Related Experiment Video
Updated: Apr 1, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.7K
Deep Learning for Automated Tumor Segmentation of Rectal Cancer on T2-Weighted Magnetic Resonance Images
Miri Seo1, YongDae Lee2, Myung-Won You3
1Department of Medicine, Kyung Hee University College of Medicine, Seoul, Korea.
Yonsei Medical Journal
|March 31, 2026
Summary
A deep learning model accurately segments rectal cancer (RC) on MRI scans. Segmentation accuracy improved with overlaid rectum data and was influenced by tumor volume.
Area of Science:
- Radiology
- Medical Imaging
- Artificial Intelligence
Background:
- Rectal cancer (RC) diagnosis relies on accurate tumor segmentation from MRI.
- Manual segmentation is time-consuming and prone to inter-observer variability.
- Automated segmentation using deep learning (DL) offers a promising solution.
Purpose of the Study:
- To develop and evaluate a DL-based automated segmentation model for rectal cancer on T2-weighted (T2W) MRI.
- To compare segmentation performance with and without rectum guidance.
Main Methods:
- Retrospective analysis of 458 patients' baseline rectal MRIs.
- Expert radiologists manually segmented tumors and rectum on T2W axial images.
- Attention U-Net model trained for voxel-wise tumor classification.
- Evaluation using Dice Similarity Coefficient (DSC) with different rectum guidance strategies.
Main Results:
- The DL model achieved a tumor segmentation DSC of 73.35% without rectum guidance.
- Segmentation accuracy improved to 75.52% with overlaid rectum data.
- Tumor volume positively correlated with segmentation accuracy across all models.
Conclusions:
- A DL-based automated segmentation model can achieve over 70% accuracy for rectal cancer on T2W MRI.
- Incorporating overlaid rectum data enhances segmentation performance.
- Larger tumor volumes are associated with improved segmentation accuracy.
