Integrating 2D Dermatological Photography with 3D Anatomical Surfaces.

Area of Science:

• Dermatology
• Computer Vision
• Medical Imaging

Background:

• Standardized 2D photography is crucial for dermatologic documentation and monitoring.
• Limited views in 2D photos reduce anatomical context and body location information.
• 3D representations offer improved spatial interpretation of multi-view medical imagery.

Purpose of the Study:

• To integrate 2D dermatological images with a 3D surface model using DensePose.
• To create an anatomically grounded 3D representation for mesh-level analysis.
• To recover spatial context and quantify body surface coverage from 2D images.

Main Methods:

• Utilized DensePose, a deep learning framework, to infer dense correspondences between 2D images and a 3D human mesh.
• Integrated multiple full-body 2D photographs (front, back, sides) from 147 subjects with chronic graft-versus-host disease into a 3D mesh.
• Quantified body surface representation and assessed spatial consistency using anatomical landmarks.

Main Results:

• A single full-body view captures an average of 28% of the body surface.
• Adding subsequent views increased average coverage to 50% (2 views), 72% (3 views), and 80% (4 views).
• Median pairwise geodesic distance between landmarks was 4.6 cm, indicating spatial consistency.

Conclusions:

• Integrating 2D dermatological images into 3D models enhances anatomical context and spatial understanding.
• The developed method quantifies body surface coverage, guiding optimal image capture strategies.
• Findings support advancements in dermatological monitoring, education, and communication through improved imaging infrastructure.