Rectal cancer response to neoadjuvant chemoradiotherapy evaluated with MRI: Development and validation of a...

Area of Science:

• Radiology and Medical Imaging
• Oncology
• Data Science in Medicine

Background:

• Locally advanced rectal cancer (LARC) requires effective treatment response assessment.
• Neoadjuvant chemoradiotherapy (CRT) is a standard treatment for LARC.
• Accurate prediction of treatment response is crucial for patient management.

Purpose of the Study:

• To develop and validate a data mining decision support model.
• To discriminate between complete responders (CR) and non-complete responders (NCR) after neoadjuvant CRT in LARC patients.
• To utilize morphologic features from MRI images for classification.

Main Methods:

• Retrospective analysis of two patient groups (n=65 for training, n=30 for validation).
• Development of a decision tree algorithm (J48) using MRI-derived features: MR-TRG, SV, TVRR, and SIRR.
• Evaluation of model performance using sensitivity, specificity, accuracy, and AUC.

Main Results:

• The J48 decision tree model achieved high accuracy in distinguishing CR and NCR patients (95.7%).
• The model demonstrated excellent performance with 85.71% sensitivity and 100% specificity in the validation group.
• Good inter- and intra-reader agreement (κ > 0.6) was observed for MRI feature analysis.

Conclusions:

• The developed decision support model shows potential for aiding in the differentiation of CR and NCR patients with LARC post-CRT.
• This model could assist clinicians in predicting treatment outcomes.
• Further validation in larger cohorts may enhance its clinical utility.