Performance evaluation of an AI-based preoperative planning software application for automatic selection of pedicle screws based on computed tomography images

Abstract

Introduction: Recent neurosurgical applications based on artificial intelligence (AI) have demonstrated its potential in surgical planning and anatomical measurement. We aimed to evaluate the performance of an AI planning software application on screw length/diameter selection and insertion accuracy in comparison with freehand surgery. Methods: A total of 45 patients with 208 pedicle screw placements on thoracolumbar segments were included in this analysis. The novel AI planning software was developed based on a deep learning model. AI-based pedicle screw placements were selected on the basis of preoperative computed tomography (CT) data, and freehand surgery screw placements were observed based on postoperative CT data. The performance of AI pedicle screw placements was evaluated on the components of screw length, diameter, and Gertzbein grade in comparison with the results achieved by freehand surgery. Results: Among 208 pedicle screw placements, the average screw length/diameters selected by the AI model and used in freehand surgery were 48.65 ± 5.99 mm/7.39 ± 0.42 mm and 44.78 ± 2.99 mm/6.1 ± 0.27 mm, respectively. Among AI screw placements, 85.1% were classified as Gertzbein Grade A (no cortical pedicle breach); among free-hand surgery placements, 64.9% were classified as Gertzbein Grade A. Conclusion: The novel AI planning software application could provide an accessible and safe pedicle screw placement strategy in comparison with traditional freehand pedicle screw placement strategies. The choices of pedicle screw dimensional parameters made by the model, including length and diameter, may provide potential inspiration for real clinical discretion.