Multi-parameter scoliosis evaluation from dual-view x-rays via a local sine-based projection model.

We proposed a dual-view labeling protocol and employed a Residual U-Net architecture to segment vertebral bodies in 700 paired anteroposterior and lateral x-rays.

A novel Local Sine-based Spinal Projection Model was introduced to calculate vertebral inclination angles and derive consecutive CAs.

The proposed method achieved a vertebral segmentation Dice coefficient of 94.7%.

The mean absolute error of the cobb angle was, outperforming the minimum bounding rectangle method () and novice manual measurements (), with an intraclass correlation coefficient of 0.958.