Ensity-based bone segmentation and osteoporosisassessment in radiological images

In this master's thesis, an artificial intelligence-based diagnostic system was developed for the early detection of osteoporosis. Osteoporosis is a systemic bone disease characterized by decreased bone mineral density, leading to increased bone fragility, and is particularly associated with a high risk of fractures in elderly individuals. Considering the limitations of conventional diagnostic methods—such as high cost, limited accessibility, and insufficient sensitivity—this study aims to provide a fast, automated, and reliable alternative. Radiographic knee images from the Knee Osteoarthritis Severity Dataset were used in the study. Bone structures were successfully segmented using a U-Net deep learning architecture. Following segmentation, morphological, statistical, and frequency-based features were extracted, and classification was performed using the Random Forest algorithm. The model's performance was evaluated through multiple metrics, including accuracy, ROC-AUC, sensitivity, specificity, precision, and F1-score. Results showed a high AUC value of 95.6% for the advanced osteoporosis class and an overall classification accuracy of 75%. These findings demonstrate that the proposed system has the potential to support early diagnosis and can be integrated into clinical decision support systems.