Novel Computerised Techniques for Recognition and Analysis of Diabetic Foot Ulcers

Goyal, Manu (2019) Novel Computerised Techniques for Recognition and Analysis of Diabetic Foot Ulcers. Doctoral thesis (PhD), Manchester Metropolitan University.

Preview

Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (24MB) | Preview

Abstract

Diabetic Foot Ulcers (DFU) that affect the lower extremities are a major complication of Diabetes Mellitus (DM). It has been estimated that patients with diabetes have a lifetime risk of 15% to 25% in developing DFU contributing up to 85% of the lower limb amputation due to failure to recognise and treat DFU properly. Current practice for DFU screening involves manual inspection of the foot by podiatrists and further medical tests such as vascular and blood tests are used to determine the presence of ischemia and infection in DFU. A comprehensive review of computerized techniques for recognition of DFU has been performed to identify the work done so far in this field. During this stage, it became clear that computerized analysis of DFU is relatively emerging field that is why related literature and research works are limited. There is also a lack of standardised public database of DFU and other wound-related pathologies. We have received approximately 1500 DFU images through the ethical approval with Lancashire Teaching Hospitals. In this work, we standardised both DFU dataset and expert annotations to perform different computer vision tasks such as classification, segmentation and localization on popular deep learning frameworks. The main focus of this thesis is to develop automatic computer vision methods that can recognise the DFU of different stages and grades. Firstly, we used machine learning algorithms to classify the DFU patches against normal skin patches of the foot region to determine the possible misclassified cases of both classes. Secondly, we used fully convolutional networks for the segmentation of DFU and surrounding skin in full foot images with high specificity and sensitivity. Finally, we used robust and lightweight deep localisation methods in mobile devices to detect the DFU on foot images for remote monitoring. Despite receiving very good performance for the recognition of DFU, these algorithms were not able to detect pre-ulcer conditions and very subtle DFU. Although recognition of DFU by computer vision algorithms is a valuable study, we performed the further analysis of DFU on foot images to determine factors that predict the risk of amputation such as the presence of infection and ischemia in DFU. The complete DFU diagnosis system with these computer vision algorithms have the potential to deliver a paradigm shift in diabetic foot care among diabetic patients, which represent a cost-effective, remote and convenient healthcare solution with more data and expert annotations.