Related Experiment Video
Updated: Jun 14, 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
2.3K
Skin cancer classification leveraging multi-directional compact convolutional neural network ensembles and gabor
1Department of Electronics and Communications Engineering, College of Engineering and Technology, Arab Academy for Science, Technology and Maritime Transport, Alexandria, 21937, Egypt. o.attallah@aast.edu.
Scientific Reports
|September 4, 2024
Summary
This study introduces SCaLiNG, a novel deep learning tool for skin cancer (SC) detection. SCaLiNG improves diagnostic accuracy by combining multiple compact Convolutional Neural Networks (CNNs) with Gabor Wavelets for comprehensive feature analysis.
Area of Science:
- Dermatology and Medical Imaging
- Artificial Intelligence in Healthcare
- Computational Pathology
Background:
- Skin cancer (SC) diagnosis relies on subjective dermatologist evaluation, necessitating objective support tools.
- Current deep learning Computer-Aided Diagnostic (CAD) platforms often use complex single Convolutional Neural Networks (CNNs), limiting their effectiveness.
- Existing CAD tools primarily utilize spatial image information, neglecting textural and frequency attributes crucial for accurate SC classification.
Purpose of the Study:
- To develop an innovative CAD tool, SCaLiNG, that overcomes the limitations of existing deep learning models for skin cancer diagnosis.
- To enhance the accuracy and efficiency of skin cancer classification by integrating diverse feature representations.
- To provide dermatologists with a more reliable and precise tool for identifying and categorizing skin cancer subtypes.
Main Methods:
- SCaLiNG employs a fusion of three compact Convolutional Neural Networks (CNNs) and Gabor Wavelets (GW) to extract spatial, textural, and frequency features.
- Gabor Wavelets decompose images into directional sub-bands, enabling CNNs to learn from multi-directional information alongside the original image.
- A feature selection approach is applied to identify and utilize the most discriminative features, optimizing model performance.
Main Results:
- SCaLiNG achieved a classification accuracy of 0.9170 in categorizing skin cancer subtypes.
- The model demonstrated superior performance compared to conventional single-CNN methodologies.
- The integration of spatial-textural-frequency attributes and feature selection significantly improved diagnostic accuracy.
Conclusions:
- SCaLiNG offers a robust and accurate solution for the computer-aided diagnosis of skin cancer.
- The innovative approach of combining compact CNNs with Gabor Wavelets provides a comprehensive feature representation for improved classification.
- SCaLiNG has the potential to significantly aid dermatologists in the swift and precise recognition and classification of skin cancer, ultimately improving patient outcomes.
