Analysis & Design of Wound Detection Using Segmentation of Improvised Watershed Segmentation and Particle Swarm Optimization

Abstract

Image analysis techniques for detecting, and classifying the diabetic wound have been proposed in this paper. Different research work has reported the methods for assessing the diabetic wound healing process using conventional processes. The development of automated methods for wound lesion detection and classification using digital images is the major contribution of this research work. The proposed techniques consist of hybrid Fuzzy clustering, watershed segmentation, optimized techniques, and DT methods to improve classification accuracy. The various statistical parameters are proposed to evaluate in terms of sensitivity, specificity and accuracy. These are calculated from a false positive rate and fall negative rate specifying the errors of over and under-segmentation in the execution process of the methods. The different wound lesions having variable sizes were segmented accurately from hyper- and hypo-intensity values. The statistical and morphological features that were characterized by taking the mean and finding the boundary on each lesion were implemented. In classifying the major wound lesions, a high-level machine learning classifier was used to get the most important information about the lesion features and characteristics. Therefore, multi-class classification is implemented and is verified based on the effectiveness in finding the accuracy of different types of wound tissue with a suitable classifier. Commonly, the existing techniques used for specific diseases find limited work on feature analysis and classification for the identification of wound tissue. The obtained results are satisfactory with minimum classification error and high accuracy which is very much comparable to the published results. The findings of our techniques can be used by clinicians for a clear understanding of the wound lesion and therapeutic intervention for curing the infected tissue. Nevertheless, the developed intelligent systems are not meant to reduce the role of physicians and radiologists would rather serve as an alternative for the clinical validation and therapeutic intervention to clinicians.