Skin lesion classification using HG-PSO and YOLOv7 based convolutional network in real time

Area of Science:

• Dermatology
• Medical Imaging
• Artificial Intelligence

Background:

• Skin cancer diagnosis relies on visual inspection and dermoscopy, demanding precise lesion localization and classification.
• Early and accurate diagnosis is critical for effective treatment and improved patient prognosis.
• Existing methods face challenges in achieving high accuracy and efficiency in classifying skin lesions.

Purpose of the Study:

• To develop and evaluate a hybrid optimization and deep learning model for enhanced skin cancer classification.
• To improve the accuracy, precision, and recall in identifying and classifying skin lesions from dermoscopic images.
• To assess the method's performance against state-of-the-art techniques on diverse datasets.

Main Methods:

• A hybrid approach combining Genetic and Particle Swarm Optimization (HG-PSO) with a You Only Look Once version 7 (YOLOv7) convolutional neural network was proposed.
• Optimized YOLOv7 was used for initial lesion detection, followed by color thresholding for segmentation.
• The segmented regions were then classified using the proposed convolutional network.

Main Results:

• The method achieved high performance across multiple datasets: 98.86% accuracy on HAM10000, 97.10% on ISIC-2019, and 97.7% on PH2.
• Excellent average precision, recall, and F1-scores were recorded, demonstrating robust classification capabilities.
• The model exhibited fast processing times (2-3 seconds), suitable for real-time applications like telemedicine.

Conclusions:

• The proposed HG-PSO and YOLOv7-based method offers a significant advancement in automated skin cancer classification.
• Its high accuracy and efficiency make it a promising tool for early detection and diagnosis of skin cancer.
• The method's speed and performance suggest potential for integration into clinical workflows and telemedicine platforms.