Vertebral Deformity Measurements at MRI, CT, and Radiography Using Deep Learning

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Spine Biomechanics

Background:

• Accurate measurement of vertebral body heights and spine angles is crucial for diagnosing spinal deformities.
• Manual landmark identification and measurement are time-consuming and prone to inter-observer variability.
• Developing automated tools can improve efficiency and consistency in spinal analysis.

Purpose of the Study:

• To develop and validate a deep learning system for automated localization of vertebral landmarks.
• To assess the system's efficacy in measuring vertebral body heights and lumbar lordosis angles (LLAs).
• To evaluate performance across multiple imaging modalities including MRI, CT, and radiography.

Main Methods:

• A retrospective study utilized 1744 images (MRI, CT, radiography) from diverse patient populations.
• A neural network model was trained using annotated images for landmark identification and measurement.
• The model's performance was evaluated on a holdout testing dataset for accuracy and speed.

Main Results:

• The deep learning system achieved high accuracy in measuring vertebral heights with mean errors ranging from 1.5% to 1.9% across modalities.
• Accurate lumbar lordosis angle (LLA) measurements were obtained with mean absolute errors between 2.26° and 3.60°.
• The system provided measurements in under 1.7 seconds per study, demonstrating rapid analysis.

Conclusions:

• The developed deep learning network enables rapid and accurate measurement of vertebral morphometrics and LLAs.
• The system's multi-modal capability makes it a versatile tool for spine analysis.
• This automated approach holds significant potential for computer-aided diagnosis in spinal imaging.