Classification of Hyperspectral Colon Cancer Images Using Convolutional Neural Networks

This paper introduces a classification system for hyperspectral images of colon cancer tissue samples. Focusing on the region of the spectrum between 360 nm and 550 nm, this system utilizes the entire spectral data to reliably differentiate between cancerous and non-cancerous cells. Using a dataset with thirteen patients, convolutional neural networks are designed to compare the classification performance of the hyperspectral images to panchromatic grayscale images of the samples and grayscale images of the individual band samples. Overall, the hyperspectral data is shown to be advantageous in classifying the cancerous and non-cancerous images, ultimately classifying the test sample images with 74.1% accuracy with an F1 score of 0.747, and classifying 85.7% of the cancerous images correctly.