A Longitudinal MRI-Based Artificial Intelligence System to Predict Pathological Complete Response After Neoadjuvant...

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Accurate prediction of treatment response is crucial for managing locally advanced rectal cancer.
• Neoadjuvant chemoradiotherapy (nCRT) is a standard treatment, but predicting response remains challenging.

Purpose of the Study:

• To develop and validate a deep learning model (DeepRP-RC) for predicting pathological complete response (pCR) after nCRT.
• To assess the model's performance using paired MRI scans before and after nCRT.

Main Methods:

• A multitask deep learning model, DeepRP-RC, was trained on longitudinal MRI data from 638 patients.
• The model captured changes in MRI scans to predict pCR and allowed simultaneous segmentation.
• Performance was validated in an internal and three external multicenter datasets involving 1201 patients.

Main Results:

• DeepRP-RC demonstrated high accuracy in predicting pCR, with AUC values ranging from 0.919 to 0.969 across validation sets.
• The model performed consistently across various subgroups and outperformed experienced radiologists.
• DeepRP-RC was also associated with improved disease-free survival, independent of other clinical factors.

Conclusions:

• DeepRP-RC serves as a highly accurate preoperative tool for predicting pCR in rectal cancer patients undergoing nCRT.
• The model's ability to predict response and its prognostic value can guide clinical decision-making.
• Further validation in diverse populations is warranted due to the study's retrospective nature and lack of multiethnic data.