Using AI to measure severity of degenerative scoliosis

In our aging society, degenerative spinal diseases are getting more and more prevalent. One of these diseases is degenerative scoliosis which is a pathological curvature of the lower spine caused by degeneration. This can lead to severe debilitating complaints such as chronic low back pain or radiating pain to the legs. To monitor the severity and progression of scoliosis, the Cobb angle is used, which is the largest angle between two vertebrae, traditionally measured on conventional AP radiographs. However, MR images are often also available for these patients and measuring the Cobb angle on these images can be useful, and possibly makes the use of additional radiograph images redundant. Besides, an automatic method for performing these measurements could reduce workload and subjectivity significantly.

Jasper van der Graaf and his promotion team created an AI algorithm which automatically measures the coronal Cobb angle based on standard sagittal MR images. A 3D deep learning neural network, trained on 477 MRI volumes, was utilized to automatically segment al vertebrae and intervertebral discs. (Method shown here.) These segmentation masks were used to calculate all possible angles in 3D between the vertebrae in the scan. The largest coronal angle found by the algorithm is equal to the Cobb angle. Validation of this method was done by comparing its measurements to those of two radiologists and one spinal surgeon, on 50 lumbar MRI scans of degenerative scoliosis patients. These manual measurements were performed on a flattened image of the vertebrae segmentations, mimicking a standard AP radiograph.

The results demonstrated strong agreement between the algorithm and human readers, with the algorithm consistently measuring the maximum Cobb angle. This research also showed that the three readers often measured different angles, since they choose different vertebral levels to perform the measurements on. Only in 9 out of 50 cases all three readers selected the same vertebral levels for their measurements, showing the inconsistency of manual measurements. Since the algorithm calculates all possible angles, and automatically extracts the largest angle, it will always choose the correct vertebral levels.

This novel AI driven algorithm can automatically measure the Cobb angle in lumbar MRI with similar accuracy as human readers. Moreover, it ensures that always the maximum angle is reliably measured. The use of such an automatic Cobb angle measurement tool will improve the accuracy of Cobb angle measurements and provides standardized extra information on top of spine MRI. Also the use of MRI could make radiographic imaging, with its related radiation dose, more redundant.

This study is published in European Radiology

van der Graaf, J. W., van Hooff, M. L., van Ginneken, B., Huisman, M., Rutten, M., Lamers, D., ... & de Kleuver, M. (2024). Development and validation of AI-based automatic measurement of coronal Cobb angles in degenerative scoliosis using sagittal lumbar MRI. European Radiology, 1-10.

https://doi.org/10.1007/s00330-024-10616-8