Related Experiment Video
Updated: Jun 4, 2025
06:08
Quantitative Visualization and Detection of Skin Cancer Using Dynamic Thermal Imaging
Published on: May 5, 2011
16.8K
The Depth Estimation and Visualization of Dermatological Lesions: Development and Usability Study.
Pranav Parekh1, Richard Oyeleke1, Tejas Vishwanath2
1Stevens Institute of Technology, Hoboken, NJ, United States.
JMIR Dermatology
|December 18, 2024
Summary
This study introduces a novel method for estimating skin lesion depth using AI and 3D holograms, improving melanoma diagnosis. The technique accurately differentiates benign and malignant lesions, offering a noninvasive tool for dermatologists.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Dermatology
Background:
- Accurate skin lesion depth estimation is crucial for clinical staging, yet current methods are often approximate.
- Noninvasive techniques for depth estimation from 2D images are needed, especially for rapidly growing malignant lesions.
Purpose of the Study:
- To propose a novel methodology for skin lesion depth estimation and visualization using AI and 3D holograms.
- To generate 3D visualizations aiding dermatologists in diagnosing melanoma and other skin conditions.
- To provide a definite estimate and visualization procedure for skin lesions beyond current approximate methods.
Main Methods:
- Convolutional Neural Network (CNN) for lesion classification, coupled with explainable AI (gradient class activation mapping) for feature localization.
- Computer graphics and depth from defocus methods for 3D structure and depth estimation from single images.
- Red spot analysis for infection degree measurement and Gabor filters for volumetric depth map representation, culminating in conical hologram generation.
Main Results:
- The combined CNN and explainable AI model achieved 86% accuracy in classifying lesions as benign or malignant.
- Benign and malignant cases were successfully mapped to distinct conical representations, with malignant lesions showing deeper structures.
- Dermatologists provided positive feedback, recognizing the potential clinical utility of the 3D hologram visualization for lesion depth estimation.
Conclusions:
- Malignant lesions exhibit a higher concentration of red spots and deeper conical sections compared to benign lesions, correlating with CNN classification.
- The qualitative assessment of the 3D holograms aligns with the initial AI-based classification, validating the method's accuracy.
- The positive reception from dermatologists indicates the method's potential as a valuable clinical tool for noninvasive lesion depth assessment.
