Related Experiment Video
Updated: Jun 26, 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
Multi-scale feature fusion and class weight loss for skin lesion classification
Zhentao Hu1, Weiqiang Mei1, Hongyu Chen1
1School of Artificial Intelligence, Henan University, Zhengzhou, 450046, China.
Computers in Biology and Medicine
|May 18, 2024
Summary
Deep learning models can now detect skin cancer more accurately using a new multi-scale fusion structure. This approach combines shallow and deep features, improving early detection of skin cancer from dermoscopic images.
Area of Science:
- Dermatology and Artificial Intelligence
- Medical Image Analysis
- Computational Pathology
Background:
- Skin cancer poses a significant health threat due to its rapid spread and difficulty in early detection.
- Deep learning shows promise for skin cancer detection in dermoscopic images, but faces challenges like inter-class similarity and intra-class variation.
- Accurate classification of skin lesions is crucial for timely diagnosis and treatment.
Purpose of the Study:
- To develop a more accurate deep learning model for skin cancer classification using dermoscopic images.
- To address challenges in skin lesion image analysis, including class imbalance and feature variations.
- To evaluate the impact of specific features, such as hair and lesion regions, on classification performance.
Main Methods:
- A novel multi-scale fusion structure combining shallow and deep features was proposed for enhanced classification.
- Techniques including class weighting, label smoothing, and resampling were implemented to handle class imbalance.
- Experiments were conducted using the HAM10000 and ISIC2019 datasets, including variations like HAM10000_RE (hair removed) and HAM10000_SE (segmented lesion regions).
Main Results:
- The proposed multi-scale fusion model achieved high accuracy on benchmark datasets.
- On the HAM10000 dataset, the model reached an accuracy (ACC) of 94.0% and an area under the curve (AUC) of 99.3%.
- The model demonstrated strong performance on the ISIC2019 dataset with an ACC of 89.8%, outperforming state-of-the-art models.
Conclusions:
- The multi-scale fusion structure effectively improves the accuracy of deep learning-based skin cancer classification.
- Addressing class imbalance and understanding feature importance (e.g., lesion region) are critical for robust diagnostic models.
- The developed model shows excellent potential for clinical application in dermoscopic image analysis for skin cancer detection.
