Discrimination of Human Skin Burns Using Machine Learning

Burns become a serious concern issue affecting thousands of lives worldwide and subjecting victims to physical deformities which usually led them to discrimination in the society due to the scary looks. High mortality rates are being reported annually and is associated with lack of healthcare facilities in most of the remote locations such as towns and villages, as well as unavailability or inadequate experienced burn surgeons. Moreover, studies have shown that experienced burn surgeons have drawback in their assessment due to visual fatigue. Therefore, we propose this study to determine whether Machine Learning (ML) can be used to discriminate between burnt skin and normal skin images. We expect to minimize the unwanted hospital delays and render service delivery improvement when conducted with ML algorithms. As such, we employed one of the variant architectures of Residual Network (ResNet) - ResNet101 for the operation. The kernels of this model were used in extracting useful features and Support Vector Machine (SVM) along with 10-fold cross-validation technique was used in classifying the images and obtained a recognition accuracy of 99.5%.