Automated endoscopic image analysis based on machine learning

Abstract

This presentation introduces automated endoscopic image diagnosis method based on machine learning. Especially we discuss effectiveness of machine learning in automated diagnosis of endoscopic images by showing automated pathological type diagnosis of colonic polyps from endocytoscopy images as examples. Machine learning research staring from perceptron or statistical pattern recognition has started in many years ago. Progress of high performance computing has enabled us to train neural network having very complex network architecture. Machine learning techniques are considered to help physicians to diagnose endoscopic images, where very higher skills are required for diagnosis. This presentation will show several automated diagnosis methods for endoscopic images. The first method classifies pathological types of colonic polyps by using hand-crafted feature values. SVM is utilized for classification. The second example is the method using CNN for automated classification. We will show these method from the technological viewpoint. Also we will discuss about training data generation.