Extended Analysis of Raman Spectra Using Artificial Intelligence Techniques for Colorectal Abnormality Classification

Area of Science:

• Biomedical Optics
• Medical Spectroscopy
• Computational Biology

Background:

• Raman spectroscopy (RS) offers real-time biochemical analysis for medical applications.
• Integrating artificial intelligence (AI) enhances RS accuracy for in vivo spectral data classification.
• AI-RS integration presents new avenues for precise medical diagnostics.

Purpose of the Study:

• To investigate an optimal preprocessing pipeline for AI-driven statistical analysis of Raman spectral data.
• To propose and evaluate preprocessing methods and algorithms for improved classification outcomes.
• To compare machine learning (ML) and deep learning (DL) algorithms for clinical applicability of RS.

Main Methods:

• Collected spectral data from healthy and cancerous colorectal specimens (n=22).
• Applied various preprocessing techniques: baseline correction, L2 normalization, filtering, and Principal Component Analysis (PCA).
• Compared ML (XGBoost, Random Forest) and DL (1D-Resnet, 1D-CNN) algorithms for tissue classification.

Main Results:

• Preprocessing techniques improved overall accuracy by 15.8%.
• ML models (XGBoost, Random Forest) effectively classified normal and abnormal tissues.
• DL models, especially 1D-CNN, showed superior performance in classifying abnormal cases.

Conclusions:

• AI-enhanced Raman spectroscopy provides accurate malignancy classification.
• Optimized preprocessing pipelines are crucial for advancing RS in clinical diagnostics.
• The study highlights the potential of AI-RS for precise medical analysis and diagnosis.