Deep learning approach for skin melanoma and benign classification using empirical wavelet decomposition.

Area of Science:

• Dermatology
• Medical Imaging
• Signal Processing

Background:

• Melanoma is a deadly skin cancer where early detection significantly improves patient outcomes.
• Computer-assisted analysis of skin lesion images can aid in differentiating melanoma from benign lesions.
• Extracting image features is crucial for early and accurate diagnosis.

Purpose of the Study:

• To introduce and evaluate a novel Empirical Wavelet Decomposition (EWD) model utilizing tan hyperbolic modulated filter banks (THMFBs) for skin lesion image feature extraction.
• To compare the performance of the new EWD-THMFBs model against existing decomposition methods.

Main Methods:

• The study employed a new Empirical Wavelet Decomposition (EWD) model based on tan hyperbolic modulated filter banks (THMFBs) for feature extraction from skin lesion images using MATLAB.
• The EWD-THMFBs model was benchmarked against the empirical short-time Fourier decomposition method based on THMFBs (ESTFD-THMFBs) and the empirical Fourier decomposition method based on THMFBs (EFD-THMFBs).

Main Results:

• The EWD-THMFBs model demonstrated superior performance, achieving 100% accuracy and an Area Under the Curve (AUC) of 1.
• In comparison, the ESTFD-THMFBs model yielded 98.89% accuracy and an AUC of 0.97.
• The EFD-THMFBs model achieved 83.33% accuracy and an AUC of 0.91.

Conclusions:

• The EWD-THMFBs model is highly effective for extracting features from skin lesion images, outperforming other tested methods.
• This model's success suggests its potential for integration into mobile applications for preliminary skin lesion detection by nurses in underserved regions.
• Early detection facilitated by this technology could significantly impact melanoma patient survival rates.