Automatic scoliosis Cobb angle estimation algorithm based on multi-scale attention
Objective Adolescent idiopathic scoliosis (AIS) is a common condition that poses a significant health risk to teenagers.The Cobb angle measurement method on X-ray images is considered a"gold standard"for assessing the severity of spinal curvature in patients.However,manual measurement of Cobb angles is complex,time-consuming,and prone to errors due to factors such as overlapping rib shadows,pelvic shadows,and morphological variations of vertebrae.Therefore,the development of fast and accurate automatic methods for Cobb angle measurement holds great clinical value.Existing segmentation-based approaches are sensitive to image quality,while keypoint detection methods often suffer from inaccurate localization due to their focus on local feature extraction.To address these limitations,this study proposes a vertebra-centered landmark detection method for automatic estimation of Cobb angles in spinal deformities.Methods A multi-scale attention M-shaped network (MSAM-Net) is constructed to detect vertebral landmarks.The network incorporates the multi-scale pyramids squeeze attention (MPSA) module and the attentional feature fusion (AFF) module to extract vertebra features and contextual information.By leveraging the vertebra center and angular offsets,the network can accurately locate the four corner landmarks,enabling the estimation of Cobb angles for the proximal thoracic,main thoracic,and thoracolumbar curves in spinal deformity assessment.Results To evaluate the accuracy of Cobb angle estimation,the proposed algorithm is assessed by using four metrics:symmetric mean absolute percentage error (SMAPE),Euclidean distance (ED),Manhattan distance (MD),and Chebyshev distance (CD) .The evaluation is conducted on the AASCE MICCAI 2019 challenge dataset.The results demonstrate SMAPE of 9.39％,ED of 4.18,MD of 5.92,and CD of 5.34.Comparative analysis of MSAM-Net with five deep learning methods based on segmentation and detection shows superior performance in Cobb angle measurement.Conclusions The experimental results highlight the capability of the proposed approach to accurately identify and locate vertebrae in X-ray images,facilitating Cobb angle measurement for AIS patients.This method has potential implications in clinical diagnosis,surgical planning,and spinal health analysis for AIS.
万方数据
维普
