Evaluation of machine learning techniques for Barret’s and dysplasia discrimination of the esophagus from in vivo optical coherence tomography images

Optical Coherence Tomography (EOCT) systems can acquire high-resolution in vivo, real-time images of the human esophagus and so, they can be an important tool for the early diagnosis and prognosis of serious esophageal diseases such as Barrett’s, dysplasia and esophageal cancer. In this study, we compare various machine learning methods for tissue segmentation and classification of esophageal tissue using in vivo OCT images. An automated algorithm is proposed able to detect normal tissue from Barrett’s Esophagus (BE) and dysplasia. The classification was based on several features extracted from each A-Scan of EOCT images. The areas of the epithelium were annotated as normal, BE or dysplasia by an expert.The results of various machine learning (ML) classifiers have shown that a neural network based approach provided the best performance, separating BE from dysplasia, for individual A-Scans, with an accuracy of 89%.