Electrode Position Mapping Using Patient-Specific ML Retraining
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current electro-anatomical mapping systems face challenges in accurately detecting magnetic and impedance signals across different subjects due to variations between subjects, leading to inaccurate impedance and cardiac calibrations, and existing solutions struggle to adjust localization models for precise sensor positioning.
Innovation Solution
A system and method using a retrained machine-learning model that combines patient-agnostic and patient-specific data to determine position signals of electrodes, incorporating a catheter with electrodes and a magnetic sensor, where the model is trained on patient-agnostic data and refined with patient-specific data to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single localization model is used for all patients, then device complexity is reduced, but measurement precision deteriorates due to patient variations
Solution Approach 1:
The localization model is segmented into two distinct components: a patient-agnostic model trained on general anatomical data from multiple patients, and a patient-specific model fine-tuned on individual patient data. This segmentation allows the system to maintain a simple base model while adding precision through targeted customization for each patient.
Solution Approach 2:
The patient-agnostic model is pre-trained on comprehensive datasets from multiple patients before deployment. This preliminary training establishes a robust baseline that captures general anatomical variations, eliminating the need to start from scratch for each new patient and reducing overall system complexity.
2Measurement precision
If patient-specific calibration is performed, then measurement precision is improved, but loss of time increases due to additional calibration steps
Solution Approach 1:
General anatomical patterns and impedance characteristics are pre-calibrated in the patient-agnostic model using data from multiple patients. This preliminary calibration captures common anatomical variations, so when a new patient is encountered, only minor patient-specific adjustments are needed rather than performing complete calibration from scratch.
Solution Approach 2:
The patient-specific model is created by fine-tuning and adapting the pre-trained patient-agnostic model rather than building a completely new model. This copying approach transfers the general knowledge already embedded in the base model, significantly reducing the time and data required for patient-specific calibration.
3Measurement precision
If more training data is collected for each patient, then measurement precision is improved, but loss of time increases due to data collection and processing
Solution Approach 1:
Instead of collecting and processing extensive new data for each patient, the system copies and adapts the knowledge already embedded in the patient-agnostic model. The pre-trained model provides a strong prior that requires minimal patient-specific data to achieve high accuracy, dramatically reducing data collection and processing requirements.
Solution Approach 2:
The system changes the approach from collecting large volumes of raw patient data to adjusting model parameters through fine-tuning. By modifying the weights and parameters of the pre-trained model based on limited patient-specific measurements, the system achieves high precision without the time cost of extensive data collection and processing.
Data Source
AI summary
A system for determining position signals of electrodes using a retrained machine-learning model includes at least a catheter including a plurality of electrodes configured to collect a plurality of potential signals and a magnetic sensor configured to collect magnetic data, and at least a computing device including a memory. The processor receives a first training set, wherein the first training set includes patient-agnostic data; receives a second training set, wherein the second training set includes patient-specific data, trains a mapping machine-learning model using the first training set, retrains the mapping machine-learning model using the second training set, receives at least a first signal, wherein the first signal includes a potential signal of the plurality of potential signal and the magnetic data, and generates, using the retrained machine-learning model, as a function of the at least a first signal, a first position signal for an electrode of the plurality of electrodes.