Noninvasive AI Tool Maps Atrial Cardiomyopathy to Support Ablation Planning

Atrial cardiomyopathy involves electrical and structural abnormalities in atrial tissue, including fibrosis, and is linked to atrial fibrillation. Identifying affected tissue can help clinicians understand the disease and plan procedures such as ablation. Current assessment may require invasive electroanatomical mapping or magnetic resonance imaging. Researchers have now developed an artificial intelligence approach for noninvasive assessment using electrical recordings from the body surface.

TResearchers at the Universitat Politècnica de València developed an artificial intelligence model based on graph neural networks to locate and quantify atrial tissue abnormalities associated with atrial cardiomyopathy. The system analyzes body surface potential maps obtained from electrodes placed across the torso. It evaluates how cardiac electrical signals are distributed spatially and how they change over time to identify patterns linked to the location and extent of affected atrial tissue.

The proof-of-concept study used simulated data and does not imply immediate clinical application. Researchers developed and evaluated the model using 14,400 simulated body surface potential maps generated from different atrial and torso anatomies. The simulations included varying degrees and locations of atrial cardiomyopathy, allowing assessment across diverse anatomical configurations.

In a baseline configuration using 128 electrodes, the model achieved 89% accuracy in locating affected tissue, with a mean sensitivity of 89% and specificity of 90%. It also reached 84% overall accuracy in determining the extent of affected tissue. The system maintained performance when analyzing anatomies not used during training, and its performance remained relatively stable when signal quality decreased.

The study was published in Discover Computing and involved the Institute of Information and Communications Technologies at the Universitat Politècnica de València, Corify Care S.L., and Karlsruhe Institute of Technology. The results also indicated that anatomical diversity in training data was important, as wider variation in atrial and torso models improved classification ability. The next step stated for the work is validation using records from real patients before transfer to treatment-planning applications.

Related Links
Universitat Politècnica de València