Deep Learning model for diagnosis and report generation of lethal chest diseases using X-rays

Abstract

Viewing medical images, diagnosing and summarizing them is a challenging task. An expert in this field gives a description of the X Ray in the form of a radiology report by distinguishing between the usual and unusual findings and provide an overview for their decision. Research shows that because of the inadequate number of experts, which in turn increases patient volumes, and the nature of human perception, radiology practice sometimes results in error. To lessen the volume of analytical errors and to assuage the job of radiologists, there is a necessity for a computer-assisted diagnosis and create a radiology report when an X Ray is given as an input. In the proposed model, chest X Rays are used for the diagnosis of diseases. Additionally, VGG16 has been used to classify the images resulting in an accuracy of 88%. For summarizing the X-rays, Encoder Decoder model has been used along with the Xception model. To further evaluate the reports, Bilingual evaluation has been used which has given a score of 96 percent for the proposed model.