TL;DR
This paper introduces a novel, lightweight method for creating statistical shape models (SSMs) of the human heart using "universal ventricular coordinates" (UVCs). This approach enables a more efficient and accurate analysis of how anatomical variations in the heart relate to electrical activity (ECG) and demographic factors, paving the way for more realistic cardiac digital twins.
What Problem Does This Paper Address?
Creating accurate digital twins of the human heart, which are virtual replicas used for research and clinical applications, requires precisely matching the model's anatomy to an individual's heart. This process, known as "anatomical twinning," is computationally intensive. While many studies have explored how anatomical variations affect simulated heart function, the impact of these variations on real-world patient measurements like ECGs remains less understood. This gap hinders the development of truly personalized and predictive cardiac digital twins. The authors aim to bridge this gap by developing a robust statistical model of healthy heart anatomy and quantifying its relationship with functional and demographic data.
Methods
The core innovation is the use of Universal Ventricular Coordinates (UVCs) to establish anatomical correspondence between different heart geometries. Instead of computationally expensive nonlinear registration, UVCs provide a standardized coordinate system (apico-basal, transmural, rotational, and chamber association) that allows for direct mapping of points. This enables the creation of lightweight surface meshes and subsequent statistical shape modeling using Principal Component Analysis (PCA).
Key methodological steps include:
- Dataset Construction: A large dataset of 271 high-resolution cardiac CT scans from healthy individuals (including athletes) was curated.
- Semantic Segmentation: A 3D U-Net-like CNN was used for automated segmentation of cardiac structures.
- Anatomical Model Generation: A specialized workflow generated volumetric biventricular meshes from the segmented data, incorporating anatomical labels and fiber orientations.
- UVC-based Correspondence: UVCs were computed for each geometry, allowing for a standardized point cloud extraction without traditional registration.
- Statistical Shape Modeling (SSM): PCA was applied to the UVC-derived point clouds to capture the main modes of anatomical variation within the population.
- Synthetic Cohort Generation: The SSM was used to generate a diverse set of synthetic biventricular meshes, suitable for downstream applications like electrophysiological simulations.
- Statistical Analysis: Correlations between the principal components (representing anatomical variations) and ECG-derived features, as well as demographic data, were investigated.
Datasets
- Training and Evaluation Dataset: A dataset of 271 high-resolution cardiac CT scans from healthy individuals, including male athletes, male healthy controls, and healthy female individuals. This dataset was derived from the Master@Heart study and a proprietary dataset from the University Hospital of Leuven.
- Synthetic Cohort: 100 ready-to-use biventricular meshes, including anatomical labels and fiber structures, generated from the developed SSM.
Key Insights and Ideas
- UVCs as a Lightweight Correspondence Method: UVCs offer a computationally efficient and effective alternative to traditional registration methods for establishing anatomical correspondence, significantly streamlining SSM construction.
- Dominant Anatomical Variations: The primary sources of anatomical variability in healthy biventricular hearts are size, followed by elongation, and then the relative positioning of the LV and RV.
- Anatomy-ECG-Demographics Link: Biventricular anatomy, as captured by the SSM, shows statistically significant correlations with both ECG-derived features (like RR interval, QT interval) and demographic variables (height, weight, BSA, BMI). This highlights the interconnectedness of cardiac structure, electrical function, and overall physiology.
- Sex-Related Differences: While the SSM is largely effective across sexes, there are subtle differences in biventricular size, with female hearts generally being smaller, which influences reconstruction accuracy.
- Synthetic Cohorts for Research: The generated synthetic cohort provides a valuable resource for researchers, enabling large-scale simulation studies and the development of data-driven calibration models for cardiac digital twins without the need for sensitive patient data.
Key Results
- The developed SSM captures 90% of shape variability with 16 principal components (PCs), 95% with 29 PCs, and 99% with 94 PCs.
- The average reconstruction error of the SSM is 0.61 mm, which is comparable to the spatial resolution of the CT scans used.
- Statistically significant correlations were found between anatomical variations (PCs) and ECG features (e.g., PC1 and RRint: r = -0.50) and demographic features (e.g., PC1 and height: r = -0.62).
- The study replicated findings that female ventricles are, on average, smaller than male ventricles.
- The impact of dataset size was demonstrated, with larger datasets leading to improved SSM accuracy.
Limitations and Future Work
- Exclusion of Non-Conductive Tissue: The current SSM excludes the non-conductive cardiac skeleton. For mechanical or hemodynamic modeling, valvular anatomy should be included.
- Valvular and Atrial Structures: The authors suggest extending the UVC system to include valvular structures and incorporating atrial anatomy for a more comprehensive four-chamber SSM. This would likely require larger datasets.
- RV Wall Thickness: The current workflow fixes RV wall thickness, which could be improved.
- Validation of Simulation Models: The authors propose using the SSM to validate cardiac digital twin simulation technology by correlating anatomical factors with clinical ECG-derived features.
- Further Exploration of Sex Differences: While sex differences were observed, further investigation is needed to understand the complexity beyond simple geometric scaling.
Summary updated