Access the full text.
Sign up today, get DeepDyve free for 14 days.
B. Al‐Qaisieh (2003)Pre- and post-implant dosimetry: an inter-comparison between UK prostate brachytherapy centres.
Radiotherapy and oncology : journal of the European Society for Therapeutic Radiology and Oncology, 66 2
Daguang Zhang, Zhiyong Yang, Shan Jiang, Lin Zhou, Zeyang Zhou, Wei Wang (2019)Individualized and inverse optimized needle configuration for combined intracavitary-interstitial brachytherapy in locally advanced cervical cancer
Journal of Cancer Research and Therapeutics, 15
B. Fionda, L. Boldrini, A. D’Aviero, V. Lancellotta, M. Gambacorta, G. Kovács, S. Patarnello, V. Valentini, L. Tagliaferri (2020)Artificial intelligence (AI) and interventional radiotherapy (brachytherapy): state of art and future perspectives
Journal of Contemporary Brachytherapy, 12
Elizabeth Watt, M. Skarsgard, M. Roumeliotis, S. Husain, T. Meyer (2019)Post-implant analysis in permanent breast seed implant: automated plan reconstruction using simulated annealing
Journal of Contemporary Brachytherapy, 11
R Mazeron (2015)257
Radiother Oncol, 114
Junjie Hu, Ying Song, Qiang Wang, S. Bai, Zhang Yi (2020)Incorporating historical sub-optimal deep neural networks for dose prediction in radiotherapy
Medical image analysis, 67
C. Haie-meder, R. Pötter, E. Limbergen, E. Briot, M. Brabandere, J. Dimopoulos, I. Dumas, T. Hellebust, C. Kirisits, S. Lang, S. Muschitz, J. Nevinson, A. Nulens, P. Petrow, N. Wachter-Gerstner (2005)Recommendations from Gynaecological (GYN) GEC-ESTRO Working Group (I): concepts and terms in 3D image based 3D treatment planning in cervix cancer brachytherapy with emphasis on MRI assessment of GTV and CTV.
Radiotherapy and oncology : journal of the European Society for Therapeutic Radiology and Oncology, 74 3
MP Schmid (2011)468
Radiother Oncol, 100
H. Pang, Xiaoyang Sun, Bo Yang, JingBo Wu (2018)Predicting the dose absorbed by organs at risk during intensity modulated radiation therapy for nasopharyngeal carcinoma.
The British journal of radiology, 91 1092
L. Appenzoller, J. Michalski, W. Thorstad, S. Mutic, K. Moore (2012)Predicting dose-volume histograms for organs-at-risk in IMRT planning.
Medical physics, 39 12
D. Reijtenbagh, J. Godart, A. Leeuw, Y. Seppenwoolde, I. Jürgenliemk-Schulz, J. Mens, R. Nout, M. Hoogeman (2022)Multi-center analysis of machine-learning predicted dose parameters in brachytherapy for cervical cancer.
Radiotherapy and oncology : journal of the European Society for Therapeutic Radiology and Oncology
R Potter (2011)116
Radiother Oncol, 100
S. Shiraishi, K. Moore (2015)Knowledge-based prediction of three-dimensional dose distributions for external beam radiotherapy.
Medical physics, 43 1
M. Schmid, C. Kirisits, N. Nesvacil, J. Dimopoulos, D. Berger, R. Pötter (2011)Local recurrences in cervical cancer patients in the setting of image-guided brachytherapy: A comparison of spatial dose distribution within a matched-pair analysis
Radiotherapy and Oncology, 100
Katherina Cortes, K. Kallis, A. Simon, J. Mayadev, S. Meyers, K. Moore (2021)Knowledge-based three-dimensional dose prediction for tandem-and-ovoid brachytherapy.
S. Shiraishi, Jun Tan, L. Olsen, K. Moore (2015)Knowledge-based prediction of plan quality metrics in intracranial stereotactic radiosurgery.
Medical physics, 42 2
R. Pötter, P. Georg, J. Dimopoulos, M. Grimm, D. Berger, N. Nesvacil, D. Georg, M. Schmid, A. Reinthaller, A. Sturdza, C. Kirisits (2011)Clinical outcome of protocol based image (MRI) guided adaptive brachytherapy combined with 3D conformal radiotherapy with or without chemotherapy in patients with locally advanced cervical cancer
Radiotherapy and Oncology, 100
B Wu (2017)57
Radiother Oncol, 123
Binbin Wu, M. Kusters, M. Kunze-Busch, T. Dijkema, T. McNutt, G. Sanguineti, K. Bzdusek, A. Dritschilo, D. Pang (2017)Cross-institutional knowledge-based planning (KBP) implementation and its performance comparison to Auto-Planning Engine (APE).
