Deep learning automates Cobb angle measurement compared with multi-expert observers

TL;DR

This work introduces a fully automated pipeline for Cobb angle measurement on 2D spine radiographs that leverages whole-spine geometry via centerline fitting instead of vertebra-by-vertebra segmentation. Using Mask-RCNN for spine ROI detection, centerline-based curvature analysis, and a tolerance-based angle computation, the method provides on-image visualizations and aims for high agreement with multiple expert readers. In a multi-reader study, the algorithm achieved a mean deviation of $4.17^\u00b0$ and ICC $>0.96$ with Pearson $r>0.944$, surpassing typical inter-reader variability, and demonstrated strong performance in both Cobb angle measurement and scoliosis severity classification (kappa ~0.65, accuracy ~0.85, F1 ~0.96). The approach reduces annotation costs, enhances interpretability, and shows promise for clinical deployment to improve consistency and efficiency in scoliosis assessment.

Abstract

Scoliosis, a prevalent condition characterized by abnormal spinal curvature leading to deformity, requires precise assessment methods for effective diagnosis and management. The Cobb angle is a widely used scoliosis quantification method that measures the degree of curvature between the tilted vertebrae. Yet, manual measuring of Cobb angles is time-consuming and labor-intensive, fraught with significant interobserver and intraobserver variability. To address these challenges and the lack of interpretability found in certain existing automated methods, we have created fully automated software that not only precisely measures the Cobb angle but also provides clear visualizations of these measurements. This software integrates deep neural network-based spine region detection and segmentation, spine centerline identification, pinpointing the most significantly tilted vertebrae, and direct visualization of Cobb angles on the original images. Upon comparison with the assessments of 7 expert readers, our algorithm exhibited a mean deviation in Cobb angle measurements of 4.17 degrees, notably surpassing the manual approach's average intra-reader discrepancy of 5.16 degrees. The algorithm also achieved intra-class correlation coefficients (ICC) exceeding 0.96 and Pearson correlation coefficients above 0.944, reflecting robust agreement with expert assessments and superior measurement reliability. Through the comprehensive reader study and statistical analysis, we believe this algorithm not only ensures a higher consensus with expert readers but also enhances interpretability and reproducibility during assessments. It holds significant promise for clinical application, potentially aiding physicians in more accurate scoliosis assessment and diagnosis, thereby improving patient care.

Figures (5)

• Figure 1: Visualization of our proposed pipeline. Our approach consists of three steps: 1) the identification of ROI and segmentation of the spine; 2) Fit the spine's central curve and identify the vertebrae that are most 'tilted' 3) Calculate the Cobb angle; all Cobb angles are presented on the image, with the main Cobb angle indicated in red.
• Figure 2: The first three examples demonstrate that there is some offset between the vertebral orientation and the spine's tangent (shown by the orange arrow). After adding the tolerance range to average the curvature, we could measure the Cobb angles.
• Figure 3: Measured examples of Cobb angles. They are automatically shown as the final outputs of our method, without any further manual sketching. The main Cobb angle is displayed in red lettering, while additional Cobb angles are displayed in green.
• Figure 4: Quantitative evaluation of our algorithm 1) The first column is the measured Cobb angle difference between the average and the median of the readers and our algorithm (blue and green bars), the average pairwise difference among readers (orange bars), and the average difference between readers and our algorithm (gray bars) on the 81 test cases. 2) The second row depicts histograms that elucidate the aforementioned Cobb Angle (CA) discrepancies across the test samples, and 3) Correspondingly, the third row offers a visualization of instances where the CA differential exceeds 5 degrees.
• Figure 5: The top left is the pairwise mean absolute difference between the readers and DL, and the top right column is Pearson's coefficient; the bottom left is the intraclass coefficient (ICC) for Cobb angle measurements. The last is the Cohen kappa score for scoliosis severity classification, respectively.