KI (Künstliche Intelligenz) - unterstützte Segmentierung und Analyse von bildgebenden Datensätzen zur Identifikation und Überprüfung von Pedikelschraubentrajektorien

In the past decade, spine surgery has seen significant innovations in both hardware and software, profoundly influencing surgical techniques. Recently, Brainlab launched its Spine Planning Software, which incorporates automatic registration and suggests pedicle screw trajectories. This report evaluates the reliability and impact of this software on instrumentation surgery. In the preoperative phase, we assess Brainlab’s registration-based software (Munich, Germany), which was tested during a two-day workshop involving four neurosurgeons. This model, while not based on artificial intelligence (AI), utilizes an elastic image fusion technique for registration. Notably, the software accelerates screw planning by approximately seven seconds per screw and achieves a clinical overall accuracy of 92.9% for automatic screw trajectory planning. In comparison, we developed an AI-based system capable of automatically and safely recognizing pedicle anatomy and executing precise screw planning within seconds. The screws were inserted into the pedicles under controlled conditions using the freely available 3D Slicer software, with individual adjustments made for screw trajectories, lengths, and diameters. Our algorithm can place ideal pedicle screws in under five seconds, achieving an accuracy of 88.9% with preoperative and intraoperative CT scans. In the intraoperative environment, we are working on a model designed to remove implants from X-ray images and replace them with anatomical textures. This approach allows the images to be utilized for 2D-3D registration without any loss of information. We have trained this model using a substantial dataset of simulated X-ray images to eliminate artifacts. Surgical instrument images (such as hooks and clamps) were utilized to enhance the realism of the simulations. Our model has significantly improved the similarity between the in-painted images and the ground truth images, with a notable enhancement in the capture range of up to 85%. The aforementioned software tools effectively reduce the workload of spine surgeons by facilitating better analysis of medical imaging and providing a system for the automated identification of vertebrae and suggested pedicle screw trajectories, even when dealing with lower-quality medical images. However, it is essential to note that such software does not replace the expertise of an experienced surgeon. Additionally, the software serves as an excellent educational tool for less experienced spine surgeons.