Abstract
Skin lesion diagnosis has recently gotten a lot of attention. Physicians spend
a lot of time analyzing these skin lesions because of their striking similarities.
Clinicians can use a deep learning-based automated classification system to identify the
type of skin lesion and enhance the quality of medical services. As deep learning
architecture progresses, skin lesion categorization has become a popular study topic. In
this work, a modern skin lesion detection system is provided using a new segmentation
approach known as wide-ShuffleNet. The entropy-based weighting technique is first
computed, and a first-order cumulative moment algorithm is implemented for the skin
picture. These illustrations are used to differentiate the lesion from the surrounding
area. The type of melanoma is then established by sending the segmentation result into
the wide-ShuffleNet, a new deep-learning structure. The proposed technique was
evaluated using multiple huge datasets, including ISIC2019 and HAM10000.
According to the statistics, EWA and CAFO wide-ShuffleNet are more accurate than
the state-of-the-art approaches. The suggested technology is incredibly light, making it
ideal for flexible healthcare management.