Fighting MRI Anisotropy: Learning Multiple Cardiac Shapes From a Single Implicit Neural Representation

The anisotropic nature of short-axis (SAX) cardiovascular magnetic resonance imaging (CMRI) limits cardiac shape analysis. To address this, we propose to leverage near-isotropic, higher resolution computed tomography angiography (CTA) data of the heart. We use this data to train a single neural implicit function to jointly represent cardiac shapes from CMRI at any resolution. We evaluate the method for the reconstruction of right ventricle (RV) and myocardium (MYO), where MYO simultaneously models endocardial and epicardial left-ventricle surfaces. Since high-resolution SAX reference segmentations are unavailable, we evaluate performance by extracting a 4-chamber (4CH) slice of RV and MYO from their reconstructed shapes. When compared with the reference 4CH segmentation masks from CMRI, our method achieved a Dice similarity coefficient of 0.91 $\pm$ 0.07 and 0.75 $\pm$ 0.13, and a Hausdorff distance of 6.21 $\pm$ 3.97 mm and 7.53 $\pm$ 5.13 mm for RV and MYO, respectively. Quantitative and qualitative assessment demonstrate the model’s ability to reconstruct accurate, smooth and anatomically plausible shapes, supporting improvements in cardiac shape analysis.

💡 Research Summary

The paper addresses the fundamental limitation of short‑axis (SAX) cardiac magnetic resonance imaging (CMRI), namely its anisotropic resolution and inter‑slice misalignment, which hinder accurate three‑dimensional cardiac shape analysis. The authors propose to leverage high‑resolution, near‑isotropic computed tomography angiography (CTA) data to learn a single implicit neural representation (INR) that can simultaneously model multiple cardiac structures—specifically the right ventricle (RV) and the myocardium (MYO), the latter encompassing both endocardial and epicardial left‑ventricular surfaces.

Two datasets are used. Set 1 consists of 153 CTA scans with automatically generated segmentations of LV blood pool, MYO, RV, and left atrium. Set 2 comprises 140 CMRI scans, each providing paired SAX and 4‑chamber (4CH) long‑axis images; segmentations for LVBP, MYO, and RV are generated automatically. Only end‑diastolic (ED) phase data are employed. From CTA, surface point clouds are extracted using Lewiner Marching Cubes; on‑surface points are sampled and off‑surface points are generated by adding Gaussian noise (σ = 0.33). For CMRI, both on‑ and off‑surface points are taken directly from voxel‑based segmentations. All coordinates are normalized to the