Scoliosis is a three-dimensional deformity characterized by lateral curvature of the spine in the coronal plane, and it is often accompanied by axial rotation and abnormalities in the sagittal plane. Idiopathic scoliosis and non-idiopathic scoliosis are two major groups of scoliosis, with adolescent idiopathic scoliosis (AIS) being a prevalent form among all idiopathic types, affecting between 0.47% to 5.2% of the population [1]. For patients with AIS, wearing braces significantly reduces the high-risk curve of progression to the surgical threshold [2]. Therefore, early detection of scoliosis and timely intervention is essential. In clinical diagnosis, the Cobb angle is the criterion used to evaluate scoliosis [3]. However, the current diagnostic method of using X-ray films to assess scoliosis has some drawbacks, as prolonged exposure to radiation can severely compromise patients’ health. In a study, it was shown that breast cancer mortality was significantly higher in women with scoliosis and other spinal disorders who underwent frequent diagnostic X-ray examinations [4]. In addition to X-ray, MRI can also be employed to image the spine and measure the Cobb angle for spinal curvature [5]. However, MRI scans are difficult to popularize in the clinical diagnosis of scoliosis because they are time-consuming and expensive. Compared to X-ray and MRI, ultrasound is a more practical choice due to its safety, convenience, and affordability. Therefore, ultrasound methods are well-suited for clinical scenarios that require multiple examinations of patients with scoliosis, and many scholars have started to use ultrasound methods for diagnosing scoliosis. Cheung et al. [6] developed a 3-D ultrasound imaging system for the radiation-free assessment of AIS, which allows manual positioning of vertebrae on a series of B-mode ultrasound images to form a spine model. Zhou et al. [7] proposed an automated measurement of spine curvature by using prior knowledge on vertebral anatomical structures in ultrasound volume projection imaging (VPI), which can be used in scoliosis assessment with free-hand 3-D ultrasound imaging. Huang et al. [8] proposed an improved 2.5-D extended field-of-view imaging technique based on the third-order Bezier interpolation for scoliosis imaging.

In recent years, with the continuous development and maturity of deep learning, it has been widely applied in medical image processing [9], [10], [11], [12], [13], [14]. Many scholars have started to use deep learning methods to recognize spine regions in ultrasound images. Ungi et al. [15] performed image segmentation on acquired ultrasound images of the spine, and they used the segmentation results of these transverse processes to visualize the spine model and then measured the degree of scoliosis. Huang et al. [16] used a target detection method to determine the location of the spinous processes and transverse processes, and they then calculated the spatial location of the vertebral landmarks after clustering the detection results, which in turn modeled the vertebrae to form the entire spine and calculated the Cobb angle. Jiang et al. [17] designed an ultrasound global guided block network (UGBNet) to better perform the task of segmenting ultrasound images of the spine. Inspired by previous studies, we realized that accurately identifying the spinal regions in ultrasound images is crucial, because it directly affects the reconstruction of the spine. However, it is challenging to accurately identify spine regions in images due to relatively small target areas and the presence of a lot of interfering information. In the field of medical image processing, it is a trend to incorporate prior knowledge to address the difficult segmentation of target regions [18], [19], [20]. Therefore, in this research, we explore the use of deep learning methods incorporating prior knowledge to identify the key regions of the spine in ultrasound images, in order to better reconstruct spine models and measure the Cobb angle, thereby better completing the tasks of diagnosing scoliosis.

Statement of significance