Estimating and Visualizing Skin Lesion Depth
Author Information
Author(s): Dellavalle Robert, Kristensen-Cabrera Alexandria, Eapen Bell, Xu Weilin, Parekh Pranav, Oyeleke Richard, Vishwanath Tejas
Primary Institution: Stevens Institute of Technology
Hypothesis
Can a novel methodology using machine learning and mixed reality improve the depth estimation and visualization of skin lesions?
Conclusion
The study successfully developed a method to estimate and visualize the depth of skin lesions, showing that malignant lesions have greater depth and concentration of infection compared to benign ones.
Supporting Evidence
- The neural model achieved an accuracy of 86% in classifying lesions.
- Positive feedback from dermatologists indicated the method's potential clinical utility.
- The study demonstrated that malignant lesions have deeper conical sections compared to benign cases.
Takeaway
This study created a way to see how deep skin problems go under the surface, helping doctors figure out if a skin issue is serious or not.
Methodology
The study used convolutional neural networks for classification, explainable AI for localization, and computer graphics for depth estimation and 3D visualization.
Potential Biases
The dataset is skewed towards lighter skin tones, which may affect the model's performance on darker skin.
Limitations
The study's dataset may not represent all skin tones, and the depth estimation relies on the color and width of lesions.
Participant Demographics
The dataset included 1497 malignant and 1800 benign skin cancer images.
Statistical Information
P-Value
0.86
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website