Detection of diabetic foot with infrared thermography

• Main Author: Muhammad Adam Abdul Rahim
• Other Authors: Ng Yin Kwee, Eddie
• Format: Theses and Dissertations
• Language: English
• Published: 2018
• Subjects: DRNTU::Engineering::Bioengineering
• Online Access: http://hdl.handle.net/10356/74240
• Institution: Nanyang Technological University

Diabetic foot is among the major complications of Diabetes Mellitus (DM) that can results in ulcerations. An early detection and appropriate treatment can prevent traumatic outcomes such as lower extremity amputation. Furthermore, studies have shown that temperature fluctuations on the plantar foot can be related to diabetic foot complications. Since the time of Hippocrates, human body temperature is being associated to health. Thermography is a non-invasive imaging method employed to view the thermal patterns using Infrared (IR) camera. In addition, it allows the qualitative and visual documentation of temperature fluctuation in vascular tissues. Absolutely, IR thermography can play a crucial role in the medical field, especially for detection of diabetic foot disease. Nonetheless, such strategy is yet to be established and many circumstances still unsolved. The first challenge is analysing the thermal changes because the risks of ulceration are associated to the rise in plantar temperatures. Nevertheless, the attempts to identify the spatial patterns of temperature remains difficult because there are extensive forms of thermal patterns. This complicates the classification process. In addition, the interpretation of thermographic patterns may be difficult due to inadequate details, even in healthy subjects, and classification techniques on the thermographic patterns. Moreover, there is no objective approach in allocating the thermogram to certain classes. Recent studies proposed the temperatures of one foot and contralateral foot to be measured and compared, and the diabetic foot risk regions are detected by the defined threshold. However, this is only applicable in without ulceration or amputation cases and in asymmetric thermal changes. Therefore, this research developed a novel computer aided detection (CAD) system for diabetic foot using nonlinear features extracted from plantar foot thermograms. A total of 33 healthy volunteers and 33 non-neuropathy diabetic patients are recruited from Ngee Ann Polytechnic (NPIRB-P0175-2017-ECE-AMA6) and Singapore General Hospital (SGH), Diabetes & Metabolism Center (DMC) (CIRB Ref: 2016/3044) respectively. The thermograms acquired are pre-processed by segmenting the plantar foot regions using polygon and then proceed to warp all the segmented plantar foot regions into uniform size. Afterward, the warped grayscale foot images are decomposed using Discrete Wavelet Transform (DWT) and Higher Order Spectra (HOS) prior to extracting texture and entropy features. The features values from left and right foot are subtracted. Subsequently, student t-test is applied on the resultant features to select and rank the significant features. Lastly, the 27 significantly ranked features (p value < 0.0001) are fed independently into support vector machine (SVM) classifier. The developed methodology achieved maximum classification accuracy of 89.39%, positive predictive value of 96.43%, sensitivity of 81.81% and specificity of 96.97% using only five features. The findings of this research will redound to benefit the diabetic patients considering that IR thermography has the potential to detect diabetic foot early. The rising occurrence and burden of diabetic foot disease prove the need for more efficient and reliable early detection methods. Thus, the developed thermography-based CAD system can be an adjunctive diabetic foot screening tool in clinics to effectively assist podiatrist in decision making and diagnosis processes. This is important as early detection of diabetic foot may prevent future complications. Equally important, this research will help to unveil critical areas in image processing techniques that many diabetic foot studies have yet to explore. Thus, the development of new and improve methodology on early diabetic foot detection. The future research intends to acquire and work on more plantar foot thermograms clinical data, which will include diabetic neuropathy patients. This is to extensively study and ascertain the variations in the plantar foot thermal imaging patterns and distributions among the healthy subjects and, diabetic subjects without and with neuropathy. Further, future research plans to develop a novel three classes methodology on diabetic foot detection with the implementation of enhanced image processing and statistical analysis methods that can efficaciously interpret and analyse the huge and diverse plantar foot thermograms data.