Related Experiment Video
Updated: Dec 31, 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
2.8K
Augmented decision-making for acral lentiginous melanoma detection using deep convolutional neural networks
1Department of Dermatology, Yonsei University Wonju College of Medicine, Wonju, Korea.
Summary
Convolutional neural networks (CNNs) can significantly improve physician accuracy in detecting acral lentiginous melanoma (ALM). Integrating CNNs as a decision-support tool enhances diagnostic performance and expert concordance.
Area of Science:
- Dermatology
- Artificial Intelligence
- Medical Imaging
Background:
- Convolutional neural networks (CNNs) show high performance in melanoma detection.
- Limited research exists on how CNNs impact physician diagnostic performance.
Purpose of the Study:
- Develop a CNN for acral lentiginous melanoma (ALM) detection.
- Evaluate if CNN implementation improves physician diagnostic decisions for ALM.
Main Methods:
- Trained a CNN on 1072 dermoscopic images (benign nevi, ALM, intermediate tumors).
- Conducted a three-stage survey with 60 physicians, progressively adding clinical info and CNN predictions.
- Assessed diagnostic accuracy and inter-physician concordance at each stage.
Main Results:
- Physician accuracy increased from 74.7% (Stage I) to 79.0% (Stage II) and 86.9% (Stage III).
- CNN integration (Stage III) improved accuracy by 12.2%p over Stage I and 7.9%p over Stage II.
- Inter-physician concordance significantly increased from Fleiss-κ 0.436 to 0.684 with CNN use.
Conclusions:
- AI-augmented decision-making enhances physician performance and agreement in ALM diagnosis.
- CNNs show potential as decision-support systems for clinicians.
- This technology can aid in improving diagnostic accuracy for challenging skin lesions.
