Artikel
Novel Artificial Intelligence Based Automatic Pedicle Screw Placement for Preoperative Planning: A Validation Study
Suche in Medline nach
Autoren
Veröffentlicht: | 22. Oktober 2019 |
---|
Gliederung
Text
Objectives: To analyze the accuracy of a novel neural network based software application in automatic placement of lumbar pedicle screws for preoperative planning
Methods: The lumbar computed tomography (CT) scans of 8 cadavers were analyzed with the autonomous planning software application. The software was developed by using neural networks and teaching the computer proper pedicle screw position placement on 7000 datasets of the lumbar spine. The output was assessed by an orthopedic surgeon using 3D viewing software and associated measuring tools. Accuracy of screw placements were evaluated according to Ravi pedicle screw position grading: I, no pedicle wall breach; II, breach less than 2 mm; III, breach equal to 2-4 mm; IV, breach more than 4 mm. Direction of breach, screw tip relative to vertebral body, and screw trajectory were also evaluated.
Results and conclusion: From a total of 58 virtual screws that were automatically placed in the virtual spine models of all 8 patients included in this experiment, 100% of all screw trajectories were deemed acceptable. Only 3 screws (5%) mildly breached the lateral pedicle wall, and only 1 screw superiorly breached the vertebral body with no contact to vital structures (1.7%).
Out of 58 planned pedicle screws, 54 (94%) of them were perfectly placed, and 100% of them were acceptably placed. The novel software has the potential to be combined with intraoperative navigation systems and reduce both preoperative planning and surgical times during lumbar pedicle screw placement.