Development of a scanner to prevent wounds in the feet of patients with diabetes.

Diabetes Mellitus (DM) is one of the most common chronic diseases globally. Currently more than 382 million people are diagnosed with DM. The population and significance of DM are expected to rise even further in the future. People with DM are at risk of developing a number of disabling and life-threatening health problems, such as diabetic foot complications, including ulceration and its pre-signs (e.g. abundant callus, inflammation, fissures, blisters). Approximately 25% of patients with DM eventually develop foot ulcers. If not treated in time, the risk for (partial) lower extremity amputations or mortality increases. These devastating consequences can be prevented by early detection and timely treatment of the diabetic foot complications. This early identification strongly depends on frequent risk assessment, preferably on a daily basis, especially for high-risk patients. However, frequent assessment by healthcare professionals is costly and not always possible. Self-examination by patients is difficult and impractical, due to other complications (e.g. limited joint mobility and reduced eyesight).

An intelligent telemedicine system, which is compact, non-invasive, non-contact and user-friendly, may provide a solution for frequent assessment, which is the ultimate objective of our project. As the first step to approach this goal, an experimental setup that combined three promising imaging modalities, namely spectral imaging, infrared thermal imaging, and photometric stereo imaging, was developed and investigated. For the design of spectral imaging system, a limited number of optical bandpass filters were selected by employing an additional spectroscopy system with a spectrometer to measure regions of interest that were assessed and annotated by clinical experts. The filter selection boiled down into feature selection in machine learning technology to signify the difference between different skin status. It was found that the required number of optical filters for a spectral imaging system ranged from three to seven, depending on additional constrains.

Following, with the seven preselected filters, the spectral imaging system was built in our experimental setup in cooperation with eleven other filters that were selected based on visual inspection of the absorption spectra of the skin chromophores. To avoid possible mechanical vibrations and to decrease the image acquisition time, a 3 × 3 cam-era array structure fitted with a sliding plate holding two 3 × 3 filter arrays was applied, instead of the traditional filter wheel structure. With the spectral images acquired, front-end pixel classifiers were designed for automatic detection of diabetic foot complications, taking image annotations based on the live assessment as the ground truth. These front-end pixel classifiers could distinguish presence or absence of diabetic foot complications with acceptable performance. How-ever, they were lack of the capability of differentiating different diabetic foot complications. Future studies on enhancing performance of the pixel classifiers and designing the back-end classifiers are needed.

With the infrared thermal imaging, the temperature difference between corresponding points on the left and right foot in the thermal images were compared to detect the risk of inflammation. As the temperature contrast between the feet and background was low, it was hard to perform the foot segmentation directly from the thermal images. To issue this, foot segmentations from digital color images were conducted for assistance. Non-rigid landmark-based registration with B-splines was applied to register the left and right foot, which helped to compare the temperature of the corresponding points on the contralateral feet. The outcomes of the thermal image analysis proved to be promising in the early detection of foot complications in patients with DM and patients who are at high risk for these complications. However, this methodology was limited to patients, for whom both feet were available in the acquired images.
With photometric stereo imaging, a feasibility study was conducted to detect diabetic foot complications. The surface curvature (derived from the reconstructed surface normal) and the reconstructed surface albedo were treated as input features to distinguish the skin regions with diabetic foot complications present or absent. The results indicated that this imaging technology was promising, although it still had some limitations to apply the photometric stereo imaging for the detection of the diabetic foot complications currently, such as the movement in patients’ foot during image acquisition. To determine the potential value of this modality in the future telemedicine system, further improvement was required.

The outcomes of the studies presented in this thesis showed the feasibility of developing a telemedicine system to detect diabetic foot complications with the three imaging modalities. The studies acted as the precursors for developing such an intelligent telemedicine system, which proposed potential detection methodologies and provided the directions for the future study. Following the conclusion and recommendations of this thesis, we expect an intelligent telemedicine system suitable for measurements at the patients’ homes or at health centres can be developed in the future.

It is possible to develop a scanner that can reliably detect (pre-signs of) ulceration by hyperspectral, photometric and thermal measurements.

N/A

A scan with hyperspectral, photometric and thermal measurements will be made from patients' feet.