-4
Publisher site
See Article on Publisher Site

Abstract

Background The evaluation of treatment response according to METastasis Reporting and Data System for Prostate Cancer (MET-RADS-P) criteria is an important but time-consuming task for patients with advanced prostate cancer (APC). A deep learning-based algorithm has the potential to assist with this assessment. Objective To develop and evaluate a deep learning-based algorithm for semiautomated treatment response assess- ment of pelvic lymph nodes. Methods A total of 162 patients who had undergone at least two scans for follow-up assessment after APC metasta- sis treatment were enrolled. A previously reported deep learning model was used to perform automated segmenta- tion of pelvic lymph nodes. The performance of the deep learning algorithm was evaluated using the Dice similarity coefficient (DSC) and volumetric similarity ( VS). The consistency of the short diameter measurement with the radiolo - gist was evaluated using Bland–Altman plotting. Based on the segmentation of lymph nodes, the treatment response was assessed automatically with a rule-based program according to the MET-RADS-P criteria. Kappa statistics were used to assess the accuracy and consistency of the treatment response assessment by the deep learning model and two radiologists [attending radiologist (R1) and fellow radiologist (R2)]. Results The mean DSC and VS of the pelvic lymph node segmentation were 0.82 ± 0.09 and 0.88 ± 0.12, respectively. Bland–Altman plotting showed that most of the lymph node measurements were within the upper and lower limits of agreement (LOA). The accuracies of automated segmentation-based assessment were 0.92 (95% CI: 0.85–0.96), 0.91 (95% CI: 0.86–0.95) and 75% (95% CI: 0.46–0.92) for target lesions, nontarget lesions and nonpathological lesions, respectively. The consistency of treatment response assessment based on automated segmentation and manual segmentation was excellent for target lesions [K value: 0.92 (0.86–0.98)], good for nontarget lesions [0.82 (0.74–0.90)] and moderate for nonpathological lesions [0.71 (0.50–0.92)]. Conclusion The deep learning-based semiautomated algorithm showed high accuracy for the treatment response assessment of pelvic lymph nodes and demonstrated comparable performance with radiologists. Xiang Liu and Zemin Zhu contributed equally to this manuscript. *Correspondence: Xiaoying Wang wangxiaoying@bjmu.edu.cn Full list of author information is available at the end of the article © The Author(s) 2023. 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. Liu et al. Cancer Imaging (2023) 23:7 Page 2 of 9 Keywords Deep learning, MET-RADS-P criteria, Pelvic lymph nodes, Metastases, DWI Background consent was waived due to its retrospective design. Advanced prostate cancer (APC) is characterized by the Two hundred and fifty-nine patients with histologi - recurrence of prostate cancer after definitive treatment or cally confirmed prostate cancer who underwent ini - by metastases without prior therapy [1]. Several therapeu- tial/curative treatment of metastases at our institution tic approaches have been approved for patients with APC. were included in this study between Jan 2017 and Jan Aside from the androgen deprivation and docetaxel treat- 2022. Pelvic MRI scans were performed before and ment, new agents with varying mechanisms of action have after at least one course of treatment (baseline and shown survival benefits in this population [2 , 3]. While the posttreatment). responses of patients with APC to these agents are various According to the MET-RADS-P criteria, lymph and treatment may cause side effects, they may result in nodes with a short diameter < 10  mm were considered the desired outcomes for patients. Therefore, early treat - nonpathological; therefore, only patients with lymph ment response assessment for patients with APC allows nodes ≥ 10  mm at baseline MRI should be included in clinicians to put a timely stop to unbeneficial treatment. the protocols. Hence, 23 of the 259 patients with APC Imagery depicting metastatic state plays a key role in were excluded because of the short diameter of all the patient management [4, 5]. There is a growing body of lesions < 10  mm. In addition, the time interval between research demonstrating how whole-body magnetic reso- baseline pelvic MRI and