Herpes Zoster Identification Using Optimized Deep Neural Network

Millions of people suffer from the effects of Herpes Zoster (shingles) each year and the reduction in patient outcomes is dependent on early diagnosis and treatment of shingles. Diagnosis of shingles is difficult due to resemblance to other skin disorders. The proposed CNN-based computer vision technique analyzes the skin images and detects the patterns that resemble shingles. The system can help healthcare professionals to make timelier and high-quality decisions concerning the treatment of patients, which leads to an earlier diagnosis and a reduction in patient outcomes. In this work, a custom shingles image set has been generated using CNN model with different combination of activation function and optimizer that compares to find the best combination to generate high performance rate. Here, the images are augmented to make it more general and finally achieved an accuracy rate of 97.1% with a binary entropy loss of 0.097 using the GELU activation function in CNN layers and sigmoid activation function in the output layer.