Image-based deep learning model for predicting pathological response in rectal cancer using post-chemoradiotherapy...

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Rectal cancer treatment response prediction is crucial for personalized therapy.
• Chemoradiotherapy is a standard neoadjuvant treatment for locally advanced rectal cancer.
• Accurate prediction of pathological response aids in treatment planning and outcome assessment.

Purpose of the Study:

• To develop a deep learning model for predicting pathological complete response (pCR) and good response (GR) in rectal cancer.
• To utilize post-chemoradiotherapy T2-weighted axial MR images for prediction.
• To compare the model's performance against human expert observers.

Main Methods:

• A cohort of 466 rectal cancer patients treated with neoadjuvant chemoradiotherapy was analyzed.
• Two deep learning models were trained on T2-weighted MR images to predict pCR (TRG 4) and GR (TRG 3 or 4).
• Model performance was evaluated on a holdout test set and compared with senior radiologist and radiation oncologist assessments.

Main Results:

• The deep learning model achieved an AUC of 0.76 for pCR and 0.72 for GR.
• It demonstrated superior predictive performance compared to human observers, with an accuracy of 85.0% for pCR and 71.7% for GR.
• Fair agreement was observed between the model's predictions and ground truth (kappa = 0.34 for pCR, 0.36 for GR).

Conclusions:

• An AI model based on post-treatment MR imaging shows promise for predicting pathological response in rectal cancer.
• The deep learning approach offers a valuable tool for assessing treatment effectiveness, potentially aiding clinical decision-making.
• The model's performance is comparable or superior to that of experienced clinicians.