treatment initiation was sug- nance imaging can be used to diagnose and evaluate APC gested to be within 4  weeks; therefore, 45 patients were tumors and determine the efficacy of treatment [6 , 7]. excluded due to an interval of more than 4 weeks. Twelve The METastasis Reporting and Data System for Prostate patients were excluded because of the unqualified scan - Cancer (MET-RADS-P) guidelines aim to reduce vari- ning range on baseline and follow-up MRI. Fifteen ability in the acquisition, interpretation, and reporting of patients were excluded for inadequate image quality. metastatic cancer by promoting standardization of prac- Finally, 162 patients who had undergone at least two tices [8]. As recommended by the Prostate Cancer Clini- scans for follow-up assessment after APC metastasis cal Trials Working Group (PCWG), MET-RADS-P allows treatment were analyzed (Fig.  1). Clinical and radiologi- the subclassification of patients based on their metastatic cal features of the enrolled patients were acquired from spread pattern (bone, nodal, visceral, or local) [5]. the electronic information system, including age, pros- Diffusion-weighted imaging (DWI) has been shown tate-specific antigen (PSA) level, PI-RADS v2.1 scores to successfully reflect tumor response and discriminate and TNM staging. between future responders and nonresponders, which could be valuable in adapting future management [9]. MRI acquisition Manual segmentation and measurement of DWI lesions Three 3.0 T scanners were used (Achieva, Philips Health - based on MET-RADS-P require a high level of exper- care; Discovery MR750, GE Healthcare; Intera, Philips tise, are time-consuming, and are subject to operator Healthcare) to perform pelvic MRI scans. The pelvic error [10, 11]. Deep learning technologies have extended MRI protocol performed in our institution included this quantitative approach with promising preliminary T2-weighted imaging (T2WI), T1WI, DWI with appar- results in the assessment of tumor response in the liver ent diffusion coefficient (ADC) maps and dynamic gad - [12, 13]. In this study, we hypothesized that the deep olinium-DTPA (Gd-DTPA)-enhanced (DCE) sequences. learning model could also be trained to estimate the The detailed scanning parameters of DWI are listed in treatment response of APC according to MET-RADS-P Table 1. guidelines. This study aimed to investigate the feasibility of deep learning-based treatment response evaluation of patients with APC, and for proof-of-concept, we focused Pelvic lymph nodes segmentation on the assessment in the pelvic lymph nodes. A previously trained 3D U-Net segmentation model developed by the same authors in this study based on deep learning was used to automatically segment the visi- Materials and methods ble pelvic lymph nodes on DWI images [14]. The training Patient enrollment data used for the model development were different from This study was approved by the local institutional the data included here. All visible lymph nodes included review board, and the requirement for informed target lesions (short diameter ≥ 15  mm), nontarget Liu  et al. Cancer Imaging (2023) 23:7 Page 3 of 9 Fig. 1 The workflow of patient enrollment Table 1 The detailed imaging parameters for diffusion-weighted made by a radiologist expert (with more than 20 years of imaging reading experience) were considered the reference stand- ard for segmentation evaluation. Parameters 3.0 T Discovery 3.0 T Intera 3.0 T Achieva B value (s/mm ) 0, 800 0, 800 0, 800 Imaging matrix 256 × 256 224 × 224 156 × 180 Treatment response assessment Echo time (ms) 60 56 54 Based on the MET-RADS-P criteria, treatment response Repetition time (ms) 4000 6628 3300 assessments of lymph nodes were conducted [15], includ- Field of view (mm ) 450 × 366 662 × 400 512 × 356 ing complete response (CR), partial response (PR), stable Section thickness (mm) 4 4 4 disease (SD), and progressive disease (PD). Number of slices 25 20 24 The radiologists who corrected the lymph nodes manually provided the reference standard for treat- ment response assessment. An algorithm for semiau- tomatic response assessment was developed using the lesions (10  mm ≤ short diameter < 15  mm) and non- MET-RADS-P criteria by automatically calculating the pathological lesions (short diameter < 10  mm). Manual diameters of the lymph nodes first and