Magnetic Marker Localization With Spurious Field Rejection

A neural network model serves as an intermediary between the magnetic sensor measurements and the final localization result. The neural network processes the raw magnetic field data, learns to distinguish between marker signals and instrument interference during training, and outputs corrected localization information that compensates for the harmful magnetic fields from surgical instruments.

The patent replaces traditional signal filtering and noise reduction techniques with a neural network-based computational approach. Instead of using physical or mathematical filters to separate signal from noise, the system uses deep learning models that have been trained to recognize and reject spurious signals from metal instruments while maintaining accurate marker localization.

The neural network model is trained in advance using simulated data that includes various combinations of marker signals and instrument interference patterns. This preliminary training allows the model to learn the characteristics of valid marker signals and distinguish them from spurious instrument signals before being deployed in the surgical environment, reducing the need for complex real-time processing.

The system creates a computational copy of the magnetic field generation process through the neural network model. The model replicates the relationship between marker positions, instrument positions, and resulting magnetic field patterns, allowing it to predict what the magnetic field should look like given the current sensor readings and compare this against actual measurements to detect anomalies.