Automated spinopelvic measurements on radiographs with artificial intelligence: a multi-reader study

Area of Science:

• Spine surgery
• Radiology
• Artificial intelligence

Background:

• Spinopelvic parameters are crucial for assessing spinal alignment.
• Manual measurement of these parameters can be time-consuming and subject to inter-observer variability.

Purpose of the Study:

• To develop and validate an AI algorithm for automated measurement of spinopelvic parameters.
• To compare the AI's performance against experienced radiologists and surgeons.

Main Methods:

• A region-based convolutional neural network (R-CNN) was trained on 295 lateral full-spine radiographs.
• The AI algorithm measured thoracic kyphosis (TK), lumbar lordosis (LL), sacral slope (SS), and sagittal vertical axis (SVA).
• Performance was evaluated against measurements from 6 human readers on 65 independent radiographs using ICC, MAE, and ANOVA.

Main Results:

• The AI achieved excellent correlation with the reference standard (ICCs within readers' range).
• Mean absolute errors (MAE) for the AI were comparable to human readers for all parameters.
• ANOVA revealed no significant differences in measurement errors between the AI and human readers (p > 0.05).
• Human reading time averaged 139 seconds per case, highlighting AI's potential time savings.

Conclusions:

• The developed AI algorithm provides accurate spinopelvic measurements within the variability of experienced clinicians.
• The AI offers potential benefits in terms of time efficiency and improved measurement reproducibility.