Seg4Reg+: Consistency Learning Between Spine Segmentation and Cobb Angle Regression

Automated methods for Cobb angle estimation are of high demand scoliosis assessment. Existing typically calculate the from landmark estimation, or simply combine low-level task (e.g., detection and spine segmentation) with regression task, without fully exploring benefits each other. In this study, we propose a novel multi-task framework, named Seg4Reg+, which jointly optimizes segmentation networks. We thoroughly investigate both local global consistency knowledge transfer between Specifically, an attention regularization module leveraging class activation maps (CAMs) image-segmentation pairs to discover additional supervision in network, CAMs can serve as region-of-interest enhancement gate facilitate turn. Meanwhile, design triangle learning train two networks optimization. The evaluations performed on public AASCE Challenge dataset demonstrate effectiveness superior performance our model state-of-the-art methods.