Prediction of Alopecia Areata using Machine Learning Techniques

Alopecia areata (AA) is a chronic, autoimmune condition that attacks anagen hair follicles, causing sudden hair loss, resulting in circular bald patches. The disorder may be developed in both adults as well as children and the prevalence of the same is approximated to be 1 in every 1000 people, putting around 2% of the broad population at a risk of developing this disease at some point in their lives. Existing techniques to assess the disorder are heavily supported by naked-eye examination and thus have demonstrated nominal accuracy. In recent years, Machine Learning has paved the way for enhanced diagnosis of diseases in various fields of healthcare, including Dermatology. This study postulates a significant role of computer-aided diagnosis of AA to equip medical practitioners with a more accurate form of prediction and classification. Our proposed framework is in relevance to the categorization of healthy hair and hair with symptoms that indicate AA. For the dataset, a 1000 images of healthy hair were collected via web scraping and the Figaro1k dataset. Over 500 AA images were web scraped along with some of the same from the Dermnet dataset. To improve the dataset further, the images were preprocessed by performing image enhancement, segmentation and data augmentation. This study aims to elevate the quality of diagnosis carried out for accurate prediction of Alopecia areata in the field of Dermatology by conducting a comparative study for classification by implementing SVM, KNN, Random Forest Classifier, Gaussian Naive Bayes and CNN algorithms which gave 85%, 78%, 88%, 80% and 92% accuracy respectively.