A CNN-Based Thoracic Vertebrae and Rib Localization Method

In the past few decades, artificial intelligence has continued to develop and has been increasingly used in medical care. Using medical images to assist doctors in diagnosing the disease and reducing the subjective errors of doctors, the computer-aided diagnosis method based on artificial intelligence is worth trying. Bone metastases often appear in the thoracic vertebrae and rib areas. Doctors determine the specific location of bone metastases mainly by taking the twelve thoracic vertebrae and their connected ribs as reference objects, and locating the thoracic vertebrae and ribs with bone metastases according to Computed Tomography (CT) scan images, so as to identify the specific location of the bone metastases. This paper proposes a thoracic vertebra and rib location method based on ResNext and YOLO V4 network models, by classifying thoracic vertebrae and rib detection combined with the characteristics of CT scan images to determine the specific positions of thoracic vertebrae and ribs on CT scan images, thereby helping doctors to reduce the heavy work of localization in problems such as diagnosing bone metastases. Our work is mainly divided into three parts: (1). Data set production and image preprocessing. (2). Classify the thoracic vertebrae using the ResNext model. (3). Locate the ribs using the YOLO V4 model. The results of the experiment were tested on 856 and 100 CT scan images collected, respectively. The accuracy, sensitivity, specificity and precision of the proposed model in thoracic vertebra classification were 94.86%, 82.79%, 98.91% and 96.22%. The Average Precision (AP), Mean Average Precision (mAP) and Average Recall (AR) rate of rib detection were 85.22%, 83.16% and 85.22%. The experiment proves that the method base on deep learning can be used as a computer aided diagnosis method to effectively reduce the workload of thoracic vertebra and rib localization.