A Computer Vision Approach for Identifying the Manufacturer of Posterior Thoracolumbar Instrumentation Systems
Background
In the thoracolumbar spine, pedicle screw fixation is the most widespread technique to achieve spinal fusion and stabilization. Despite pedicle screws being used as clinical best practice, screw loosening and breakage are recurring mechanical issues of spinal fixation that lead to revision surgery in about 6% of cases. Additional reasons for revision surgery are disc herniation, scar tissue, hardware issues, and bone fragments. During revision surgery, previous hardware is often removed and replaced (or extended) to allow the new bone graft to heal properly or to extend the fusion construct.
Information about previous hardware is often unavailable due to patients being referred by other centers or missing information in the patient's records. A surgeon or other expert medical professional can sometimes decipher which fixation component was used from AP/LAT radiographs of the levels of interest but doing so can take significant time and is still subject to error.
Technology Overview
Researchers at Baylor College of Medicine have developed a new and innovative machine learning-based system for classifying implanted thoracolumbar pedicle screws from DICOM images (e.g. radiograph) of a patient in which the thoracolumbar pedicle screws are implanted. The classification system provides information regarding the pedicle screw including manufacturer and provides that information for a surgeon to use prior to a revision surgery. This process improves identification accuracy and speed compared to manual image review. Input for the novel classifier is simply AP/LAT x‑ray data. And the output is the manufacturer of the hardware.
Development status
436 images were isolated, containing five different manufacturers. When classifying binarily between the two most common manufacturers used at BCM, accuracy was 91.08% ± 5.30% (mean ± standard deviation). Classification accuracy was 81.98% ± 4.80%, 72.86% ± 5.51%, and 65.90% ± 5.14% for 3-, 4-, and 5-way classification, respectively.
Benefits
- An automated computer vision system that can detect features on DICOM images
- Able to identify manufacturers of previously implanted screw and rod systems
- Classification accuracy is over 85% when comparing the top two manufacturers
- Classification accuracy is over 73% in three-way classification
- Nearly instantaneous results
- Ability for automated improvement with continued use due to machine learning
- Provides meaningful knowledge prior to performing revision surgery
Applications
Identification of the manufacturer of previously implanted pedicle screws and rod systems using an automated computer vision system