Deep Learning With Optical Coherence Tomography for Melanoma Identification and Risk Prediction
Pei-Yu Lai1, Tai-Yu Shih1, Yu-Huan Chang1
1Institute of Biophotonics, National Yang Ming Chiao Tung University, Taipei, Taiwan.
Journal of Biophotonics
|October 27, 2024
Summary
This study developed a convolutional neural network (CNN) for early melanoma detection using optical coherence tomography (OCT) imaging. The AI model achieved high accuracy in identifying melanoma and assigning risk scores, aiding clinical diagnosis.
Area of Science:
- Biomedical Imaging
- Artificial Intelligence in Medicine
- Dermatology
Background:
- Malignant melanoma incidence is rising, necessitating advanced diagnostic tools.
- Current noninvasive imaging techniques for melanoma often rely on dermoscopic images.
- Few studies utilize prospective datasets for developing melanoma diagnosis models.
Purpose of the Study:
- To develop and evaluate a convolutional neural network (CNN) for melanoma identification and risk prediction.
- To utilize optical coherence tomography (OCT) imaging for melanoma diagnosis.
- To assess the CNN's performance on longitudinal data from animal models.
Main Methods:
- Development of a CNN model for image analysis.
- Application of the CNN to optical coherence tomography (OCT) images of mice skin.
- Longitudinal testing on four animal models: melanoma, dysplastic nevus, and control groups.
Main Results:
- The CNN achieved high sensitivity (0.99) and specificity (0.98) in classifying melanoma versus healthy tissue.
- The model successfully assigned risk scores based on melanoma probability.
- Accurate classification and risk stratification were demonstrated on prospective datasets.
Conclusions:
- The developed CNN shows significant potential for early melanoma detection and risk stratification using OCT imaging.
- This AI-driven approach may enhance clinical management of melanoma.
- OCT-based CNN models offer a promising avenue for noninvasive skin cancer diagnosis.
