Machine learning based local recurrence prediction in colorectal cancer using polarized light imaging

Area of Science:

• Oncology
• Biophysics
• Medical Imaging

Background:

• Stage III colorectal cancer (CRC) treatment often requires adjuvant therapy post-surgery.
• Identifying patients at high risk for recurrence is crucial for personalized treatment but lacks quantitative tools.
• Peri-tumoral collagen's role in CRC recurrence is an area of interest.

Purpose of the Study:

• To quantitatively analyze the prognostic value of peri-tumoral collagen in CRC using Mueller matrix (MM) polarized light microscopy.
• To correlate MM-derived collagen features with 5-year local recurrence (LR) in stage III CRC patients.
• To develop a machine learning (ML) model for predicting LR based on polarimetric biomarkers.

Main Methods:

• Utilized a simple MM microscope to image surgical resection samples from stage III CRC patients.
• Derived potential LR biomarkers from MM elements through decomposition and transformation.
• Employed supervised ML models (e.g., XGBoost) with polarimetric biomarkers as features to differentiate between recurrent and non-recurrent cases.

Main Results:

• The best-performing XGBoost model achieved 86% accuracy at the patient level in predicting 5-year local recurrence.
• A sub-cohort analysis within stage III CRC patients yielded an accuracy of 96%.
• Top five polarimetric biomarkers, identified by feature importance, were crucial for model performance.

Conclusions:

• ML-aided polarimetric analysis of collagenous stroma shows prognostic value in CRC.
• This approach may enhance clinical management and personalized therapy selection for CRC patients.
• Quantitative assessment of peri-tumoral collagen could improve prediction of recurrence risk.