MRI-Based Radiomics to Predict Response to Neoadjuvant Therapy in Locally Advanced Rectal Cancer: A Retrospective...

Area of Science:

• Radiology
• Oncology
• Machine Learning

Background:

• Neoadjuvant therapy response in locally advanced rectal cancer (LARC) is variable.
• Early identification of non-responders is crucial for treatment optimization and toxicity reduction.

Purpose of the Study:

• To develop and validate a machine learning model using baseline MRI radiomic features.
• To predict treatment response in LARC patients based on MRI tumor regression grade (mrTRG).

Main Methods:

• Retrospective analysis of 86 LARC patients.
• Radiomic feature extraction from baseline MRI using PyRadiomics.
• LASSO and elastic net logistic regression for feature selection and model building.
• 5-fold cross-validation for performance evaluation.

Main Results:

• The model achieved an AUC-ROC of 0.73.
• Accuracy was 72.5%, sensitivity 79.2%, and specificity 50% in predicting responders vs. non-responders (mrTRG ≤ 2 vs. ≥ 3).

Conclusions:

• Baseline MRI-based radiomics shows promise in identifying LARC patients likely to non-respond to neoadjuvant therapy.
• Limited specificity and lack of external validation hinder immediate clinical application.
• Further multicenter validation and integration with clinical data are needed to enhance model generalizability.