Infarct Localization From Myocardial Deformation: Prediction and Uncertainty Quantification by Regression From a Low-Dimensional Space

Abstract

Diagnosing and localizing myocardial infarct is crucial for early patient management and therapy planning. We propose a new method for predicting the location of myocardial infarct from local wall deformation, which has value for risk stratification from routine examinations such as (3D) echocardiography. The pipeline combines non-linear dimensionality reduction of deformation patterns and two multi-scale kernel regressions. Confidence in the diagnosis is assessed by a map of local uncertainties, which integrates plausible infarct locations generated from the space of reduced dimensionality. These concepts were tested on 500 synthetic cases generated from a realistic cardiac electromechanical model, and 108 pairs of 3D echocardiographic sequences and delayed-enhancement magnetic resonance images from real cases. Infarct prediction is made at a spatial resolution around 4 mm, more than 10 times smaller than the current diagnosis, made regionally. Our method is accurate, and significantly outperforms the clinically-used thresholding of the deformation patterns (on real data: sensitivity/specificity of 0.828/0.804, area under the curve: 0.909 versus 0.742 for the most predictive strain component). Uncertainty adds value to refine the diagnosis and eventually re-examine suspicious cases.

Model-Based Feature Augmentation for Cardiac Ablation Target Learning From Images

“…One way to achieve this is to use synthetic data with realistic characteristics similar to what was performed in [48]. These samples could include realistic myocardial infarctions with different grey zone to scar core ratios, sizes, shapes and locations and intracardiac electrograms could be computed from them.…”

Transfer Learning From Simulations on a Reference Anatomy for ECGI in Personalized Cardiac Resynchronization Therapy

“…Third, in order to capture the conductivity heterogeneity we modelled scar tissue as having no reaction term in the Mitchell-Schaeffer model and a diffusivity reduction of 80%. A varying scar location on the LV with a random and realistic shape [26] was added in 50% of the simulations.…”

Strain-Based Parameters for Infarct Localization: Evaluation via a Learning Algorithm on a Synthetic Database of Pathological Hearts

“…The aligned strain parameters and infarct location were treated as column vectors of length equal to the number of elements for the reference mesh. The link between each strain parameters and infarct location was learned via kernel ridge regression, inspired by the algorithm described in [4]. Direct regression was preferred over going through an intermediate space of reduced dimensionality, whose purpose is mainly for uncertainty modeling without substantially a↵ecting the localization performance.…”

“…However, due to the complex strain patterns, simple thresholding is not su cient and further processing is required [12]. Learning-based algorithms have been investigated for better detecting [9] and locating [2,4] infarcts from strain data. These methods are promising and can be generalized to di↵erent modalities and subjects, in addition to showing high accuracy of infarct detection.…”

“…They were assessed in terms of their localization performance (sensitivity and specificity) given a learning algorithm to locate the infarcts. The algorithm uses a regression to find the transfer function between the tested parameters and the infarct location [4]. To do so, we designed a large database of synthetically-generated pathological cases that incorporates infarcts with di↵erent locations, shapes, sizes, and degrees of severity.…”