then assessing corrections of the automatically segmented lymph nodes the treatment response by a rule-based program. More Liu et al. Cancer Imaging (2023) 23:7 Page 4 of 9 Fig. 2 The flowchart of treatment response assessment of lymph nodes details about the algorithm development of pelvic PI-RADS scores and T/N/M staging were recorded lymph nodes were shown in our previous study [14]. from the baseline MRI reports, and PI-RADS 5 In addition, an attending radiology radiologist (R1) (74.07%), T4 (30.86%), N1 (56.79%) and M0 (38.89%) and a fellow radiology radiologist (R2), with 8 and accounted for the largest proportion. The Gleason 4  years of pelvic imaging experience, performed the scores were obtained from the pathological report, treatment response assessments on all patients by pri- and Gleason 4 + 5 (37.65%) accounted for the largest mary review of the MRI images. The two radiologists percentage. compared baseline scans before treatment and subse- All patients had received at least one course of post- quent scans after treatment for every patient. The defi- treatment MRI examination, 63 patients had two nition and evaluation rules are shown in Fig. 2. posttreatment examinations, 23 patients had three posttreatment examinations, 8 patients had four post- treatment examinations, 3 patients had five post - Statistical analysis treatment examinations, and 1 patient had seven The “median (interquartile range)” values are used for posttreatment examinations. In the baseline pelvic the description of continuous variables, and descrip- MRI, 112 patients had target lesions, 129 patients had tive statistics of the categorical data are presented with nontarget lesions, and all patients had nonpathological “n (%)”. The segmentation results are quantitatively lymph nodes. evaluated by the overlap-based metric [Dice similar- ity coefficient (DSC)] and the volume-based metric Assessment of automated lymph node segmentation [volumetric similarity (VS)] [16]. The independent One hundred and sixty-two APC patients with 162 base- t-test was applied to determine the difference in the line pelvic MRI scans and 260 posttreatment MRI scans evaluation metrics between the subgroups. We used were used to perform automated lymph node segmen- the Kappa statistic to evaluate the consistency of treat- tation. As shown in Table  3, the mean DSC and VS are ment response. A P value less than 0.05 was treated 0.82 ± 0.09 and 0.88 ± 0.12, respectively. In the subgroup as significant. Statistical analysis was performed with analyses, the DSC and VS values of the target lesions and MedCalc (version 14.8; MedCalc Software, Ostend, nontarget lesions showed no significant difference (DSC: Belgium). 0.85 vs. 0.82, P > 0.05; VS: 0.88 vs. 0.86, P > 0.05) but were significantly higher than those of nonpathological lesions Results (all P values > 0.05). The subgroups of baseline and post - Study population treatment MRI scans showed no significant difference In this study, 162 eligible APC patients with metasta- (all P values > 0.05). The explementary segmentation of ses were included. The baseline characteristics of the lymph nodes is shown in Fig. 3. enrolled patients are shown in Table  2. The median T-PSA level in this population was 35.39  ng/ml. The Liu  et al. Cancer Imaging (2023) 23:7 Page 5 of 9 Table 2 Main baseline demographics and clinical characteristics (interquartile range, 15.93—26.77  mm), respectively of patients (P = 0.231). The mean short diameters of automatically segmented and manually segmented nontarget lesions Characteristics Value were 11.91  mm (interquartile range, 10.85—13.14  mm) Age(y) 69 (64, 75) and 12.33  mm (interquartile range, 11.07—13.59  mm), PSA (ng/ml) respectively (P = 0.082). The agreement between the T-PSA 35.39 (14.2, 70) automatically segmented and manually segmented target F-PSA 4.32 (1.53, 9.00) lesions and nontarget lesions in terms of short diameter is F/T-PSA 0.11 (0.07, 0.17) shown in Fig. 4. The Bland–Altman analysis showed good PI-RADS scores (n%) consistency between the automated segmentation and 3 7 (4.32%) manual segmentation, and most values were within the 4 35 (21.60%) upper and lower limits of agreement (LOA). 