Radiotherapy and oncology : journal of the European Society for Therapeutic Radiology and Oncology, 123 1
BS Gill (2015)540
Int J Radiat Oncol Biol Phys, 91
José Rubio (2020)Stability Analysis of the Modified Levenberg–Marquardt Algorithm for the Artificial Neural Network Training
IEEE Transactions on Neural Networks and Learning Systems, 32
Y. Ge, Q. Wu (2019)Knowledge‐based planning for intensity‐modulated radiation therapy: A review of data‐driven approaches
Medical Physics, 46
R. Mazeron, P. Castelnau-Marchand, I. Dumas, Eleonor Campo, L. Kom, F. Martinetti, G. Farha, Anne Tailleur, P. Morice, Cyrus Chargari, D. Lefkopoulos, C. Haie-meder (2015)Impact of treatment time and dose escalation on local control in locally advanced cervical cancer treated by chemoradiation and image-guided pulsed-dose rate adaptive brachytherapy.
Radiotherapy and oncology : journal of the European Society for Therapeutic Radiology and Oncology, 114 2
Amy Frederick, M. Roumeliotis, P. Grendarova, S. Quirk (2021)Performance of a knowledge‐based planning model for optimizing intensity‐modulated radiotherapy plans for partial breast irradiation
Journal of Applied Clinical Medical Physics, 23
Chenyang Shen, Y. Gonzalez, P. Klages, N. Qin, Hyunuk Jung, Liyuan Chen, D. Nguyen, Steve Jiang, X. Jia (2018)Intelligent inverse treatment planning via deep reinforcement learning, a proof-of-principle study in high dose-rate brachytherapy for cervical cancer
Physics in Medicine & Biology, 64
Dashan Jiang, Hui Yan, Na Chang, Teng Li, R. Mao, Chi Du, Bin Guo, Jianfei Liu (2020)Convolutional Neural Network Based Dosimetry Evaluation for Esophageal Radiation Treatment Planning.
Tahir Yusufaly, K. Kallis, A. Simon, J. Mayadev, C. Yashar, J. Einck, L. Mell, Derek Brown, D. Scanderbeg, S. Hild, Brent Covele, K. Moore, S. Meyers (2020)A knowledge-based organ dose prediction tool for brachytherapy treatment planning of patients with cervical cancer.
F. Bray, J. Ferlay, I. Soerjomataram, R. Siegel, Lindsey Torre, A. Jemal (2018)Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries
CA: A Cancer Journal for Clinicians, 68
A. Viswanathan, S. Beriwal, J. Santos, D. Demanes, D. Gaffney, J. Hansen, E. Jones, C. Kirisits, B. Thomadsen, Beth Erickson (2012)American Brachytherapy Society consensus guidelines for locally advanced carcinoma of the cervix. Part II: high-dose-rate brachytherapy.
Brachytherapy, 11 1
F Bray (2018)394
CA Cancer J Clin, 68
A. Damato, A. Viswanathan, R. Cormack (2013)Validation of mathematical models for the prediction of organs-at-risk dosimetric metrics in high-dose-rate gynecologic interstitial brachytherapy.
Medical physics, 40 10
Yuliang Jiang, Z. Ji, F. Guo, R. Peng, Haitao Sun, J. Fan, Shuhua Wei, Weiyan Li, Kai Liu, Jinghua Lei, Junjie Wang (2018)Side effects of CT-guided implantation of 125I seeds for recurrent malignant tumors of the head and neck assisted by 3D printing non co-planar template
Radiation Oncology (London, England), 13
B. Gill, Hayeon Kim, C. Houser, J. Kelley, P. Sukumvanich, R. Edwards, J. Comerci, A. Olawaiye, Marilyn Huang, M. Courtney-Brooks, S. Beriwal (2015)MRI-guided high-dose-rate intracavitary brachytherapy for treatment of cervical cancer: the University of Pittsburgh experience.
International journal of radiation oncology, biology, physics, 91 3
C. Kirisits, S. Lang, J. Dimopoulos, D. Berger, D. Georg, R. Pötter (2006)The Vienna applicator for combined intracavitary and interstitial brachytherapy of cervical cancer: design, application, treatment planning, and dosimetric results.
