LesionNet: an automated approach for skin lesion classification using SIFT features with customized convolutional
Sarah A Alzakari1, Stephen Ojo2, James Wanliss2
1Department of Computer Sciences, College of Computer and Information Sciences, Princess Nourah Bint Abdulrahman University, Riyadh, Saudi Arabia.
Frontiers in Medicine
|November 5, 2024
Summary
This study introduces LesionNet, a computer-aided diagnosis tool for skin lesion detection. Combining Scale-Invariant Feature Transform (SIFT) with a custom convolutional neural network, it achieves 99.28% accuracy in identifying melanocytic lesions.
Area of Science:
- Dermatology and Medical Imaging
- Artificial Intelligence in Healthcare
Background:
- Manual skin lesion analysis faces inefficiencies and errors.
- Automated diagnostic systems require robust solutions for image variability and artifacts.
Purpose of the Study:
- To develop an accurate computer-aided diagnosis system for skin lesion detection.
- To enhance the precision of distinguishing melanocytic lesions from normal skin.
Main Methods:
- Utilized the HAM10000 dataset for training and validation.
- Integrated Scale-Invariant Feature Transform (SIFT) for feature extraction.
- Developed and applied a custom convolutional neural network model named LesionNet.
Main Results:
- Achieved a high accuracy of 99.28% in skin lesion classification.
- Demonstrated the model's robustness against common image artifacts.
- Successfully distinguished melanocytic lesions from normal skin conditions.
Conclusions:
- Combining SIFT features with LesionNet significantly improves skin lesion detection accuracy.
- The developed model shows strong potential for clinical application in dermatology.
- Advanced neural network models are effective for precise automated skin lesion analysis.