5 120 (74.07%) T staging (n%) Accuracy of the treatment response assessment T2 42 (25.93%) In this population, 75 APC patients with 112 pairs of T3a 21 (12.96%) pelvic MRI performed the target lesion evaluation; 129 T3b 49 (30.25%) APC patients with 209 pairs of pelvic MRI performed the T4 50 (30.86%) nontarget lesion evaluation, and 162 APC patients with N staging (n%) 260 pairs of pelvic MRI performed the nonpathological X 52 (32.10%) lesion evaluation. As shown in Fig.  5, the accuracies of 0 18 (11.11%) the automated segmentation-based response assessment 1 92 (56.79%) were 0.92 (95% CI: 0.85–0.96), 0.91 (95% CI: 0.86–0.95) M staging (n%) and 75% (95% CI: 0.46–0.92) for target lesions, nontarget X 56 (34.57%) lesions and nonpathological lesions, respectively. 0 63 (38.89%) 1a 3 (1.85%) Consistency of the treatment response assessment 1b 40 (24.69%) As shown in Table  4, the agreement of treatment Gleason score (n%) response assessment based on automated segmentation 3 + 3 10 (6.17%) and manual correction was excellent for target lesions [K 3 + 4 10 (6.17%) value: 0.92 (0.86–0.98)], good for nontarget lesions [0.82 3 + 5 6 (3.70%) (0.74–0.90)] and moderate for nonpathological lesions 4 + 3 24 (14.81%) [0.71 (0.50–0.92)], which were approximately equal to the 4 + 4 25 (15.43%) agreement between R1 and manual correction [0.89, 0.81 4 + 5 61 (37.65%) and 0.68 for target lesions, nontarget lesions and non- 5 + 4 23 (14.20%) pathological lesions, respectively] but slightly higher than 5 + 5 3 (1.85%) the agreement between R2 and the reference standard [0.86, 0.82 and 0.60 for target lesions, nontarget lesions and nonpathological lesions, respectively]. Quantitative measurement of the lymph node segmentation Discussion The mean short diameters of the automatically segmented MET-RADS-P is a guideline for the treatment response and manually segmented target lesions were 23.53  mm evaluation of systemic metastases of patients with APC, (interquartile range, 17.61- 26.55  mm) and 27.94  mm which involves the evaluation of primary focus, bone Table 3 Segmentation results of pelvic lymph nodes Metrics All Subgroup analysis Target lesions Nontarget lesions Nonpathological Baseline lesions Post- lesions treatment lesions DSC 0.82 ± 0.09 0.85 ± 0.09 0.82 ± 0.09 0.78 ± 0.09 0.81 ± 0.10 0.82 ± 0.09 VS 0.88 ± 0.12 0.88 ± 0.09 0.86 ± 0.08 0.80 ± 0.08 0.87 ± 0.09 0.88 ± 0.08 DSC Dice similarity coefficient, VS volumetric similarity. The DSC and VS values were used to evaluate the performance of the automated lymph node segmentation by comparison with the manual annotation Liu et al. Cancer Imaging (2023) 23:7 Page 6 of 9 Fig. 3 Explementary results of lymph node segmentation and correction. Light green: target lesion; light blue: nontarget lesion; light yellow: nonpathological lesion Fig. 4 Agreement between the automatically segmented and manually segmented lymph nodes. a target lesions; b nontarget lesions Fig. 5 Confusion matrix of treatment response assessment Liu  et al. Cancer Imaging (2023) 23:7 Page 7 of 9 Table 4 The consistency of treatment response assessment Comparison Target lesions Nontarget lesions Nonpathological lesions Automated segmentation vs. Manual correction 0.92 (0.86–0.98) 0.82 (0.74–0.90) 0.71 (0.50–0.92) R1 vs. Automated segmentation 0.89 (0.81–0.96) 0.81 (0.73–0.89) 0.68 (0.47–0.88) R2 vs. Automated segmentation 0.86 (0.79–0.94) 0.78 (0.69–0.87) 0.60 (0.37–0.82) R1 vs. R2 0.90 (0.85–0.97) 0.95 (0.90–0.99) 0.84 (0.68–0.99) R1 vs. Manual correction 0.96 (0.93–1.00) 0.99 (0.97–1.00) 0.96 (0.88–1.00) R2 vs. Manual correction 0.94 (0.89–0.99) 0.96 (0.92–0.99) 0.88 (0.73–1.00) R1 an attending radiologist with 8 years of reading experience, R2 a fellow radiologist with 4 years of reading experience metastases, lymph node metastases and organ metasta- according to MET-RADS-P criteria included two parts. ses. In this study, we established a semiautomatic pelvic First, a previously established pelvic lymph node seg- lymph node treatment response evaluation process for mentation model was used to perform the automatic patients with APC through lymph node segmentation segmentation of lymph nodes. The model achieved based on deep learning. Our results showed that the good segmentation performance here, which is similar accuracies of automated segmentation-based response to the segmentation results reported in previous litera- assessment were high for all the target lesions, nontarget ture (the DSC and VS values for all visible lymph nodes