International journal of radiation oncology, biology, physics, 65 2
C Kirisits (2006)624
Int J Radiat Oncol Biol Phys, 65
N. Kriegeskorte, Tal Golan (2019)Neural network models and deep learning
Current Biology, 29
Objective A neural network method was employed to establish a dose prediction model for organs at risk (OAR) in patients with cervical cancer receiving brachytherapy using needle insertion. Methods A total of 218 CT‑based needle ‑insertion brachytherapy fraction plans for loco ‑regionally advanced cervical cancer treatment were analyzed in 59 patients. The sub‑ organ of OAR was automatically generated by self‑ written MATLAB, and the volume of the sub‑ organ was read. Correlations between D2cm of each OAR and volume of each sub‑ organ—as well as high‑risk clinical target volume for bladder, rectum, and sigmoid colon—were analyzed. We then established a neural network predictive model of D2cm of OAR using the matrix laboratory neural net. Of these plans, 70% were selected as the training set, 15% as the validation set, and 15% as the test set. The regression R value and mean squared error were subsequently used to evaluate the predictive model. Results The D2cm /D90 of each OAR was related to volume of each respective sub‑ organ. The R values for blad‑ der, rectum, and sigmoid colon in the training set for the predictive model were 0.80513, 0.93421, and 0.95978, respectively. The ∆D2cm /D90 for bladder, rectum, and sigmoid colon in all sets was 0.052 ± 0.044, 0.040 ± 0.032, and 0.041 ± 0.037, respectively. The MSE for bladder, rectum, and sigmoid colon in the training set for the predictive model −3 −3 −3 was 4.779 × 10 , 1.967 × 10 and 1.574 × 10 , respectively. Conclusion The neural network method based on a dose‑prediction model of OAR in brachytherapy using needle insertion was simple and reliable. In addition, it only addressed volumes of sub‑ organs to predict the dose of OAR, which we believe is worthy of further promotion and application. Keywords Brachytherapy, Needle insertion, Neural network, Dose prediction Introduction The incidence and mortality rates of cervical cancer are high among women worldwide , and concurrent radio- therapy and chemotherapy can significantly improve the local control rate in these patients . Brachytherapy is Huai‑ wen Zhang and Xiao‑ming Zhong contributed equally to this work. a key technique used in the radical radiotherapy of cer- vical cancer, and possesses the distinct advantages of *Correspondence: Haow ‑ en Pang physical dosimetry, enabling the tumor to receive rela- email@example.com tively high doses without causing serious complications 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. Zhang et al. BMC Cancer (2023) 23:385 Page 2 of 7 to surrounding normal tissues . However, locally Amsterdam, The Netherlands) and gynecological exami - advanced tumors are relatively large and frequently nations were performed to identify tumor areas before invade nearby cervical tissues. Conventional brachy- external beam intensity-modulated radiation therapy therapy does not adequately enclose the target volume, (IMRT) and brachytherapy. All patients were adminis- often resulting in uncontrolled or recurrent tumors. In tered diluted iohexol (10 mL in 1000 mL of water) as a contrast, brachytherapy using needle insertion improves gastrointestinal contrast medium before brachytherapy, target coverage , thereby enhancing local control and and patients also received an enema and underwent overall patient survival [5, 6]. bladder catheterization prior to treatment. Experienced In designing a brachytherapy plan, the choice of con- physicians determined the location of the uterus and straint parameters is of utmost importance and directly tumor characteristics based upon the results of the MRI influences the quality of the final plan. However, this and gynecological examinations, and the site of needle information is unknown prior to designing the clini- insertion. cal brachytherapy plan. The plan designer usually refers to the optimization definition goal provided by the doc - Target volume delineation and treatment planning tor, which is based on data from the general population All patients underwent external beam IMRT at a pre- data and radiation therapy group. The Radiation Therapy scribed dose of 45 Gy/25 F, which was increased to Oncology Group guidelines or statistically derived clini- 60 Gy/25 F for patients with biopsy-positive lymph cal norms serve as the target for OAR dose optimization nodes. After external beam IMRT, Ir high-dose-rate . In general, these reference objectives are universal. brachytherapy was administered at a dose of 6 or 7 Gy/ However, an optimal brachytherapy plan should be based fraction, 1–2 fractions/week, for a total of 4 or 5 frac- upon each patient’s unique anatomic structures, and, tions. An appropriate intrauterine tube was inserted into therefore, methods that apply universal clinical norms the uterine cavity based on the depth and angle of the cannot meet the needs of the individual patient. uterine cavity. Next, based on the tumor location, metal In recent years, machine learning methods have been insertion