Machine learning pipeline with custom grid search for colorectal Raman spectroscopy data.

Area of Science:

• Biomedical Engineering
• Spectroscopy
• Machine Learning

Background:

• Colorectal cancer (CRC) is a significant global health issue.
• Early detection of CRC is critical for successful treatment outcomes.
• Current diagnostic methods can be invasive or lack real-time feedback.

Purpose of the Study:

• To evaluate the efficacy of handheld Raman spectroscopy coupled with machine learning for classifying colorectal tissue.
• To develop and optimize classification models for distinguishing between healthy and cancerous colorectal tissues.
• To explore advanced model selection strategies for imbalanced biomedical datasets.

Main Methods:

• A dataset of 330 Raman spectra from 155 participants undergoing colonoscopy was collected.
• Standardized preprocessing pipeline applied to spectral data.
• Custom grid search optimized hyperparameters and preprocessing, prioritizing balanced accuracy for imbalanced data.

Main Results:

• Decision Tree (DT) and Support Vector Classifier (SVC) models achieved the highest balanced accuracy.
• DT achieved 71.77% balanced accuracy, and SVC achieved 70.77% balanced accuracy.
• These models outperformed those trained with traditional methods, demonstrating improved classification performance.

Conclusions:

• Handheld Raman spectroscopy shows potential as a rapid, non-destructive screening tool for colorectal cancer.
• Tailored model selection strategies, focusing on balanced accuracy, are crucial for imbalanced biomedical data.
• This approach represents a promising step towards clinical validation for improved CRC detection.