lesions and nonpathological lesions according to MET- were 0.76 ± 0.15 and 0.82 ± 0.14, respectively) [14], espe- RADS-P criteria and achieved good consistency with the cially the target lesions, further highlighting its potential attending radiologist and fellow radiologist. usefulness. Based on the morphology and signal characteristics of Second, based on the quantitative measurements all acquired images, the MET-RADS-P system mapped obtained from the automated segmentation, we can unequivocal diseases to 14 predefined body regions [8, directly evaluate the treatment response according to 15]. Analysis of lymph node metastases in the pelvis is MET-RADS-P criteria, which can be more practical in crucial for clinical practice and drug studies in patients clinical settings. A clinical radiology report provides a with APC, which is the most common metastatic site qualitative narrative, but does not provide standardized, [17]. A lymph node’s size is highly correlated with sur- quantitative information about the patient’s progress or vival time, a measurement that radiologists and clini- response to treatment [21]. Natural language processing cians perform to monitor disease progression or assess and deep learning models have been employed in previous therapeutic options, due to the fact that many malig- studies to estimate responses from clinical text [22, 23]. nancies can enlarge lymph nodes [18]. According to These approaches can be feasible for quantitative assess - the Response Evaluation Criteria in Solid Tumors 1.1 ment related to MET-RADS-P criteria but can be indirect. (RECIST 1.1) Guidelines, lymph nodes with a short-axis Our proposed semiautomated algorithm achieved high diameter of at least 10 mm are considered to be enlarged Kappa values in terms of treatment response assessment lymph nodes and are clinically significant [19]. The size with attending and fellow radiologists when measuring standard of pathological lymph nodes defined by MET- the same set of target and nontarget lesions. The con - RADS-P based on MRI was similar to RECIST 1.1, while sistency of nonpathological lesions was lower, which MET-RADS-P provides a more complete assessment of may be due to the relatively poor segmentation perfor- nodal metastases response including the nontarget nodes mance. Tang et  al. [24] proposed a deep learning-based and nonpathologic nodes, which was usually qualitatively method for semiautomated RECISTS measurement and assessed by RECIST 1.1 criteria. assessed using a mean difference between the deep learn - According to the MET-RADS-P criteria, the core whole ing algorithm and manual measurement in the unit of body MRI protocol designed for bone and lymph node pixels. Scores using pixel difference, however, may not metastasis detection included T1WI (GRE Dixon tech- be reliable, as scores are largely determined by data com- nique) and axial DWI [8]. DWI is a well-recognized and position. In this study, we used Bland–Altman plotting used sequence for pelvic lymph node imaging, that is able based on percent measurement difference to address the to offer qualitative and quantitative assessments for dis - issue as suggested by Woo et  al. [25]. As demonstrated, ease characterizations [14, 20]. Therefore, in this study, the Bland–Altman analysis indicated good consistency we performed the treatment response assessment only between the automated segmentation and manual seg- on DWI images. mentation, and most values were within the upper and In this study, the established semiautomatic pelvic lower LOA. lymph node treatment response evaluation process Liu et al. Cancer Imaging (2023) 23:7 Page 8 of 9 was written by Xiang Liu and Zemin Zhu, and all authors commented on There are some limitations that need to be addressed. previous versions of the manuscript. All authors read and approved the final First, in this study, the deep learning-based treatment manuscript. response assessment was only focused on the pelvic Funding lymph node, and other regions of the body according to This work was supported by the Capital’s Funds for Health Improvement and the MET-RADS-P guideline need to be investigated in Research (2020–2-40710) and Innovation Fund for Outstanding Doctoral Can- the future. Second, we acknowledge that there remain didates of Peking University Health Science Centre (BMU2022BSS001). opportunities