Classification of human stomach cancer using morphological feature analysis from optical coherence tomography images

Optical coherence tomography is radiation-free, and it is considered a tool of optical biopsy. Classification of normal and cancerous tissues is very important for the guidance of surgeons. Here, we develop the morphological feature analysis-based classification (MFAC) method, combining it with machine learning to identify cancerous tissues. We extract five quantitative morphological features from one OCT image through the structured analysis. Five classifiers are involved to make a classification: the support vector machine, the K-nearest neighbor, the random forest, logic regression, and the conventional threshold method. Sensitivity, specificity, and accuracy are used to evaluate these classifiers and are compared with each other. We launched the experimental research of the imaging of ex vivo patients' stomach cancerous tissue with the OCT system. The results showed the three additional features specially designed for stomach cancer are remarkably better than the traditional image feature. The best feature demonstrated over 95% accuracy under all five classifiers. The designed feature based on the layer structure of the stomach tissue is significantly effective. This MFAC method will be used to image the in vivo tissue in clinical applications in the future.