Recent Methods on Medical Image Inpainting and Multi-task Learning Using Deep Learning Models

Medical images are essential in Computer Aided Diagnosis (CAD) systems. Based on the patient’s medical visual data, an automatic CAD system is required for disease categorization, confirmation of severity level, and prevention of future threats. However, medical images that are distorted can considerably impede medical diagnosis. As a result, enhancing diagnostic imaging accuracy and rebuilding damaged areas are critical for the categorization of medical images. These challenges have recently been widely researched for medical image inpainting utilizing Deep Learning (DL) approaches. The medical image inpainting for automatic and accurate medical disease classification becomes a multi-task learning problem. It requires first automatically restoring the original medical image and then its classification. We present a systematic study of the recent solutions for medical image classification considering image inpainting and multi-task learning. The reviewed methods mainly targeted the medical image inpainting using the DL methods. We categorized these methods into Enhanced Generative Adversarial Networks (EGAN), Generative Adversarial Networks (GAN), and other DL methods. The purpose of this survey study is to classify the recent multi-task learning DL-based medical image inpainting methods for efficient medical image classification problems. Then, we discussed the current research gaps according to the functionality and outcomes of the reviewed techniques in this paper.