Using Deep Learning to Classify Burnt Body Parts Images for Better Burns Diagnosis

摘要

Several deaths occur each year because of burns. Despite advancements in burn care, proper burns diagnosis and treatment of burn patients still remains a major challenge. Automated methods to give an early assessment of the total body surface area (TBSA) burnt and/or the burns depth can be extremely helpful for better burns diagnosis. Researchers are considering the use of visual images of burn patients to develop these automated burns diagnosis methods. As the skin architecture varies across different parts of the body, and so the burn impact on different body parts. So, it is likely that the body part specific visual images based automatic burns diagnosis assessment methods would be more effective than generic visual images based methods. Considering this, we explore this problem of classifying the body part of burn images. To the best of our knowledge, ours is the first attempt to classify burnt body part images. In this work, we consider 4 different burnt body parts: face, hand, back, and inner arm, and we present the effectiveness of independent and dependent deep learning models (using ResNet-50) in classifying the different burnt body parts images.