Skin Cancer Classification Using Deep Spiking Neural Network.

Area of Science:

• Medical imaging
• Artificial intelligence
• Computational neuroscience

Background:

• Skin cancer is a leading global cause of death, with diagnosis relying on visual inspection, which can be error-prone.
• Early detection of skin cancer significantly reduces mortality rates, necessitating improved diagnostic tools.
• Computer-aided diagnosis (CAD) systems, particularly those using deep learning, are crucial for enhancing skin cancer identification from lesion images.

Purpose of the Study:

• To develop and evaluate a deep spiking neural network (SNN) model for accurate classification of skin lesions.
• To compare the performance of the proposed SNN model against conventional deep learning models like VGG-13 and AlexNet.
• To explore the potential of SNNs for power-efficient hardware implementation in medical diagnostics.

Main Methods:

• Utilized the ISIC 2019 dataset comprising 3670 melanoma and 3323 non-melanoma images.
• Employed deep spiking neural networks with surrogate gradient descent for image classification.
• Developed a spiking VGG-13 model and compared its performance with standard VGG-13 and AlexNet architectures.

Main Results:

• The proposed spiking VGG-13 model achieved an accuracy of 89.57% and an F1 score of 90.07%.
• The SNN model demonstrated superior performance compared to VGG-13 and AlexNet.
• The spiking VGG-13 model achieved these results with fewer trainable parameters than the compared conventional models.

Conclusions:

• Deep spiking neural networks offer a viable and efficient approach for skin cancer classification.
• The developed spiking VGG-13 model presents a promising advancement in computer-aided diagnosis for dermatology.
• SNNs hold potential for efficient hardware deployment, paving the way for advanced, low-power diagnostic tools.