How much data is needed to train a medical image deep learning system to achieve necessary high accuracy?

• DOI: 10.48550/arxiv.1511.06348
• Publication Date: 2015-01-01
• Authors: Junghwan Cho; Kyewook Lee; Ellie Shin; Garry Choy; Synho Do
• Keywords: Data set; Training set; Sample (material); Contextual image classification

The use of Convolutional Neural Networks (CNN) in natural image classification systems has produced very impressive results. Combined with the inherent nature medical images that make them ideal for deep-learning, further application such to holds much promise. However, usefulness and potential impact a system can be completely negated if it does not reach target accuracy. In this paper, we present study on determining optimum size training data set necessary achieve high accuracy low variance systems. CNN was applied classify axial Computed Tomography (CT) into six anatomical classes. We trained using different sizes (5, 10, 20, 50, 100, 200) then tested resulting total 6000 CT images. All were acquired from Massachusetts General Hospital (MGH) Picture Archiving Communication System (PACS). Using data, employ learning curve approach predict at given sample size. Our research will general methodology certain easily other problems within