A Multiphase CT-Based Integrated Deep Learning Framework for Rectal Cancer Detection, Segmentation, and Staging:...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Oncology
• Radiology

Background:

• Accurate rectal cancer staging is vital for treatment planning but challenging with computed tomography (CT) due to radiologist dependency.
• Current CT interpretation for rectal cancer staging requires significant expertise, highlighting a need for improved diagnostic tools.

Purpose of the Study:

• To develop and evaluate an AI-assisted system for enhancing rectal cancer detection and staging using CT images.
• To integrate lesion detection, segmentation, and staging into a unified AI framework for rectal cancer management.

Main Methods:

• A three-component AI framework was developed: RCD-CNN for lesion detection, U-Net for segmentation, and RCS-3DCNN for staging.
• The system was trained and validated on CT scans from 223 rectal cancer patients, analyzing both non-contrast and contrast-enhanced studies.

Main Results:

• The AI system demonstrated high performance in detection (accuracy 0.976) and segmentation (Dice scores 0.897 and 0.856).
• AI-based staging (80.4% concordance) showed no significant difference compared to radiologist-based staging (82.6% concordance) against pathology.
• The AI system achieved staging accuracy comparable to that of expert radiologists.

Conclusions:

• The developed AI-assisted system is a feasible decision-support tool for rectal cancer management.
• This novel AI framework offers a unified workflow for CT-based rectal cancer detection, segmentation, and staging.
• The AI system shows potential to assist radiologists in improving the accuracy and efficiency of rectal cancer staging.