Detection of spine deformity deep from X-ray images with deep learning

As spinal deformities reach advanced levels, they reduce the patient's quality of life and make diagnosis and treatment difficult. Therefore, early diagnosis is very important in order to stop the progression of the deformity and start the treatment process on time. The diagnostic procedure for deformities is based on manual measurements made by specialist doctors on medical images. Manual measurements are error-prone, time-consuming and they rely on subjective opinion. In this study, a deep learning-based automatic diagnosis method is presented to eliminate these disadvantages in the diagnosis of spinal deformities. In the light of literature research, it has been observed that deep learning-based convolutional neural networks produce high-accuracy results in image processing and segmentation processes. In this study, the X-ray image classification performances of ResNet and GoogleNet architectures, which are convolutional neural network architectures reported to have high accuracy and performance, will be compared in terms of accuracy, sensitivity, precision, F1 score, specificity, training and testing times. As a result of the experiments conducted within the scope of this study, it was seen that ResNet showed superior performance than GoogleNet in terms of accuracy, precision, F1 score, specificity and sensitivity, and when a comparison was made in terms of training and testing times, it was seen that GoogleNet processed faster than ResNet.