Skin cancer detection through attention guided dual autoencoder approach with extreme learning machine

Area of Science:

• Dermatology
• Medical Imaging
• Artificial Intelligence

Background:

• Early skin cancer detection is crucial for preventing metastasis and improving patient outcomes.
• Artificial Intelligence (AI) offers promising automated solutions for enhanced diagnostic accuracy.

Purpose of the Study:

• To present DualAutoELM, a novel AI-based approach for the effective identification of various skin cancer types.
• To leverage a dual autoencoder network with attention mechanisms for improved feature learning in skin lesion images.

Main Methods:

• The DualAutoELM method utilizes a spatial autoencoder and a Fast Fourier Transform (FFT)-autoencoder to learn distinct image features.
• Attention modules are integrated within the autoencoder encoders to enhance discriminative feature extraction.
• A single-layer feedforward Extreme Learning Machine (ELM) classifies skin malignancies based on learned features.

Main Results:

• The DualAutoELM approach achieved high performance on the HAM10000 dataset (AUC: 0.98, Precision: 97.68%, Accuracy: 97.66%).
• On the ISIC-2017 dataset, the method demonstrated strong results (AUC: 0.95, Precision: 86.75%, Accuracy: 86.68%).
• Experimental findings confirm the accuracy and robustness of the proposed AI technique for skin cancer detection.

Conclusions:

• The DualAutoELM approach shows significant potential for accurate and early detection of skin cancer.
• The integration of spatial and frequency domain analysis via autoencoders enhances diagnostic capabilities.
• This AI-driven method offers a promising tool for improving skin cancer diagnosis and patient care.