Spect image analysis with deep learning method for diagnosis of parkinson's disease

Parkinson's disease which develops as a result of nerve cells devastation, is the most commonly diagnosed disease after Alzheimer's disease. It is seen in 1 percent of individuals over the age of 55-60 years. Although this ratio seems low, it causes an increase in individuals with Parkinson's disease considering the prolongation of life expectancy. Today, traditional methods are used in the early diagnosis of Parkinson's disease. However, using artificial intelligence technologies, researchers are studying to be detected this disease earlier and more accurately. In this study, it has been tried to contribute to the early and more accurate diagnosis of Parkinson's disease by using SPECT images with deep learning technology. In order to analyze SPECT images, a model was created using convolutional neural network structure and tested with optimization options which is within the Keras library. For each optimization value, different activation functions and epoch have been tested and tried to reach the most successful accuracy rate. Sensitivity and specificity values as well as accuracy rate as the criterion of success for disease diagnosis are calculated and presented as a table. As a result of the deep learning process applied on SPECT image data, which is in the three different classes, the most successful classification method is obtained with Adamax optimization method 94.15% accuracy,91.03% sensitivity and 95.71% specificity. This thesis consists of 5 chapters in general. In the first part, Parkinson's disease and deep learning methods are mentioned and the current studies regarding topic in the related literature are presented. Artificial intelligence and the deep learning, newly developing technology in its subfield, take place in the second part. In this part, deep learning methods and areas of usage are explained with examples. In the third section, detailed information is given about the data set and the structures used in the application are introduced. In the fourth chapter, the accuracy, sensitivity and specificity values obtained according to the optimization methods used were evaluated visually. In the last section, the results obtained from the diagnostic application have been compared with similar or connected studies in the literature and an evaluation has been made and some prospective suggestions have been made in the light of this study.