Deep convolutional neural network accurately classifies different types of bladder cancer cells based on their pH fingerprints and morphology

Abstract

Non-invasive identification of different populations of cells at the single-cell level carries significant biomedical implications. We recently developed a novel fast and cost-effective method that, combining pH imaging and machine learning, enabled us to classify normal and cancer cells based on their different intracellular acidity. Here, we sought to capture and utilize intracellular structural features, known to undergo changes during cancer progression, in addition to intracellular pH patterns in order to make robust predictions. Leveraging both the biophysical and biochemical markers acquired via pH imaging with deep learning allowed us to classify cancer cells, at single-cell resolution, with very high accuracy. Specifically, the deep Convolutional Neural Network (CNN)-based strategy classified individual cells from the RT4 and J82 bladder cancer cell lines with an accuracy of 99.9%, compared to 94% achieved with our previously reported pH-based method.