Development of a hybrid machine learning method for the detection of non-traumatic spine compression fractures from computed tomography (CT) image

*Non-traumatic vertebral compression fractures are a common pathological condition in the human spine. The decrease in bone mass with age causes non-traumatic vertebral compression fractures. Since non-traumatic vertebral compression fractures usually remain asymptomatic for a long time, early diagnosis in this period prevents the disease from progressing and provides more effective treatment. Radiological images of patients are used for diagnosis and treatment. Data sets obtained from radiological images make significant contributions to scientific developments today. Data sets create effective solutions for the segmentation of pathological conditions in the field of healthcare, detection, classification with high accuracy and precision, and disease diagnosis by using machine learning methods. In recent years, machine learning methods have begun to be used more frequently in research on the detection of non-traumatic vertebral compression fractures. In this thesis study, a unique data set was created for the application of machine learning methods by processing the raw computed tomography image data of the patients. The original data set was created from retrospective images of patients who received radiological diagnosis with computed tomography at Istanbul University-Cerrahpaşa, Cerrahpaşa Faculty of Medicine, Department of Radiology. Within the scope of the thesis, image features were extracted with the VGG16 method and selected with the SVM method. A new hybrid adversarial generative Network -quantum convolutional neural network (QCNN) machine learning classification model has been developed. While creating the hybrid model, quantum features, DenseNet121 and GAN were inspired. The model was tested with the created data set. The classification results of the hybrid model on the original dataset were compared with DenseNet121, GAN, MobileNetV3, EfficientNetB1, VGG16 and ResNet50V2, and these results were discussed.