needles were inserted into the tumor. Brachy- widely applied to external beam radiotherapy, and have therapy was performed on a Ir source (mHDR, Elekta, shown promising results [8–11]. For the present study, The Netherlands) using a Microelectron v2 afterloader we implemented the authorized Chinese invention patent (Elekta, The Netherlands) (the needle manufacturer was method (patent number: ZL201610529290.8), which can Elekta [Elekta, Veenendaal, the Netherlands], part num- predict the OAR dose distribution prior to the design of ber 083.920; diameter of 1.5 mm, and length of 140 mm). the brachytherapy plan. This can help the plan designer All CT images were transferred to Onecentra 4.3 in evaluating the quality of the brachytherapy plan, deter- treatment planning software (Elekta Brachytherapy, mine whether the brachytherapy plan meets the require- Veenendaal, The Netherlands) to formulate the brachy - ments, provide standards for dosimetry verification and therapy plan. Target volumes—including high-risk clini- quality control, meet the specific needs of the individual cal target volume (HR-CTV), intermediate-risk clinical patient, and provide a basis for the automation of tumor target volume (IR-CTV), and OARs (bladder, rectum, radiotherapy planning. To our knowledge, this work is and sigmoid colon)—were depicted on CT images. The the first to apply brachytherapy using needle insertion. HR-CTV included the entire cervix and residual tumor This method can ensure dose distribution with high pre - during brachytherapy. In contrast, IR-CTV included all cision and thus improve the efficiency of brachytherapy. components of the HR-CTV and areas with tumor infil - tration before external beam IMRT . The dose curves Materials and methods were optimized repeatedly using manual/graphic optimi- Patients zation with Oncentra 4.3 treatment-planning software A total of 218 CT-based needle-insertion brachytherapy (Elekta Brachytherapy, Veenendaal, The Netherlands) fraction plans were analyzed in 59 patients with loco- to ensure that the prescribed doses for the curves were regionally advanced cervical cancer in the Department of lower around the HR-CTV and OARs. Tumor target Oncology of the Affiliated Hospital of Southwest Medical and OAR doses were calculated based on the equivalent University. Of these plans, 70% were selected as the train- dose delivered in 2 Gy fractions (EQD2). The doses were ing set, 15% as the validation set, and 15% as the test set as follows: for HR-CTV D90, EQD2 ≥ 85 Gy; for blad- applying the matrix laboratory (MATLAB) neural net-der D2cm , EQD2 ≤ 90 Gy; and for rectum and sigmoid fitting application. D2cm , EQD2 ≤ 75 Gy . D90 referred to the dose All patients were scanned with contrast-enhanced at 90% of the HR-CTV, and D2c m referred to the dose magnetic resonance images (MRI, Achieva 3.0 T, Philips, received by 2 cm of the OAR. Zhang et al. BMC Cancer (2023) 23:385 Page 3 of 7 Deriving the sub‑organs from the OAR application (R2017a, MathWorks, Inc., Natick, MA, The HR-CTV was externally expanded to a plurality of USA). We selected the Levenberg–Marquardt (LM) rings (ring –ring ) with a width of 0.3 cm. R ing –ring algorithm to train 100 iterations through the MATLAB 1 n 1 n and different OAR (bladder, rectum, and sigmoid colon) neural net-fitting application, and we chose the best intersection regions (r ing –ring ∩ OAR) were used result to establish the neural network prediction model 1 n as independent sub-organs, with r ing ∩ OAR defined of D2cm /D90 (output) with V and the HR- n nsub-organ as sub-organ-n (e.g., ring ∩ OAR was defined as sub- CTVs of the bladder, rectum, and sigmoid colon as the organ-1). The total number of sub-organs was kept input variables. The network consisted of three layers: under 15 (the intersecting regions for sub-organ-1 to (1) the input layer (transmission rate), which receives sub-organ-5 of the rectum are presented in Fig. 1). The input data to the network through a set of neurons; sub-organ volume was normalized for improved data (2) the hidden layer, which runs a set of algorithms to analyses. The normalized sub-organ volume (V ) compute the input data; and (3) the output layer linear nsub-organ was equal to the sub-organ volume divided by the OAR transformation, which iteratively calculates the desired volume. In this study, the MATLAB application written result through a set of linear output neurons. The out - by us was used to automatically generate the sub-organ put depends on the weighted sum of the input variables and automatically read as V . One of the patients plus bias to ensure numerical stability . The LM algo - nsub-organ automatically generated a CT tomogram of the sub-organ rithm applied in the network was a back-propagation for the rectum through a self-written MATLAB app pro- algorithm, which is a combination of two minimization gram (Fig. 1). algorithms and a gradient descent algorithm—i.e., the Gauss–Newton algorithm . Dose prediction model based on the neural network method Predictive accuracy evaluation The D2cm /D90 of the OAR was used as the prediction The neural network prediction model was evaluated target to