Artificial Intelligence Models Using Magnetic Resonance Imaging to Predict Response to Chemoradiotherapy in Rectal...

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Pathological complete response (pCR) after neoadjuvant chemoradiotherapy (nCRT) is crucial for managing locally advanced rectal cancer (LARC).
• Magnetic resonance imaging (MRI) is vital for assessing response, but distinguishing residual tumor from treatment effects is difficult.
• AI and ML models applied to MRI show potential for predicting pCR, yet methodological variability impedes clinical adoption.

Purpose of the Study:

• To systematically review the performance of MRI-based AI and ML models in predicting pCR in LARC patients.
• To identify current limitations and areas for improvement in AI/ML model development and validation for this application.

Main Methods:

• A comprehensive literature search was performed across major databases (Embase, Medline, Cochrane, Web of Science) following PRISMA guidelines.
• Studies utilizing MRI-only AI or ML models for pCR prediction in rectal cancer post-chemoradiotherapy were included.
• Data extraction and risk of bias assessment (QUADAS-2) were conducted independently by two reviewers.

Main Results:

• Twenty-two studies with 94 predictive models were included, predominantly retrospective, using T2-weighted MRI.
• Significant variability was observed in MRI protocols, modeling techniques, and validation strategies; only five studies performed external validation.
• The median area under the curve (AUC) was 0.801, with model performance varying widely (AUC 0.49–0.997).

Conclusions:

• MRI-based AI models exhibit moderate discriminative ability for predicting pCR in LARC following neoadjuvant therapy.
• Methodological heterogeneity, inconsistent reporting, and insufficient external validation currently limit the generalizability of these models.
• Standardization of methods and multicenter external validation are essential for clinical implementation.