Image Synthesis-based Late Stage Cancer Augmentation and Semi-Supervised Segmentation for MRI Rectal Cancer Staging

Saeko Sasuga,Akira Kudo,Yoshiro Kitamura,Satoshi Iizuka,Edgar Simo-Serra,Atsushi Hamabe,Masayuki Ishii,Ichiro Takemasa

2023-12-08

Abstract:Rectal cancer is one of the most common diseases and a major cause of mortality. For deciding rectal cancer treatment plans, T-staging is important. However, evaluating the index from preoperative MRI images requires high radiologists' skill and experience. Therefore, the aim of this study is to segment the mesorectum, rectum, and rectal cancer region so that the system can predict T-stage from segmentation results. Generally, shortage of large and diverse dataset and high quality annotation are known to be the bottlenecks in computer aided diagnostics development. Regarding rectal cancer, advanced cancer images are very rare, and per-pixel annotation requires high radiologists' skill and time. Therefore, it is not feasible to collect comprehensive disease patterns in a training dataset. To tackle this, we propose two kinds of approaches of image synthesis-based late stage cancer augmentation and semi-supervised learning which is designed for T-stage prediction. In the image synthesis data augmentation approach, we generated advanced cancer images from labels. The real cancer labels were deformed to resemble advanced cancer labels by artificial cancer progress simulation. Next, we introduce a T-staging loss which enables us to train segmentation models from per-image T-stage labels. The loss works to keep inclusion/invasion relationships between rectum and cancer region consistent to the ground truth T-stage. The verification tests show that the proposed method obtains the best sensitivity (0.76) and specificity (0.80) in distinguishing between over T3 stage and underT2. In the ablation studies, our semi-supervised learning approach with the T-staging loss improved specificity by 0.13. Adding the image synthesis-based data augmentation improved the DICE score of invasion cancer area by 0.08 from baseline.

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

What problem does this paper attempt to address?

The paper aims to address the issue of rectal cancer T-staging assessment. Specifically, the goal of the paper is to develop a computer-aided diagnosis (CAD) system that can segment rectal cancer from preoperative magnetic resonance imaging (MRI) images and predict its T-stage. Since evaluating T-stage requires radiologists to have high skills and experience, this study proposes two methods to address the problems of insufficient datasets and annotation difficulties:

1. **Semi-supervised Learning**: A new T-stage loss function is proposed, enabling the model to train the segmentation model with only image-level T-stage labels, rather than relying solely on pixel-level labels.

2. **Image Synthesis-based Advanced Cancer Data Augmentation**: By simulating the cancer progression process to generate advanced cancer images, this method aims to overcome the scarcity of advanced cancer cases, thereby improving the model's performance in handling rare cases.

Through these two methods, the researchers hope to improve the accuracy and specificity of rectal cancer T-staging, ultimately helping doctors diagnose rectal cancer stages more accurately. Experimental results show that the proposed methods achieve good performance in terms of sensitivity and specificity in distinguishing T3 and above stages from T2 and below stages.