eliminate the influence of different D90 of the using the regression R-value (Regression R values meas- HR-CTV. Correlations between D2cm /D90 for each ure the correlation between outputs and targets. An OAR and their V —as well as the HR-CTVs for R value of 1 indicates a close relationship, and 0 indi- nsub-organ the bladder, rectum, and sigmoid colon—were analyzed. cates a random relationship). In addition, the mean The neural network prediction model was estab - squared error (MSE) was the average squared differ - lished based on a correlation, and the predictive model ence between outputs and targets (lower values are bet- was established using the MATLAB neural net-fitting ter, and zero indicates no error). Model performance Fig. 1 Sub‑ organ of the rectum The blue line indicates HR‑ CTV; the shadow area indicates the sub‑ organ Zhang et al. BMC Cancer (2023) 23:385 Page 4 of 7 3 3 respectively; in the testing set for the predictive model was also quantified as follows: ∆D2cm /D90 =|D2cm / were 0.80144, 0.88250, and 0.85583, respectively; and in D90(actual)-D2cm /D90(predicted)| (mean and standard all sets for the predictive model were 0.80077, 0.92075, deviation). and 0.92880, respectively. We observed no statistical dif- ference between the D2c m /D90 predicted value for each Statistical methods OAR and the actual planned value in all sets (P > 0.05). We conducted Pearson’s correlation test to analyze the The p-value in our paired t test results for D2cm /D90 correlation between D2cm /D90 of each OAR with HR- for the bladder, rectum, and sigmoid colon were 0.630, CTV volume, each OAR volume, and their V nsub-organ 0.185, and 0.638, respectively (P > 0.05). The MSEs for the using Statistical Product Service Solutions (SPSS) 19.0. bladder, rectum, and sigmoid colon in the training set for Statistical differences between predicted and planned −3 −3 the predictive model were 4.779 × 10 , 1.967 × 10 , and values of D2cm /D90 of each OAR were compared −3 1.574 × 10 , respectively. The regression diagram of the using paired t tests. P < 0.05 was considered statistically prediction model is shown in Fig. 2; the MSE for the pre- significant. dictive model is presented in Table 3; and the predicted and actual plots for all set are depicted in Fig. 3. Results The average number of needles inserted per patient was Discussion 3.7, with a maximum of eight and a minimum of two nee- With the continuous development of artificial intelli - dles. The correlation analysis results of D2cm /D90 for gence (AI) in radiation oncology (RO) and interventional the bladder, rectum, and sigmoid colon with HR-CTV radiotherapy (IRT, brachytherapy), AI and automation volume, bladder volume, rectal volume, sigmoid colon in RO and IRT are able to successfully facilitate all the volume, and their respective sub-organ volumes are steps of treatment workflow. Compared to traditional shown in Tables 1. and 2. As our results showed a signifi - approaches, AI exhibits potential benefits in reducing cant correlation between D2c m /D90 and related param- time-consuming repetitive tasks and improving treat- eters, we then used the neural network-based method to ment-plan quality assurance . Implementing AI in establish the predictive model. IRT might also result in significant advantages for physi - The R values for the bladder, rectum, and sigmoid colon cians, allowing them more time to interact with patients. in the training set for the predictive model were 0.80513, Several recent studies have underlined the concept that 0.93421, and 0.95978, respectively. The mean values for AI can automatically adjust the weight parameters of the ∆D2cm /D90 for the bladder, rectum, and sigmoid treatment plans, assist in optimizing applicator loca- colon in all sets were 0.052 ± 0.044, 0.040 ± 0.032, and tion in treatment planning phases, and predict the opti- 0.041 ± 0.037, respectively. The R values for the blad - mal source position in targets—thus avoiding irradiation der, rectum, and sigmoid colon in the validation set for of OARs in brachytherapy [17–19]. Through machine the predictive model were 0.85809, 0.92256, and 0.90246, Table 1 Correlationcoefficient between D2cm /D90 of each OAR and the volumes of HR‑ CTV,bladder, rectum and sigmoid colon D2cm /D90 Variable Volume of HR‑ CTV Volume of bladder Volume of rectum Volume of sigmoid colon Volume of small intestine 3 a a D2cm D90 (bladder) 0.539 0.349 \ \ \ 3 a a a D2cm /D90 (rectum) 0.454 0.195 0.154 \ \ 3 a a a D2cm /D90 (sigmoid colon) 0.424 0.189 \ 0.391 \ Note: significant correlation at 0.01 level (bilateral) Table 2 Correlation coefficient between D2cm/D90 and V of each OAR nsub‑ organ D2cm /D90 V nsub‑ organ V V V V V nsub‑ organ1 nsub‑ organ2 nsub‑ organ3 nsub‑ organ4 nsub‑ organ5 3 a a a a a D2cm /D90 (bladder) 0.391 0.485 0.476 0.419 0.334 3 a a a a a D2cm /D90 (rectum) 0.220 0.601 0.743 0.773 0.753 3 a a a a a D2cm /D90 (sigmoid colon) 0.286 0.402 0.487 0.506 0.536 Significant correlation at 0.01 level (bilateral) Zhang et al. BMC Cancer (2023) 23:385 Page 5 of 7 Fig. 2 Regression diagram of Dn0–Dn50 for the predictive model: A bladder, B rectum, and C sigmoid colon with