Investigating Different Deep learning Models for Classification of Folliculitis

Folliculitis is a common skin condition that happens when a hair follicle(s) become inflamed. The cause of inflammation can be bacterial or fungal infection, ingrown hair due to hair removal etc. Based on the cause of inflammation in follicle, folliculitis can be categorized into eight different types. If left untreated, it may spread, cause deep infections which can further cause permanent hair loss, scarring, cellulitis and even pass into the bloodstream and become fatal. Dermatologists can usually diagnose folliculitis only by glancing at a patient’s skin. However, in order to find the cause of folliculitis, dermatologists may recommend taking a tissue sample, skin swab and having laboratory tests done. Additionally, a sample of the extracted hair from the troubled areas may be obtained for testing. Using potassium hydroxide, hair samples are microscopically examined to identify a potentially infectious cause. In this paper, we investigate how accurate CNNs are in identifying the type of folliculitis. We will use different Convolutional Neural Network (CNN) models like AlexNet, DenseNet201, GoogLeNet, InceptionV3, ResNet50, VGG19 and Xception. The results show that GoogLeNet performs best in identifying the type of folliculitis.