الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 116 |  |  |
| | | from | 116 | | |
Abstract
Diabetic foot (DF) is one of the most common chronic complications of poorly controlled diabetes mellitus (DM). Early diagnosis and adequately treated is difficult by traditional methods. Lately, it has been found a strong relationship between temperature variation and diabetic foot ulcer emergence. In this thesis, there are three techniques that are followed to get an early diagnosis and detection of the diabetic foot grades using thermal images and distinguish among its grades easily. The first technique depends on thermal statistics and the other two techniques depend on image analysis. The thermal statistics technique, by measuring the temperature of the region of interest points in the insole then computing the mean temperature difference between the corresponding points and finally, the decision is taken depends on the mean temperature result values. Experimental results showed that the value of classification accuracy is 84%, sensitivity is 56.71%, and specificity is 95.59%. To improve the thermal statistics performances and avoid errors which result from manually selecting ROI, the other two proposed systems analysis are applied. The first proposed system is based on a machine learning classification and the second proposed system is based on deep learning classification. In the first proposed system, the thermal images are initially segmented then textural; histogram, fan-beam, and discrete wavelet feature are extracted, selected and combined to get more efficient feature vectors. Finally, classified them using k-nearest neighbor (KNN), Support vector machine (SVM), and Decision tree classifiers. Experimental results showed that Fine KNN had a maximum accuracy of 96.8%, a sensitivity of 88.3%, a specificity of 99.1%, Area under Curve (AUC) of 0.972, and losses score of 0.004 using a combined feature vector of textural and histogram features. To improve the first proposed system performance, the second proposed system is introduced to employ deep convolutional neural network (CNN) in this field. In the second proposed system, the segmented thermal foot images are classified by Alexnet, Googlenet, and proposed CNN. Experimental results showed that the proposed CNN has a maximum accuracy of 99.3%, a sensitivity of 97.4%, and a specificity of 99.8%. When comparing the second proposed system to the thermal statistic system, the first proposed system, and the previous studies, the second proposed system approved greater accuracy. Moreover, it offers an automatic diagnostic tool for the diabetic foot and differentiates among its types easily and accurately.