the parameters they used for prediction being the Table 3 MSE value of neural network prediction model for overlapping volume of the organ at risk with the targeted D2cm /D90 area or a 1-cm expansion of the target area. In contrast to OAR Set the work of Damato et al., we possessed a larger patient plan with additional parameter input for prediction. Training Validation Testing all Yusufaly et al.  applied an approach closely related –3 –3 –3 –3 bladder 4.779 × 10 3.499 × 10 5.543 × 10 4.702 × 10 to that developed for IMRT to predict bladder, rectum, –3 –3 –3 –3 rectum 1.967 × 10 2.965 × 10 5.028 × 10 2.576 × 10 and sigmoid D2cm for tandem and ovoid treatments. –3 –3 –3 –3 sigmoid colon 1.574 × 10 3.934 × 10 8.815 × 10 3.027 × 10 Reijtenbagh et al.  used 3D U-NET CNN to perform voxel-based dose prediction on OARs, and Cortes et al.  used overlap volume histograms (OVHs) to evalu- ate dose prediction on OARs. In contrast to their work, learning analysis of pre- and post-plan seed configura - our model only included inputs for the volume of sub- tions, effective algorithms have been developed to obtain organs of the OAR and did not require voxel information sufficient target coverage and optimal OAR avoidance in to establish a predictive model; thus, even planners with- brachytherapy [20, 21]. OAR dose prediction has devel- out programming experience can use the model based on oped into an exciting area in the application of AI in radi- open-source software (MATLAB, etc.). otherapy, and dose prediction has been widely applied The current method was successfully applied in pre - to IMRT to reduce time-consuming repetitive tasks and vious studies to the dose prediction of IMRT for naso- assuring the quality of IMRT plans [22–24]. For OAR pharyngeal carcinoma and achieved favorable results dose prediction in cervical cancer brachytherapy, Damato . In contrast to the work of machine learning meth- et al.  deployed a dataset of 20 patients to develop a ods [8–11], our model was easy to establish for radiother- simple mathematical model and predict bladder and rec- apy planners, and it does not require complex modeling. tal D2cm for gynecological interstitial brachytherapy, Fig. 3 Predicted value and the actual planned value for all sets. Red indicates the predicted value of the neural network model; blue indicates the actual planned value: A bladder, B rectum, and C sigmoid colon Zhang et al. BMC Cancer (2023) 23:385 Page 6 of 7 With this study, we were the first to uncover a correla - brachytherapy. Open-source software was used in this tion between D2cm /D90 for each OAR in brachyther- study, and the required arguments were read directly apy using needle insertion and their sub-organs, and to from the planning system or from a written MATLAB establish a predictive model based on the neural network application. For developing countries (particularly for method. Any institution can enter the sub-organ volume newly established radiotherapy departments), the qual- for each OAR to predict OAR doses. Furthermore, we ity control tool for the brachytherapy plan can be estab- did not note any significant differences in model perfor - lished without using any other new software modules mance among training, validation, and test sets, indicat- for radiotherapy planning. ing that the model was not biased towards patients in the We adopted a new and simpler dose prediction training set and did not overfit. method to predict the critical OAR D2cm quality indi- We also analyzed ∆D2cm /D90 to evaluate model cator for patients with cervical cancer receiving brachy- performance, where the absolute value was used for therapy. To our knowledge, this study was the first to evaluation; thus, this indicator accurately reflected show a correlation between D2cm /D90 for each OAR model performance. ∆D2cm /D90 for the bladder, rec- and its sub-organ in cervical cancer brachytherapy tum, and sigmoid colon in all sets were 0.052 ± 0.044, using needle insertion. Based on these conditions, we 0.040 ± 0.032, and 0.041 ± 0.037, respectively; with no postulate that the OAR dose prediction model that we statistical difference detected between the D2cm /D90 have established will greatly improve the quality con- predicted value for each OAR and the actual planned trol and automation of patient treatment plans. This value. The MSE of the bladder, rectum, and sigmoid method has also obtained a Chinese invention patent. colon in the training set for the predictive model were Acknowledgements −3 −3 −3 4.779 × 10 , 1.967 × 10 , and 1.574 × 10 , respectively; Not applicable. and the R value for the predictive model was greater than Presentation 0.8, while in the rectum and sigmoid colon, the R value We were not making a Presentation at any Conference. was greater than 0.9. Based on these results, we hypoth- esize that the predictive model is valid and stable. Disclaimers All study participants provided informed consent, and the study design was If the radiotherapy planning system is used to divide approved by the appropriate ethics review board. We have read and under‑ each OAR into sub-organs and record their volumes, stood your journal’s