Gastric Lesion Classification Using Deep Learning Based on Fast and Robust Fuzzy C-Means and Simple Linear Iterative Clustering Superpixel Algorithms

Abstract

Gastric diseases are a common medical issue; they can be detected using endoscopy equipment. Computer-aided diagnosis (CADx) systems can help internists identify gastric diseases more accurately. In this paper, we present a CADx system that can detect and classify gastric diseases such as gastric polyps, gastric ulcers, gastritis, and cancer. The system uses a deep learning model as a GoogLeNet based on an Inception module. The fast and robust fuzzy C-means (FRFCM) and simple linear iterative clustering (SLIC) superpixel algorithms are applied for image segmentation during preprocessing. The FRFCM algorithm, which is based on morphological reconstruction and membership filtering, is much faster and more robust than fuzzy C-means. In addition, the SLIC superpixel algorithm adapts the k-means clustering method to efficiently generate superpixels. These two approaches produce a feasible method of classifying normal and abnormal gastric lesions. The areas under the receiver operating characteristic curves were 0.85 and 0.87 for normal and abnormal lesions, respectively. The proposed CADx system also performs reliably.