Image-based skin diseases diagnosis using deep learning

Abstract

Skin diseases are the fourth most common cause of human illness that results in enormous non fatal burden in daily life activities. There are more than 3000 known skin diseases that are caused by chemical, physical and biological factors. Visual assessment in combination with clinical information is the common diagnostic procedure for skin diseases. However, these procedures are manual and require experience and excellent visual perception. Even though there are sophisticated imaging devices in the market which provide better diagnosis results, their cost limits affordability in a low-resource setting. Different Artificial intelligence-based computer-aided diagnosis systems have been proposed in the literature to diagnose skin disease from clinical images. However, most of them were based on the availability of online datasets rather than the prevalence of diseases, mainly concentrated on the diagnosis of skin tumors and cancers, and accuracy needs to be further improved. In this study, an automated system is proposed for the diagnosis of five common skin diseases using data from clinical images and patient information based on the pre-trained mobilenet-v2 model. Clinical images were acquired using different smartphone cameras and patient information was collected during patient registration. Shades of gray color constancy algorithm were applied to remove the color cast resulting from different illumination sources. The training images were rotated by 90°, flipped horizontally and vertically to increase the size of the training set. Multi class classification accuracy of 87.9% has been achieved using a multiclass classifier using the clinical images only. Integrating patient information with clinical images improve the multiclass classification by 9.6%, resulting in an overall accuracy of 97.5%. A smartphone android application has been developed for ease of use for the proposed skin disease diagnosis system. The proposed system can be used as a decision support system in low resource setting where both expert dermatologists and the means are limited. Even though the proposed work achieved the best performance, further improvement is required by expanding the size of the dataset, including other common skin diseases.