MRI is a common tool for radiographic diagnosis of spinal stenosis, but it is expensive and requires long scanning time. CT is also a useful tool to diagnose spinal stenosis, yet interpretation can be time-consuming with high inter-reader variability even among the most specialized radiologists. In this study, the investigators aim to develop a deep-learning algorithm to automatically detect and classify lumbar spinal stenosis.

MRI is a common tool for radiographic diagnosis of spinal stenosis, but it is expensive and requires long scanning time. CT is also a useful tool to diagnose spinal stenosis, yet interpretation can be time-consuming with high inter-reader variability even among the most specialized radiologists. In this study, the investigators aim to develop a deep-learning algorithm to automatically detect and classify lumbar spinal stenosis. It would be a time-saving workflow if the software can assist the radiologists to detect and locate the suspected lesion.

Inclusion Criteria:

• Age >18 years
• with radiologists' CT reports on cervical, thoracic and lumbar stenosis

Exclusion Criteria:

• not applicable (only specific levels with extensive infections, fractures, tumor, high-grade spondylolisthesis would be excluded for analysis).

Spinal stenosis is a narrowing of the spaces within the spine, which can put pressure on the nerves that travel through the spine. Spinal stenosis occurs most often in the back, the neck, and sometimes the thoracic spine.

Some people with spinal stenosis may not have symptoms. Others may experience pain, tingling, numbness and muscle weakness. Symptoms can worsen over time.