Rectification of Perspective and Scale Distortion in Wound Images and Integrated Evaluation of the SmartWAnDS System

Raghav, Fnu

Etd

Rectification of Perspective and Scale Distortion in Wound Images and Integrated Evaluation of the SmartWAnDS System

Public

Wound assessment using digital images has recently emerged as a viable method of remote analysis and assessment, which provides healthcare providers with a longitudinal view of the wound's healing progress. Prior work has proposed methods to analyze wound images including detecting the wound, classifying its type, segmenting the wound area and generating comprehensive assessments based on validated wound grading rubrics. To assist wound nurses with their decisions, the Smartphone Wound Analysis and Decision Support system (SmartWAnDS) is being researched and developed with the aim of generating wound care recommendation decisions to wound nurses by analyzing chronic wound images taken using smartphones along with Electronic Health Records (EHR) information of patients. The research done in this thesis has two parts 1) Rectification of geometric distortion in wound image and 2) Comprehensive, integrated evaluation of the SmartWAnDS system. The accuracy of wound image assessment tools relies on the quality of the image captured with a camera. Prior work has found wound size as the most important single visual attribute for wound assessment. The image captured using a digital camera utilizes perspective projection in which parallel lines converge to a vanishing point, which suffers from geometric and scale distortion due to non-standard camera angles and orientations. This distortion introduces errors into wound sizes calculated. The orthogonal image is the image captured with a camera placed at a surface normal and fixed distance from the wound surface. The orthogonal image is free from perspective and scale distortions and facilitates more accurate wound size assessment. In this thesis, We propose and evaluate using a Faster RCNN for wound localization and a deep Spatial Transformation Network (STN) based system for automatic rectification of the perspective and scale distortion from wound images. We consider both real and artificial wounds data with images captured by smartphone from different angles and distances. Secondly, the integrated SmartWAnDS system prototype consisting of multiple subsystems was evaluated: Image light correction, Image geometric correction, Wound segmentation, wound score generator and care decision recommendation system. We perform multiple evaluation steps consisting of different experiments to evaluate the integrated system performance of various system configurations and the addition of each subsystem. We discuss our findings and present conclusions.

Creator