Designing a decision support system on the sacrum bone using machine learning methods

Caudal epidural injection is the most well-known and frequently applied anesthesia technique for dealing with orthopedic injuries, fractures and chronic back pain, generally in the sacro-iliac region. Classification of the sacral hiatus (SH) is a very important and challenging task due to variences in its shape and size. It is clinically important in traumas where surgeons need to make quick and accurate decisions. Past studies have focused on morphometric and statistical analyzes for SH classification. Therefore, accurate and rapid autonomous classification of SH types using machine learning methods is vital. For this purpose, in the thesis titled Designing a Decision Support System on the Sacrum Bone using Machine Learning Methods, a new SH classification approach, Multi-Stage Process (MSP), was proposed. Initially; a small medical tabular dataset was obtained from computed tomography scans of the sacrum by manual feature extraction. In the next stage, this dataset was synthetically augmented through the Generative Adversarial Network (GAN) and applied to traditional machine learning classifiers. In the third stage of the MSP approach; a two-dimensional (2D) embedding algorithm was applied to transform, synthetically augmented tabular features into images. Finally; these images were applied to pre-trained deep Convolutional Neural Networks (CNN). As a result of applying the MSP approach to six different CNN models, significant classification success rates of approximately 90% to 93% were achieved. Recommended MSP approach; in addition to the many innovations it has brought to the field of medicine, especially in bone classification, it has also provided a solution to the problem of insufficient data set encountered in deep models.