Assessing Cardiac Functionality via Interpretable AI

A fully interpretable, rule-based machine learning model using myocardial strain data to accurately estimate Left Ventricular Ejection Fraction (LVEF).

Cardiac Imaging is a powerful methodology for assessing heart health, but translating raw imaging data into actionable clinical insights requires robust—and transparent—predictive models.

The most important metric for evaluating heart function is the Left Ventricular Ejection Fraction (LVEF), which measures the percentage of oxygen-rich blood pumped out of the left ventricle with each heartbeat (typically 55% to 70% in a healthy individual). By tracking the movement and deformation of the myocardium during the cardiac cycle (Myocardial Strain) via Cardiac Magnetic Resonance, we developed a predictive machine learning model to estimate LVEF.

Crucially, to ensure clinical trust and applicability, we moved away from "black-box" neural networks. Instead, we engineered a fully interpretable model based on a rule-based Fuzzy Inference System, coupled with a novel methodology (RuDe) to disambiguate the rules, providing physicians with clear, readable insights into how strain features affect cardiac functionality.

Interpretable AI Cardiac Imaging Fuzzy Logic

Methodological Approach

Estimating LVEF accurately is critical, but understanding why a model predicts a certain value is just as important for medical professionals. Our approach tackles this by bridging high-accuracy machine learning with human-readable logic:

  • Myocardial Strain Analysis: We utilize high-fidelity strain data, which captures the mechanical deformation of the heart muscle, offering deeper insights than simple volumetric measurements.
  • Fuzzy Inference Systems (FIS): Instead of hidden layers, the model’s core relies on fuzzy logic rules that map directly to human reasoning (e.g., “IF strain is X AND volume is Y, THEN LVEF is Z”).
  • Rule Disambiguation (RuDe): To prevent overlapping or contradictory rules from confusing the output, we introduced a novel disambiguation methodology that streamlines the rule base, ensuring the model remains strictly interpretable without sacrificing predictive power.

Research Highlights

Transparent AI

Replaces opaque algorithms with a transparent Fuzzy Inference System, allowing clinicians to trace exact decision pathways.

Strain-Based LVEF

Accurately predicts Left Ventricular Ejection Fraction purely from the physical deformation metrics of the myocardium.

RuDe Methodology

Introduces a novel approach to automatically clean and disambiguate logical rules, improving both model accuracy and readability.

Request Information

Core Publication

2025

  1. Assessing Cardiac Functionality by Means of Interpretable AI and Myocardial Strain
    Marco S Nobile, Amalia Lupi, Leone Bacciu, and 4 more authors
    In 2025 IEEE Conference on Computational Intelligence in Bioinformatics and Computational Biology (CIBCB), 2025
    DOI