for further model refinement, including Availability of data and materials the achievement of lymph node registration between The datasets used and/or analyzed during the current study are available from baseline and posttreatment images, thus realizing fully the corresponding author on reasonable request. automated lymph node treatment response evaluation. Finally, our results demonstrated that the semiautomated Declarations treatment response assessment can be achieved on the Ethics approval and consent to participate DWI sequence, but the values of other sequences (e.g. This study was performed in accordance with the principles of the Declaration T1WI, DCE or T2WI) on response assessment also need of Helsinki and was approved by the Committee for Medical Ethics, Peking University First Hospital (2021–060). Informed consent was waived according to be investigated in further studies. to its retrospective design. Consent for publication Not applicable. Conclusion In conclusion, we have developed a semiautomated deep Competing interests learning-based model to estimate response assessments Yaofeng Zhang, Jialun Li and Xiangpeng Wang are from a medical technical corporation provided technical support for model development. The authors of pelvic lymph nodes in patients with APC. The accu - declare that they have no competing interests. racy of response assessments based on the automati- cally segmented lymph nodes showed close similarity to Author details Department of Radiology, Peking University First Hospital, No.8 Xishiku the manually segmented lymph nodes and yielded out- Street, Xicheng District, Beijing 100034, China. Department of Hepatobiliary put comparable to the radiologists. These initial results and Pancreatic Surgery, Zhuzhou Central Hospital, Zhuzhou 412000, China. provide a promising way to achieve a fully automated School of Basic Medical Sciences, Capital Medical University, Beijing 100069, China. Beijing Smart Tree Medical Technology Co. Ltd, Beijing 100011, China. treatment response assessment algorithm according to MET-RADS-P criteria. Received: 4 October 2022 Accepted: 5 January 2023 Abbreviations ADC Apparent diffusion coefficient APC Advanced prostate cancer References CR Complete response 1. Teo MY, Rathkopf DE, Kantoff P. Treatment of Advanced Prostate Cancer. DCE Dynamic contrast-enhanced Annu Rev Med. 2019;70:479–99. DSC Dice similarity coefficient 2. Komura K, Sweeney CJ, Inamoto T, Ibuki N, Azuma H, Kantoff PW. DWI Diffusion-weighted imaging Current treatment strategies for advanced prostate cancer. Int J Urol. LOA Limits of agreement 2018;25(3):220–31. MET-RADS-P METastasis Reporting and Data System for Prostate Cancer 3. Swami U, McFarland TR, Nussenzveig R, Agarwal N. Advanced pros- PD Progressive disease tate cancer: treatment advances and future directions. Trends Cancer. PR Partial response 2020;6(8):702–15. PCWG P rostate Cancer Clinical Trials Working Group 4. Halabi S, Kelly WK, Ma H, Zhou H, Solomon NC, Fizazi K, et al. Meta- PSA Prostate-specific antigen analysis evaluating the impact of site of metastasis on overall sur- SD Stable disease vival in men with castration-resistant prostate cancer. J Clin Oncol. T2WI T2-weighted imaging 2016;34(14):1652–9. VS Volumetric similarity 5. Scher HI, Morris MJ, Stadler WM, Higano C, Basch E, Fizazi K, et al. Trial design and objectives for castration-resistant prostate cancer: updated Acknowledgements recommendations from the prostate cancer clinical trials working group Not applicable. 3. J Clin Oncol. 2016;34(12):1402–18. 6. Abern MR, Tsivian M, Polascik TJ. Focal therapy of prostate cancer: Code availability evidence-based analysis for modern selection criteria. Curr Urol Rep. The codes used for the development the algorithm are available from the 2012;13(2):160–9. corresponding author on reasonable request. 7. Lecouvet FE, Talbot JN, Messiou C, Bourguet P, Liu Y, de Souza NM. Monitoring the response of bone metastases to treatment with Magnetic Authors’ contributions Resonance Imaging and nuclear medicine techniques: a review and posi- All authors contributed to the study conception and design. Material prepa- tion statement by the European Organisation for Research and Treatment ration, data collection and analysis were performed by Xiang Liu and Kexin of Cancer imaging group. Eur J Cancer. 