Abstract:
Skin cancer is a common ailment that arises predominantly from regular exposure to sunlight
and manifests across various areas of the body. The accurate recognition and Classification of
skin lesions present considerable challenges due to their morphological diversity and the oftensimilar
features between different types of skin malignancies. In current times, the application
of deep learning methodologies in image-based diagnosis of skin lesions has shown promising
results, achieving levels of diagnostic accuracy comparable to those of professional dermatologists.
Researchers developed an automated system that uses Artificial Intelligence (AI) and deep
learning models to enable early detection of skin disease. This research collected skin disease
photographs from recognized online sources. A new framework called Skin-D was then developed
to assess different kinds of skin conditions. To create this model, we combined MobileNet
architecture, residual blocks and dense blocks with a transition layer in a deep neural network.
The Skin-D algorithm is trained using a dataset of 79,665 skin pictures. The Skin-D classification
approach achieved 99%, 98.5%, 97.5% and 89% accuracy on four separate datasets. These
findings indicate that the model delivers positive results and might be utilized by healthcare
practitioners as a diagnostic tool. In terms of accuracy, the Skin-D approach outperformed the
cutting-edge models SkinLesNet and MobileNet V2-LSTM.