policies, and we believe that neither the manuscript nor then the amount of data is enlarged and the time cost the study violates any of these. There are no conflicts of interest to declare. remains high, and this is not conducive to the promotion Authors’ contributions of the current research method. Therefore, in this study H.P. conceived of the presented idea. H.P and H.Z. collected the planning we independently wrote our own MATLAB application, data of all patients in this study. H.P and H.Z. took the lead in writing the manuscript. All authors provided critical feed‑back and helped shape the with the compilation automatically generating the sub- research, analysis, and manuscript. The author(s) read and approved the final organ for each OAR and automatically reading V . nsub-organ manuscript. The entire process did not require manual participation, Funding greatly improving overall efficiency. We acknowledge funding from the Gulin County People’s Hospital, Southwest Our research is presently limited to one institution; Medical University Affiliated Hospital Science and Technology Strategic thus, our proposed method may reflect certain limita - Cooperation Project (project number: 2022GLXNNYDFY05), the Sichuan Medical Association Scientific Research Project (project number: S21004), the tions. We expect that additional studies will further Luzhou Municipal People’s Government‑Southwest Medical University Sci‑ increase multicenter research in this area. As the number ence and Technology Strategic Cooperation Fund (number: 2020LZXNYDJ12) of patient plans increases, more plan data can be merged and the Open Fund for Scientific Research of Jiangxi Cancer Hospital (number:2021J15). to ensure that we obtain a more accurate predictive model. Availability of data and materials Despite the limitations to the current model, its pre- All data generated and analyzed during this study are included in this pub‑ lished article. dictive accuracy indicates that the model can still be used as a quality control tool for brachytherapy plans. Declarations In the actual application of brachytherapy plans, the threshold value of the difference between the predicted Ethics approval and consent to participate value and planned value can be set with respect to the The study was conducted according to the ethical guide‑lines of the Helsinki Declaration and was approved by the institutional review board of The Affili‑ quality control of the plan. If the difference between the ated Hospital of Southwest Medical University(Ethics number: KY2021002). two exceeds the threshold, then the plan should be fur- Written informed consents were obtained from all patients prior to treatment. ther optimized so as to reduce the influence of subjec - Consent for publication tive factors for planners and to carry out quality control Consent for publication is not applicable in this study, because there is not and quality assurance for the individualized plan of any individual person’s data. Zhang et al. BMC Cancer (2023) 23:385 Page 7 of 7 Competing interests 16. Fionda B, Boldrini L, D’Aviero A, et al. Artificial intelligence (AI) and inter ‑ There are no conflicts of interest to declare. ventional radiotherapy (brachytherapy): state of art and future perspec‑ tives. J Contemp Brachytherapy. 2020;12(5):497–500. Author details 17. Shen C, Gonzalez Y, Klages P, et al. Intelligent inverse treatment planning Department of Radiotherapy, Jiangxi Cancer Hospital, The Second Affiliated via deep reinforcement learning, a proof‑ of‑principle study in high dose ‑ Hospital of Nanchang Medical College, Jiangxi Clinical Research Center rate brachytherapy for cervical cancer. Phys Med Biol. 2019;64(11):115013. for Cancer, Nanchang 330029, China. Department of Oncology, The third 18. Jiang Y, Ji Z, Guo F, et al. Side effects of CT ‑ guided implantation of I seeds people’s hospital of Jingdezhen, Jingdezhen 333000, China. Depar t‑ for recurrent malignant tumors of the head and neck assisted by 3D ment of Oncology, The Affiliated Hospital of Southwest Medical University, printing non co‑planar template. Radiat Oncol. 2018;13(1):18. Luzhou 646000, China. 19. Zhang D, Yang ZY, Jiang S, et al. Individualized and inverse optimized nee‑ dle configuration for combined intracavitary‑interstitial brachytherapy in Received: 31 January 2023 Accepted: 21 April 2023 locally advanced cervical cancer. J Cancer Res Ther. 2019;15(7):1589–96. 20. Al‑ Qaisieh B, UK Prostrat Brachytherapy Group. Pre‑ and post ‑implant dosimetry: An inter‑ comparison between UK prostate brachytherapy centres. Radiother Oncol. 2003;66(2):181–3. 21. Watt E, Skarsgard M, Roumeliotis M, et al. Post‑implant analysis in permanent breast seed implant: Automated plan reconstruction using References simulated annealing. J Contemp Brachytherapy. 2019;11(1):61–8. 1. Bray F, Ferlay J, Soerjomataram I, et al. Global cancer statistics 2018: GLO‑ 22. Appenzoller LM, Michalski JM, Thorstad WL, et al. Predicting dose‑ BOCAN estimates of incidence and mortality worldwide for 36 cancers in volume histograms for organs‑at ‑risk in IMRT planning. Med Phys. 185 countries. CA Cancer J Clin. 2018;68(6):394–424. 