Hybrid Model with Wavelet Decomposition and EfficientNet for Accurate Skin Cancer Classification

Area of Science:

• Medical imaging analysis
• Artificial intelligence in dermatology
• Computational pathology

Background:

• Deep learning faces significant challenges in medical image analysis, particularly with detecting, diagnosing, and classifying pathologies.
• Data imbalance, variability, and complexity are major hurdles in developing effective deep learning models for medical imaging.
• Skin disease detection is complex due to the high variability in lesion appearance, texture, color, and localization.

Purpose of the Study:

• To develop an innovative and robust hybrid deep learning architecture for enhanced skin disease detection and classification.
• To address the challenges of data variability and complexity in medical image analysis using deep learning.
• To improve the accuracy and speed of identifying skin pathologies from medical images.

Main Methods:

• Proposed a hybrid architecture integrating wavelet decomposition with EfficientNet models.
• Employed advanced data augmentation techniques to handle data variability.
• Utilized optimized loss functions and training strategies for improved model performance.

Main Results:

• The hybrid model achieved an accuracy of 94.7% on the HAM10000 dataset.
• The model demonstrated 92.2% accuracy on the ISIC2017 dataset.
• The proposed approach effectively addressed challenges related to data imbalance and complexity in skin lesion classification.

Conclusions:

• The synergistic integration of wavelet decomposition and EfficientNet offers a robust solution for medical image analysis.
• This hybrid deep learning approach significantly improves the accuracy of skin disease detection and classification.
• The findings highlight the potential of advanced deep learning techniques in overcoming limitations in current medical imaging diagnostics.