Comparing Adversarial and Supervised Learning for Organs at Risk Segmentation in CT images

Leonardo Crespi,Mattia Portanti,Daniele Loiacono

DOI: https://doi.org/10.1109/CBMS58004.2023.00280

2023-03-31

Abstract:Organ at Risk (OAR) segmentation from CT scans is a key component of the radiotherapy treatment workflow. In recent years, deep learning techniques have shown remarkable potential in automating this process. In this paper, we investigate the performance of Generative Adversarial Networks (GANs) compared to supervised learning approaches for segmenting OARs from CT images. We propose three GAN-based models with identical generator architectures but different discriminator networks. These models are compared with well-established CNN models, such as SE-ResUnet and DeepLabV3, using the StructSeg dataset, which consists of 50 annotated CT scans containing contours of six OARs. Our work aims to provide insight into the advantages and disadvantages of adversarial training in the context of OAR segmentation. The results are very promising and show that the proposed GAN-based approaches are similar or superior to their CNN-based counterparts, particularly when segmenting more challenging target organs.

Image and Video Processing,Computer Vision and Pattern Recognition,Machine Learning

What problem does this paper attempt to address?

This paper aims to solve the problem of automatically segmenting Organs at Risk (OARs) from CT images during radiotherapy. Specifically, the author compared the performance of Generative Adversarial Networks (GANs) and supervised learning methods in the OARs segmentation task in CT images. The paper proposed three GAN - based models and compared these models with existing mature CNN models (such as SE - ResUnet and DeepLabV3), using the StructSeg dataset containing 50 annotated CT scans. The main objective of the study was to explore the advantages and disadvantages of adversarial training in OARs segmentation, especially when dealing with more challenging target organs. ### Main problems: 1. **Automation of OARs segmentation**: Manual segmentation of OARs is a time - consuming and error - prone process, especially when dealing with a large number or multiple targets. Therefore, automating this process is crucial for improving the efficiency of the clinical workflow. 2. **GANs vs. supervised learning**: Currently, supervised learning is the most commonly used method for training CNNs for medical image segmentation. However, GANs, as an emerging generative model, perform well in image generation tasks, but their performance in semantic segmentation tasks, especially compared with supervised learning methods, still needs further research. ### Research methods: - **Dataset**: Use the StructSeg dataset, which contains 50 annotated CT scans, and each scan contains the contours of six OARs. - **Models**: Propose three GAN - based models. These models share the same generator architecture but use different discriminator networks. At the same time, compare with two mature CNN models, SE - ResUnet and DeepLabV3. - **Evaluation metrics**: Use the Dice Similarity Coefficient (DSC) and Hausdorff Distance (HD) as evaluation metrics. These two metrics are common standards for evaluating model performance. ### Main findings: - **Overall performance**: GAN - based models perform equally well or better than CNN - based models in the segmentation tasks of most organs. - **Performance of specific organs**: When dealing with more challenging organs (such as the trachea and esophagus), GAN - based models show better performance. - **Multi - class segmentation**: In multi - class segmentation tasks, GAN - based models also show similar advantages, especially in the segmentation of the trachea and esophagus. ### Conclusions: - Generative Adversarial Networks (GANs) show performance equal to or better than supervised learning methods in the OARs segmentation task in CT images, especially when dealing with more challenging organs. - This research provides a basis for future research in the field of medical image segmentation, helps to further explore the advantages and limitations of adversarial training and supervised learning, and may ultimately develop more accurate and robust segmentation models, enhancing their potential in clinical applications.

Comparing Adversarial and Supervised Learning for Organs at Risk Segmentation in CT images

Generative Adversarial Network Based Cerebrovascular Segmentation for Time-of-flight Magnetic Resonance Angiography Image

Automatic Multi-Organ Segmentation in Thorax CT Images Using U-Net-GAN

Shape Constrained Fully Convolutional DenseNet with Adversarial Training for Multi-organ Segmentation on Head and Neck Low Field MR Images

Automatic Segmentation of Prostate Magnetic Resonance Imaging Using Generative Adversarial Networks

Semi-supervised semantic segmentation of prostate and organs-at-risk on 3D pelvic CT images

Incorporating Adaptive Sparse Graph Convolutional Neural Networks for Segmentation of Organs at Risk in Radiotherapy

Deep Learning-Augmented Head and Neck Organs at Risk Segmentation From CT Volumes

Comprehensive Evaluation of a Deep Learning Model for Automatic Organs-at-Risk Segmentation on Heterogeneous Computed Tomography Images for Abdominal Radiation Therapy

Adversarial deep learning for improved abdominal organ segmentation in CT scans

Multi-organ segmentation of organ-at-risk (OAR's) of head and neck site using ensemble learning technique

SA-GAN: Structure-Aware GAN for Organ-Preserving Synthetic CT Generation

Auto-segmentation of pelvic organs at risk on 0.35T MRI using 2D and 3D Generative Adversarial Network models

Comprehensive evaluation of a deep learning model for automatic organs at risk segmentation on heterogeneous computed tomography images for abdominal radiotherapy

Synthetic CT-aided Multiorgan Segmentation for CBCT-guided Adaptive Pancreatic Radiotherapy

LGAN: Lung segmentation in CT scans using generative adversarial network

Organ at Risk Segmentation in Head and Neck CT Images Using a Two-Stage Segmentation Framework Based on 3D U-Net

Enhancing Organ at Risk Segmentation with Improved Deep Neural Networks

Deep Learning-based Multi-organ CT Segmentation with Adversarial Data Augmentation

URO-GAN: an Untrustworthy Region Optimization Approach for Adipose Tissue Segmentation Based on Adversarial Learning

Transformer based Generative Adversarial Network for Liver Segmentation