Preoperative MRI-based deep learning reconstruction and classification model for assessing rectal cancer.
Yuan Yuan1, Shengnan Ren2, Haidi Lu1
1Department of Radiology, Changhai Hospital, Naval Medical University, 168 Changhai Road, Shanghai, 200433, China.
BMC Medical Imaging
|July 2, 2025
Summary
Deep learning reconstruction significantly improved rectal MRI image quality and lesion visualization. This enhancement improved the accuracy of TN staging for rectal cancer, aiding in better diagnosis and treatment planning.
Area of Science:
- Radiology
- Medical Imaging
- Artificial Intelligence
Background:
- Rectal MRI quality impacts cancer staging.
- Deep learning reconstruction (DLR) is a novel technique for image enhancement.
- Evaluating DLR's effect on rectal MRI and TN staging is crucial.
Purpose of the Study:
- To assess if DLR improves rectal MRI image quality.
- To compare TN staging discrimination using DLR vs. conventional MRI.
- To evaluate deep learning models for TN staging.
Main Methods:
- Retrospective analysis of rectal cancer MRI (T2WI, DWI, CE-T1WI) with and without DLR.
- Image quality assessment (SNR, CNR, visual scoring) by five readers.
- TN staging evaluation and comparison with deep learning models.
Main Results:
- DLR significantly increased SNR and CNR across all sequences (p < 0.0001).
- Overall image quality and lesion display were significantly improved with DLR (p < 0.0001).
- Deep learning models with DLR showed strong TN stage discrimination (AUC 0.937 and 0.824).
Conclusions:
- DLR enhances rectal MRI image quality and lesion visualization.
- DLR-based deep learning models improve TN staging accuracy for rectal cancer.
- DLR offers a promising approach for improved rectal cancer diagnosis.
