Detection of knee osteoarthritis with thermal image processing

In people with osteoarthritis (OA) disease, the temperature in the knee area where osteoarthritis is present is higher than in normal people. In our study, an examination will be carried out on the early diagnosis of the disease with images obtained by thermography by taking advantage of the temperature feature of OA disease. In this study, CNN, Support Vector Machines and VGG-16 architecture will be used as a method. Our study aims to design a disease diagnosis system using thermal imaging that can assist the doctor in the diagnostic process as it provides flexible system and effective tools. "Phyton" will be used as the programming language in the design and programming of the proposed system. When we load the image into the designed interface program, by clicking the diagnostic button, the program takes the guesswork out of diagnosing the disease. It is aimed to find the method that can predict the disease with the highest accuracy by applying these methods to the images obtained by thermography. In this study, a total of 998 images were obtained from different people using the FLIR E45 type thermal camera. Of these thermal images, 284 are patient images and 714 are healthy images. In the study, deep learning and machine learning algorithms libraries were used. While the color difference in the images taken with thermography cannot reveal whether there is Osteoarthritis disease on its own, it is possible to detect this disease with the help of the methods mentioned above. Thousands of images are required to train deep learning methods. However, since it would take a long time to create such a dataset in the medical environment, image enhancement methods were used. Among the applied methods, the best classification result was achieved in convolutional neural networks method and using image augmentation with 90% accuracy. The obtained results reveal that deep learning methods are very successful in classifying thermographic images.