SADASNet: A Selective and Adaptive Deep Architecture Search Network with Hyperparameter Optimization for Robust Skin...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Skin cancer is a significant public health issue requiring early diagnosis and effective treatment.
• Deep learning and computer vision show promise for improving skin cancer diagnostic accuracy.
• A gap exists in optimizing deep learning architectures for high accuracy and low computational complexity in skin cancer detection.

Purpose of the Study:

• To develop novel deep learning architectures for multi-class skin cancer classification using metaheuristic optimization.
• To address the need for high accuracy and reduced computational complexity in skin cancer diagnosis.
• To introduce the Selective and Adaptive Deep Architecture Search Network by Hyperparameter Optimization (SADASNet) algorithm.

Main Methods:

• Utilized Particle Swarm Optimization (PSO) to develop the SADASNet algorithm for deep learning architecture search.
• Applied innovative data augmentation techniques to the HAM10000 dataset to address class imbalance.
• Designed SADASNet to accommodate various image sizes, resulting in six distinct deep learning models.

Main Results:

• Achieved high performance metrics: 99.31% accuracy, 97.58% F1 score, 97.57% recall, 97.64% precision, and 99.59% specificity.
• Demonstrated superior performance over state-of-the-art methods in accuracy and computational complexity.
• Maintained a broad solution space during hyperparameter optimization.

Conclusions:

• The proposed SADASNet method effectively enhances skin cancer classification accuracy.
• This research contributes to the advancement of deep learning applications in medical diagnostics.
• The findings support improved early detection and treatment strategies for skin cancer.