A mobile AI-enabled platform for screening human skin diseases.

Abstract

For this project, a novel approach for automatic segmentation and classification of Acne and Eczema skin lesions using a machine learning model hosted on a mobile application is proposed. Initially, skin images are filtered to remove unwanted hairs, scars, and noise, and then the segmentation process is carried out to extract lesion areas. A region growing method is applied for segmentation by automatically initializing seed points. The segmentation performance is measured with well-known measures such as accuracy, loss, and precision, and the results are appreciable. Subsequently, the extracted lesion areas are represented by color and texture features. A segmented CNN classifier is used along with its fusion layers for the classification using the extracted features. The system's performance is tested on a dataset obtained online that contains 2779 images consisting of 2 classes of diseases, i.e., Acne and Eczema. The results are encouraging, with an accuracy on the training and validation data of 90.3% and 88% and an AUC of 0.93.