2012;39(12):7446‑61. 2. Schmid MP, Kirisits C, Nesvacil N, et al. Local recurrences in cervical cancer 23. Shiraishi S, Tan J, Olsen LA, et al. Knowledge‑based prediction of plan patients in the setting of image‑ guided brachytherapy: a comparison of quality metrics in intracranial stereotactic radiosurgery. Med Phys. spatial dose distribution within a matched‑pair analysis. Radiother Oncol. 2015;42:908. 2011;100(3):468–72. 24. Shiraishi S, Moore KL. Knowledge‑based prediction of three dimen‑ 3. Potter R, Georg P, Dimopoulos JC, et al. Clinical outcome of protocol sional dose distributions for external beam radiotherapy. Med Phys. based image (MRI) guided adaptive brachytherapy combined with 3D 2016;43:378. conformal radiotherapy with or without chemotherapy in patients with 25. Damato AL, Viswanathan AN, Cormack RA. Validation of mathemati‑ locally advanced cervical cancer. Radiother Oncol. 2011;100(1):116–23. cal models for the prediction of organs‑at ‑risk dosimetric metrics 4. Kirisits C, Lang S, Dimopoulos J, et al. The Vienna applicator for combined in high‑ dose‑rate gynecologic interstitial brachytherapy. Med Phys. intracavitary and interstitial brachytherapy of cervical cancer: design, 2013;40(10):101711. application, treatment planning, and dosimetric results. Int J Radiat Oncol 26. Yusufaly TI, Kallis K, Simon A, et al. A knowledge‑based organ dose Biol Phys. 2006;65(2):624–30. prediction tool for brachytherapy treatment planning of patients with 5. Mazeron R, Castelnau‑Marchand P, Dumas I, et al. Impact of treatment cervical cancer. Brachytherapy. 2020;19(5):624–34. time and dose escalation on local control in locally advanced cervical 27. Reijtenbagh D, Godart J, Leeuw AD, et al. Multi‑ center analysis of cancer treated by chemoradiation and image‑ guided pulsed‑ dose rate machine‑learning predicted dose parameters in brachytherapy for cervi‑ adaptive brachytherapy. Radiother Oncol. 2015;114(2):257–63. cal cancer. Radiother Oncol. 2022;170:169–75. 6. Gill BS, Kim H, Houser CJ, et al. MRI‑ guided high‑ dose‑rate intracavitary 28. Cortes KG, Kallis K, Simon A, Mayadev J, Meyers SM, Moore KL. Knowl‑ brachytherapy for treatment of cervical cancer: the University of Pitts‑ edge‑based three ‑ dimensional dose prediction for tandem‑and‑ ovoid burgh experience. Int J Radiat Oncol Biol Phys. 2015;91(3):540–7. brachytherapy. Brachytherapy. 2022;21(4):532–42. 7. Wu B, Kusters M, Kunze‑Busch M, Dijkema T, McNutt T, Sanguineti G, et al. 29. Pang H, Sun X, Yang B, et al. Predicting the dose absorbed by organs at Cross‑Institutional Knowledge ‑Based Planning (KBP) Implementation and risk during intensity modulated radiation therapy for nasopharyngeal its Performance Comparison to Auto‑Planning Engine (APE). Radiother carcinoma. Br J Radiol. 2018;91(1092):20170289. Oncol. 2017;123(1):57–62. 8. Ge Y, Wu QJ. Knowledge‑based planning for intensity‑modulated radiation therapy: a review of data‑ driven approaches. Med Phys. Publisher’s Note 2019;46(6):2760‑75. Springer Nature remains neutral with regard to jurisdictional claims in pub‑ 9. Jiang D, Yan H, Chang N, et al. Convolutional Neural Network Based lished maps and institutional affiliations. Dosimetry Evaluation for Esophageal Radiation Treatment Planning. Med Phys. 2020;47(10):4735–42. 10. Jh A, Ying SB, Qiang WB, et al. Incorporating historical sub‑ optimal deep neural networks for dose prediction in radiotherapy. Med Image Anal. 2021;67:101886. 11. Frederick A, Roumeliotis M, Grendarova P, et al. Performance of a knowledge‑based planning model for optimizing intensity‑modulated radiotherapy plans for partial breast irradiation. J Appl Clin Med Phys. 2022;23(3):e13506. 12. Haie‑Meder C, Pötter R, Van Limbergen E, et al. Recommendations From 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: : Gynaecological (GYN) GECESTRO Working Group (I): Concepts and Terms in 3D Image Based 3D Treatment Planning in Cervix Cancer Brachyther‑ fast, convenient online submission apy With Emphasis on MRI Assessment of GTV and CTV. Radiother Oncol. thorough peer review by experienced researchers in your ﬁeld 2005;74(3):235–45. rapid publication on acceptance 13. Viswanathan AN, Beriwal S, De Los Santos J, et al. The American brachy‑ • therapy society treatment recommendations for locally advanced carci‑ support for research data, including large and complex data types noma of the cervix Part II: high dose‑rate brachytherapy. Brachytherapy. • gold Open Access which fosters wider collaboration and increased citations 2012;11(1):47‑52. maximum visibility for your research: over 100M website views per year 14. Kriegeskorte N, Golan T. Neural network models and deep learning. Curr Biol. 2019;29(7):R231–6. At BMC, research is always in progress. 15. Rubio J. Stability Analysis of the Modified Levenberg‑Marquardt Algo ‑ rithm for the Artificial Neural Network Training. IEEE Trans Neural Netw Learn more biomedcentral.com/submissions Learn Syst. 2021;32(8):3510–24.
BMC Cancer – Springer Journals
Published: Apr 28, 2023
Keywords: Brachytherapy; Needle insertion; Neural network; Dose prediction
Access the full text.
Sign up today, get DeepDyve free for 14 days.