Chronic Wound Healing Assessment System Based on Different Features Modalities and Non-Negative Matrix Factorization (NMF) Feature Reduction

Identification and treatment of chronic wounds (CWs) are considered economic and social challenges, especially with respect to bedridden and elderly persons. CWs do not follow a predictive course of healing within a particular period. Their treatment and management costs are very high. Also, CWs decrease quality of life for patients, which cause severe pain and discomfort. In this paper, we proposed a comprehensive wound healing assessment framework based on current appearance, texture, and prior visual appearance analysis to handle different types of CWs depending on extracting various tissue types. The framework provided an accurate evaluation tool for the CW healing process depending on extracting and fusing significant features from the CWs RGB images. Non-negative matrix factorization (NMF) was used to retrieve the most significant features to reduce computation time. The gradient boosted trees (GBT) classifier was used to classify different tissue types. Finally, the healing assessment of the CWs depended on calculating the improvement in the area of the necrotic eschar, slough, granulation, and healing epithelial tissues. The framework was trained and tested using 377 RGB images from Medetec wound database and national pressure ulcer advisory panel website. The proposed system achieved an average accuracy of 96% for tissue classification, which helps in obtaining an accurate CW healing assessment. This result can be considered as a promising result when compared to the other state-of-the-art techniques.