Detecting vertebra landmarks from ultrasound image using single shot multibox detector

In the diagnosis of scoliosis, the position and geometric parameters of the vertebra or spine are important to doctors' diagnoses. Many imaging techniques such as X-ray and Magnetic Resonance Imaging (MRI) can be used for scoliosis detection. Ultrasound imaging (US) is a radiation-free and low-cost way on clinical application in contrast to other imaging techniques. Many methods on Ultrasound imaging were reported to assess the severity of scoliosis. In this paper, we employ Single Shot MutilBox Detector, an end-to-end object detection algorithm based on deep learning, on detecting the vertebra landmarks with ultrasound image. The automatic detection and location of vertebra landmarks is important basis for further analysis and diagnosis of scoliosis and contributes to development on computer aided diagnosis system. The preliminary experiment results on phantom show that our method is high accuracy and feasible in detecting vertebra landmarks.