Fast Nail Segmentation for Teledermatology Encounters with an End-to-End Masked R-CNN Architecture

Abstract

Image segmentation is the process of identifying boundaries of objects in an image. Modern applications of computer vision in dermatology rely heavily on these methods to standardize data inputs and focus directly on regions of interest. A major barrier to the development of performant computer vision models for point-of-care clinical use is the heterogeneity of image quality, especially in the context of teledermatology when patients are tasked to show dermatological conditions without specialized cameras. Previous authors have developed nail-bed segmentation methodologies trained on clinical photography, which is of much higher consistency and quality than patient-generated images in clinical practice. Bridging the gap between advances in computer vision and its use in clinical practice, we examine whether using sub-clinical-quality photography hinders image segmentation and detection of nail beds using a near real-time architecture more suited to on-demand clinical practice.

Accepted as a Poster Presentation to the Atlantic Derm Conference 2023

Previous
Previous

Pemphigus Foliaceus following treatment with Tirzepatide

Next
Next

Role of RNA-LPX in Melanoma Immunotherapy