Related Experiment Video
Updated: May 29, 2025
09:37
Combining Reflectance Confocal Microscopy with Optical Coherence Tomography for Noninvasive Diagnosis of Skin Cancers via Image Acquisition
Published on: August 18, 2022
2.2K
Towards unbiased skin cancer classification using deep feature fusion
Ali Atshan Abdulredah1, Mohammed A Fadhel2, Laith Alzubaidi3
1National School of Electronics and Telecoms of Sfax, University of Sfax, Sfax, Tunisia.
BMC Medical Informatics and Decision Making
|January 31, 2025
Summary
SkinWiseNet (SWNet), a novel deep learning model, achieves high accuracy in detecting malignant skin cancer. Its feature fusion approach enhances classification, addressing biases for diverse skin tones and hair conditions.
Area of Science:
- Artificial Intelligence
- Medical Imaging
- Dermatology
Background:
- Skin cancer detection remains a critical challenge in healthcare.
- Existing deep learning models may exhibit biases related to skin tone and hair.
- Accurate and early diagnosis is crucial for improving patient outcomes.
Purpose of the Study:
- To introduce SkinWiseNet (SWNet), a deep convolutional neural network for skin cancer detection and classification.
- To enhance model efficiency and address potential biases in datasets.
- To improve the accuracy and interpretability of AI-driven skin cancer diagnosis.
Main Methods:
- Developed SWNet, a deep convolutional neural network with optimized feature extraction and network width augmentation.
- Implemented feature fusion to integrate insights from diverse datasets, mitigating bias.
- Utilized Explainable Artificial Intelligence (XAI) techniques, specifically Grad-CAM, for model interpretability.
- Conducted experiments on four public datasets: Mnist-HAM10000, ISIC2019, ISIC2020, and Melanoma Skin Cancer.
Main Results:
- SWNet achieved superior performance compared to EfficientNet, MobileNet, and Darknet.
- Attained an accuracy of 99.86% and an F1 score of 99.95%.
- Demonstrated effective gradient propagation and feature capture across multiple levels.
Conclusions:
- SWNet shows significant potential for advancing skin cancer detection and classification.
- Feature fusion effectively enhances accuracy and reduces biases related to hair and skin tones.
- The model offers a robust tool for accurate, early skin cancer diagnosis, improving patient care.
