Rapid Colorectal Tissue Classification Using Data-Driven Raman Techniques

Abstract

Colorectal cancer is among the most widespread cancers globally, and the risk of developing this disease increases with age. This has led to the recommendation that screening should begin in middleaged patients. Consequently, the implementation of prevention programs has resulted in a greater number of samples being available for histological analysis. Raman spectroscopy offers a diagnostic solution to this challenge. In this study, we present a rapid procedure for Raman tissue analysis and classification using machine learning methods, including both supervised and hybrid supervised-unsupervised approaches, to distinguish between healthy and pathological cases based on Raman spectroscopy data. Raman spectra were acquired using a handheld portable spectrometer with an automatic BubbleFill preprocessing algorithm, covering at spectral range of 500 to 1800 cm-1. Various machine-learning algorithms have been evaluated for classification, including Long Short-Term Memory (LSTM), Multi-Layer Perceptron (MLP), and Extreme Gradient Boosting (XGBoost), and hybrid supervised-unsupervised methods involving dimensionality reduction, clustering, and classification. The experimental results indicated promising classification accuracy, achieving up to 82% accuracy using the MLP algorithm. This suggests the potential effectiveness of neural network methodologies for the classification of Raman spectroscopy data for the diagnosis of colorectal cancer.