2014;50(15):2519–31. Wang. Xiang Liu performed manual segmentation under the supervision 8. Padhani AR, Lecouvet FE, Tunariu N, Koh DM, De Keyzer F, Collins DJ, of Xiaoying Wang. Xiaodong Zhang participated in the image interpreta- et al. METastasis reporting and data system for prostate cancer: practical tion. Zemin Zhu, Jialun Li, Yaofeng Zhang, and Xiangpeng Wang performed guidelines for acquisition, interpretation, and reporting of whole-body data interpretation and statistical analysis. The first draft of the manuscript Liu  et al. Cancer Imaging (2023) 23:7 Page 9 of 9 magnetic resonance imaging-based evaluations of multiorgan involve- ment in advanced prostate cancer. Eur Urol. 2017;71(1):81–92. 9. Cook GJR, Goh V. Molecular Imaging of Bone Metastases and Their Response to Therapy. J Nucl Med. 2020;61(6):799–806. 10. Liu X, Han C, Cui Y, Xie T, Zhang X, Wang X. Detection and segmentation of pelvic bones metastases in MRI images for patients with prostate cancer based on deep learning. Front Oncol. 2021;11:773299. 11. Liu X, Wang X, Zhang Y, Sun Z, Zhang X, Wang X. Preoperative prediction of pelvic lymph nodes metastasis in prostate cancer using an ADC-based radiomics model: comparison with clinical nomograms and PI-RADS assessment. Abdom Radiol (NY ). 2022;47(9):3327–37. 12. Aerts HJ. The Potential of Radiomic-Based Phenotyping in Precision Medicine: A Review. JAMA Oncol. 2016;2(12):1636–42. 13. Xu X, Zhang HL, Liu QP, Sun SW, Zhang J, Zhu FP, et al. Radiomic analysis of contrast-enhanced CT predicts microvascular invasion and outcome in hepatocellular carcinoma. J Hepatol. 2019;70(6):1133–44. 14. Liu X, Sun Z, Han C, Cui Y, Huang J, Wang X, et al. Development and validation of the 3D U-Net algorithm for segmentation of pelvic lymph nodes on diffusion-weighted images. BMC Med Imaging. 2021;21(1):170. 15. Padhani AR, Tunariu N. Metastasis reporting and data system for prostate cancer in practice. Magn Reson Imaging Clin N Am. 2018;26(4):527–42. 16. Taha AA, Hanbury A. Metrics for evaluating 3D medical image segmenta- tion: analysis, selection, and tool. BMC Med Imaging. 2015;15:29. 17. Kim YJ, Song C, Eom KY, Kim IA, Kim JS. Lymph node ratio determines the benefit of adjuvant radiotherapy in pathologically 3 or less lymph node-positive prostate cancer after radical prostatectomy: a popu- lation-based analysis with propensity-score matching. Oncotarget. 2017;8(66):110625–34. 18. Alheejawi S, Xu H, Berendt R, Jha N, Mandal M. Novel lymph node segmentation and proliferation index measurement for skin melanoma biopsy images. Comput Med Imaging Graph. 2019;73:19–29. 19. Eisenhauer EA, Therasse P, Bogaerts J, Schwartz LH, Sargent D, Ford R, et al. New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur J Cancer. 2009;45(2):228–47. 20. Eiber M, Beer AJ, Holzapfel K, Tauber R, Ganter C, Weirich G, et al. Pre- liminary results for characterization of pelvic lymph nodes in patients with prostate cancer by diffusion-weighted MR-imaging. Invest Radiol. 2010;45(1):15–23. 21. Arbour KC, Luu AT, Luo J, Rizvi H, Plodkowski AJ, Sakhi M, et al. Deep learning to estimate RECIST in patients with NSCLC treated with PD-1 blockade. Cancer Discov. 2021;11(1):59–67. 22. Chen MC, Ball RL, Yang L, Moradzadeh N, Chapman BE, Larson DB, et al. Deep learning to classify radiology free-text reports. Radiology. 2018;286(3):845–52. 23. Bozkurt S, Alkim E, Banerjee I, Rubin DL. Automated detection of meas- urements and their descriptors in radiology reports using a hybrid natural language processing algorithm. J Digit Imaging. 2019;32(4):544–53. 24. Chlebus G, Schenk A, Moltz JH, van Ginneken B, Hahn HK, Meine H. Automatic liver tumor segmentation in CT with fully convolutional neural networks and object-based postprocessing. Sci Rep. 2018;8(1):15497. 25. Woo M, Devane AM, Lowe SC, Lowther EL, Gimbel RW. Deep learning for semi-automated unidirectional measurement of lung tumor size in CT. Cancer Imaging. 2021;21(1):43. Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in pub- 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: : lished maps and institutional affiliations. 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

Cancer Imaging – Springer Journals

Published: Jan 17, 2023

Keywords: Deep learning; MET-RADS-P criteria; Pelvic lymph nodes; Metastases; DWI

There are no references for this article.