Automated reporting system using deep convolutional neural network in the medical domain

Abstract

Nowadays, in the healthcare sector, a massive volume of medical data sources is available. The data is growing at 153 Exabytes in 2013 and an estimated 2,314 exabytes in 2020 (Turner, Gantz et al. 2014). The medical data is composed of the patients' information, medications, follow-up, recommendation, and more other information. One of the medical sector's significant issues is medical experts' need besides their jobs to document and write different medical reports. On the other hand, machine learning plays a vital role in analyzing the large volume of medical data available in the healthcare sector to help diagnose and predict diseases. In the future, deep learning-based methods are considered the most promising machine learning methods in the medical field. To analyze and classify the medical images and present the medical reports based on these images/videos with minimum assistance from the medical staff and with high accuracy and some understandable form. There is a need to apply deep learning methods to produce an automatic documenting and reporting system. This thesis uses a deep convolutional neural network (CNN) to improve the automated reporting and documenting system of the gastrointestinal tract (such as inflammation and colorectal cancer). The report is designed under the principle of universal design, and accessibility procedure is also maintained to generate the report. The thesis also focuses on representing the internal processes of generating medical reports using CNNs.