Deep Learning-Based Prediction System for Surgical Difficulty in Rectal Cancer Patients Using MRI Pelvimetry

Area of Science:

• Medical Imaging and Artificial Intelligence
• Surgical Oncology
• Rectal Cancer Treatment

Background:

• Laparoscopic total mesorectal excision (TME) is a complex procedure for rectal cancer.
• Accurate prediction of surgical difficulty is crucial for patient outcomes and surgical planning.

Purpose of the Study:

• To develop an automated system for predicting laparoscopic TME difficulty.
• To utilize magnetic resonance imaging (MRI) pelvimetry and clinical data with deep learning (DL).

Main Methods:

• Colorectal MRI data were used for DL segmentation and automatic pelvimetry measurements.
• Statistical analysis identified predictors of surgical difficulty.
• A logistic nomogram was generated using selected predictors.

Main Results:

• Segmentation models achieved high accuracy (92.7%-95.7% DSC).
• Automatic measurements correlated well with manual measurements.
• Key predictors included age, tumor location, interspinous distance, angle Δ, and mesorectal fat area (C-index: 0.852).

Conclusions:

• The developed system offers standard guidance for assessing TME difficulty.
• Potential for clinical application in objective surgical difficulty prediction.
• Future work aims for full automation with a user-friendly interface.