Deep Learning-Based Identification of Surgical Candidacy for Cervical Spinal Cord Decompression.

Area of Science:

• Artificial intelligence in medical imaging
• Deep learning for clinical decision support
• Spine imaging analysis

Background:

• Artificial intelligence (AI) has shown promise in interpreting cervical spine imaging.
• Triage of patients into operative and nonoperative groups is complex due to background findings.

Purpose of the Study:

• To determine if deep learning can triage patients into operative and nonoperative groups based on cervical spine imaging.
• To develop an AI-driven biomarker for cervical stenosis.

Main Methods:

• A deep-learning algorithm was trained to segment spinal canal and cord on 100 cervical spine MRI scans.
• A biomarker for cervical stenosis was generated based on the cross-sectional area difference.
• The model was tested on 147 outpatient patients evaluated by neurosurgeons.

Main Results:

• The mean minimum cross-sectional area difference was 35.90 mm² for surgical patients and 48.55 mm² for non-surgical patients (P=0.005).
• The deep learning model distinguished between surgical and non-surgical patients with an area under the curve of 0.79.

Conclusions:

• A proof-of-concept deep learning model can triage patients with cervical stenosis into surgical and non-surgical cohorts.
• This AI tool offers a supportive metric for providers considering surgical referrals.
• Deep learning can assist in the radiographic assessment for operative candidacy, potentially expediting care.