MelanomaNet: An Effective Network for Melanoma Detection.

Area of Science:

• Dermatology
• Medical Imaging
• Artificial Intelligence

Background:

• Melanoma is a dangerous skin cancer requiring accurate detection.
• Manual dermoscopy for melanoma diagnosis is time-consuming and prone to errors due to subtle image variations.
• Existing automated methods lack robustness and generalization capabilities.

Purpose of the Study:

• To develop an effective and robust automated algorithm for melanoma detection.
• To enhance the accuracy and efficiency of skin lesion classification using deep learning.

Main Methods:

• Proposed MelanomaNet architecture integrating Inception-v4 with residual-squeeze-and-excitation (RSE) blocks.
• Inception-v4 extracts rich spatial features and increases feature diversity.
• RSE blocks recalibrate feature channels, improving feature representation.
• Support Vector Machine (SVM) used for final skin lesion classification.

Main Results:

• The MelanomaNet model demonstrated superior performance compared to existing state-of-the-art methods.
• Evaluated on the ISIC 2018 skin lesion challenge dataset, confirming its effectiveness.
• The RSE blocks enhanced feature learning and classification accuracy.

Conclusions:

• MelanomaNet offers a promising solution for accurate and efficient melanoma detection.
• The proposed architecture effectively addresses limitations of traditional methods.
• This approach has the potential to aid dermatologists in early melanoma diagnosis.