Integration of handcrafted and deep-level features to improve skin disease detection.

As per the World Health Organization, skin disorders are some of the most common health issues, constituting a significant source of non-fatal disease burden and greatly affecting both quality of life and healthcare systems. Melanoma skin cancer is the most fatal disease in the world, and nearly 60,000 people died due to melanoma in 2022. Skin disorders affect millions of people globally and pose a serious threat to public health. These disorders can affect the skin's composition, function, and appearance. To detect and predict skin diseases, a comprehensive physical examination, a review of the patient's medical history, and appropriate laboratory testing are required. Noise, blur, uneven illumination, inadequate feature representation, and overlapping features between distinct skin disorders all reduce the efficacy of skin disease identification. The Skin Disease Detection Network (SDNet), a deep convolutional neural network-based two-way feature depiction system for multiclass skin disease identification, is presented in this study. To depict hierarchical features in skin images, the SDNet employs two parallel arms. The first arm uses a 2D CNN in conjunction with pre-processed original images, while the second arm uses a 1D CNN that accepts Gray Level Co-occurrence Matrix features (GLCM), improved Local Binary Pattern (ILBP), and Histogram of Oriented Gradient (HOG) features to depict texture and shape attributes of skin images. The accuracy, recall, precision, and F1-score of SDNet results are evaluated using the DermNet dataset. The proposed SDNet achieves an overall accuracy of 99.1%, a recall of 99.1%, a precision of 98.96%, and an F1-score of 98.95% for the 5-class skin disease detection, demonstrating a notable improvement over the performance of traditional state-of-the-art methods. This study marks a notable step forward in leveraging the reliability of SDNet for precise and efficient skin disease identification through the Explainable Artificial Intelligence (XAI) approach.