Evaluation of Neoadjuvant Chemoradiotherapy Response in Rectal Cancer Using MR Images and Deep Learning Neural...

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence in Medicine

Background:

• Rectal cancer treatment response evaluation is crucial for guiding therapy.
• Neoadjuvant chemoradiotherapy (nCRT) is a standard treatment for rectal cancer.
• Accurate assessment of tumor response post-nCRT is essential for patient management.

Purpose of the Study:

• To develop deep-learning neural networks for evaluating tumor response to nCRT in rectal cancer.
• To utilize magnetic resonance (MR) images for accurate prediction of treatment response.
• To guide treatment decisions and improve patient outcomes in rectal cancer.

Main Methods:

• Retrospective analysis of 59 stage 2 or 3 rectal cancer tumors treated with nCRT.
• Development of a deep neural network (DNN) with long short-term memory (LSTM) units using MR image features.
• Comparison of DNN-based tumor regression grading (DNN-TRG) with pathological grading (Dw-TRG) and traditional MR imaging grading (MR-TRG).

Main Results:

• DNN-TRG demonstrated higher accuracy (89.8%) in predicting good or complete response compared to MR-TRG (76.3%).
• DNN-TRG achieved higher sensitivity (84.6%) and accuracy (89.8%) for predicting tumor response than MR-TRG.
• DNN-TRG showed strong agreement with pathological grading (Cohen's kappa=0.79 for good/complete response).

Conclusions:

• Deep LSTM neural networks offer a promising, accurate method for evaluating tumor response to nCRT in rectal cancer.
• AI-driven analysis of MR images can significantly enhance the prediction of treatment response.
• This approach has the potential to optimize treatment strategies and improve outcomes for rectal cancer patients.