Deep Learning Algorithm for Identifying Cervical Cord Compression Due to Degenerative Canal Stenosis on Radiography

Area of Science:

• Radiology and Artificial Intelligence
• Spinal Imaging Diagnostics
• Deep Learning in Medicine

Background:

• Degenerative cervical myelopathy diagnosis is often delayed, leading to suboptimal patient management.
• Screening tools are needed to identify patients requiring further physical evaluation for suspected cervical stenosis.

Purpose of the Study:

• To validate the diagnostic accuracy of a deep-learning algorithm for detecting cervical cord compression caused by degenerative canal stenosis on radiography.
• To compare the algorithm's diagnostic performance against that of spine physicians.

Main Methods:

• A deep-learning algorithm (EfficientNetB2) was developed using radiography and MRI data from 240 patients.
• The algorithm was trained on 198 patients and tested on 42, identifying suspected cervical stenosis levels.
• Diagnostic accuracy and AUC were calculated; algorithm performance was compared to 10 physicians.

Main Results:

• The deep-learning algorithm achieved a diagnostic accuracy of 0.81 and an AUC of 0.81.
• The algorithm's correct diagnosis rate (81.0%) was significantly higher than the physician consensus (66.2%; P = 0.034).
• The algorithm demonstrated superior accuracy compared to individual physicians.

Conclusions:

• A deep-learning algorithm was successfully developed to detect cervical spinal cord compression on radiography.
• The algorithm can highlight suspected levels of compression on radiographic images.
• The developed algorithm exhibits greater diagnostic accuracy than spine physicians for this condition.