RefineNet‐based 2D and 3D automatic segmentations for clinical target volume and organs at risks for patients with cervical cancer in postoperative radiotherapy

Abstract

Purpose An accurate and reliable target volume delineation is critical for the safe and successful radiotherapy. The purpose of this study is to develop new 2D and 3D automatic segmentation models based on RefineNet for clinical target volume (CTV) and organs at risk (OARs) for postoperative cervical cancer based on computed tomography (CT) images. Methods A 2D RefineNet and 3D RefineNetPlus3D were adapted and built to automatically segment CTVs and OARs on a total of 44 222 CT slices of 313 patients with stage I–III cervical cancer. Fully convolutional networks (FCNs), U‐Net, context encoder network (CE‐Net), UNet3D, and ResUNet3D were also trained and tested with randomly divided training and validation sets, respectively. The performances of these automatic segmentation models were evaluated by Dice similarity coefficient (DSC), Jaccard similarity coefficient, and average symmetric surface distance when comparing them with manual segmentations with the test data. Results The DSC for RefineNet, FCN, U‐Net, CE‐Net, UNet3D, ResUNet3D, and RefineNet3D were 0.82, 0.80, 0.82, 0.81, 0.80, 0.81, and 0.82 with a mean contouring time of 3.2, 3.4, 8.2, 3.9, 9.8, 11.4, and 6.4 s, respectively. The generated RefineNetPlus3D demonstrated a good performance in the automatic segmentation of bladder, small intestine, rectum, right and left femoral heads with a DSC of 0.97, 0.95, 091, 0.98, and 0.98, respectively, with a mean computation time of 6.6 s. Conclusions The newly adapted RefineNet and developed RefineNetPlus3D were promising automatic segmentation models with accurate and clinically acceptable CTV and OARs for cervical cancer patients in postoperative radiotherapy.

Review of Deep Learning Based Autosegmentation for Clinical Target Volume: Current Status and Future Directions

“…DLAS models performed well in contouring cervical cancer CTV, with a DSC range of 0.68 to 0.89. 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 , 38 , 39 , 40 , 41 Only Chang et al showed one pretrained DLAS model that had DSC of 0.68. 31 Compared with manual contouring, DLAS models used demonstrated satisfactory, if not better, performance than manual contouring with improvements in DSC and HD.…”

Feasibility of clinical target volume (CTV) automatic delineation using deep learning network for cervical cancer radiotherapy: a study with external validation

“…4, whether it is cervical cancer or endometrial cancer, AttResCNet achieved the better automatic delineation performance among the evaluated models in the most of the slices. Additionally, in our study, we added oncologist clinical evaluation, which were not included in numerous studies [30][31][32][33].…”

“…Kim et al [10] evaluated the feasibility of automatic delineation of CTV for patients with endometrial and cervical cancer using atlas-based auto-segmentation (ABAS) algorithms, which are used in commercial software, making the research not accessible for many radiotherapy centers. DL-based networks were open-source, which were developed in many researches [11][12][13]. In addition, the cervical cancer automatic delineation DL models studies did not have external validation [10][11][12][13] and only few researches had clinical evaluation [9].…”

A prior‐information‐based automatic segmentation method for the clinical target volume in adaptive radiotherapy of cervical cancer

“…Recently, deep‐learning‐based (DL) methods using Convolutional neural networks (CNN) have been applied to the segmentation of images of a variety of cancers 6–8 . In terms of cervical cancer, it has shown comparable performance to the expert delineation for some cervical cancer segmentation 9–14 . However, the performance of these auto‐segmentation networks is highly dependent on the training dataset, which is affected by the difference in physician contouring styles and the anatomic structure differences among patients 15,16 .…”

“… 6 , 7 , 8 In terms of cervical cancer, it has shown comparable performance to the expert delineation for some cervical cancer segmentation. 9 , 10 , 11 , 12 , 13 , 14 However, the performance of these auto‐segmentation networks is highly dependent on the training dataset, which is affected by the difference in physician contouring styles and the anatomic structure differences among patients. 15 , 16 In order to generalize the trained network to a variety of unseen patient cases, network training is usually based on a large enough training set of different patient cases.…”