Validation of diffusion tensor MRI measurements of cardiac microstructure with structure tensor synchrotron radiation imaging

Abstract

Background: Diffusion tensor imaging (DTI) is widely used to assess tissue microstructure non-invasively. Cardiac DTI enables inference of cell and sheetlet orientations, which are altered under pathological conditions. However, DTI is affected by many factors, therefore robust validation is critical. Existing histological validation is intrinsically flawed, since it requires further tissue processing leading to sample distortion, is routinely limited in field-of-view and requires reconstruction of three-dimensional volumes from two-dimensional images. In contrast, synchrotron radiation imaging (SRI) data enables imaging of the heart in 3D without further preparation following DTI. The objective of the study was to validate DTI measurements based on structure tensor analysis of SRI data.

Methods: One isolated, fixed rat heart was imaged ex vivo with DTI and X-ray phase contrast SRI, and reconstructed at 100 μm and 3.6 μm isotropic resolution respectively. Structure tensors were determined from the SRI data and registered to the DTI data.

Results: Excellent agreement in helix angles (HA) and transverse angles (TA) was observed between the DTI and structure tensor synchrotron radiation imaging (STSRI) data, where HA_DTI-STSRI = -1.4° ± 23.2° and TA_DTI-STSRI = -1.4° ± 35.0° (mean ± 1.96 standard deviation across all voxels in the left ventricle). STSRI confirmed that the primary eigenvector of the diffusion tensor corresponds with the cardiomyocyte long-axis across the whole myocardium.

Conclusions: We have used STSRI as a novel and high-resolution gold standard for the validation of DTI, allowing like-with-like comparison of three-dimensional tissue structures in the same intact heart free of distortion. This represents a critical step forward in independently verifying the structural basis and informing the interpretation of cardiac DTI data, thereby supporting the further development and adoption of DTI in structure-based electro-mechanical modelling and routine clinical applications.

Keywords: Cardiovascular magnetic resonance; Diffusion tensor imaging; Structure tensor; Synchrotron radiation imaging; Tissue microstructure; Validation.

Fig. 1

a Schematic diagram and b photograph of the SRI setup. The coherent X-ray beam passes through the heart and into the detector, which is comprised of a scintillator, that converts X-rays into visible light, a mirror, lens system and charge-coupled device (CCD) camera

Fig. 2

(Top) Non-diffusion-weighted image in a mid-ventricular short-axis slice acquired at 100 μm isotropic resolution, including magnified views in the lateral wall of the left ventricle as seen from the apex – base (AB), anterior wall - posterior wall (AP) and lateral wall - septal wall (LS), along with corresponding diffusion tensors. (Bottom) Matching SRI images reconstructed at 3.6 μm isotropic resolution reveal cellular (dark grey) and extracellular (extracellular fluid and vasculature; light grey) structures, along with corresponding structure tensors. Tensors are coloured by helix angle as determined by v _1,DT and v _3,ST

Fig. 3

a Maps of helix angle (HA) and transverse angle (TA) in whole ex vivo rat heart based on DTI and STSRI, including difference maps; b, c Bland-Altman plots of HA and TA measured with DTI and STSRI in all voxels in the global myocardium

Fig. 4

Transmural variation in helix angle (HA) and transverse angle (TA) in global myocardium as measured with DTI and STSRI. Regions were defined by the 17-segment American Heart Association model (bottom right). Means (solid lines) and standard deviations (dashed lines) are given over all voxels in each region. Angles were normalised to wall thickness in each region, with 0% and 100% corresponding to the LV subepicardium and subendocardium respectively

Fig. 5

Tracking of eigenvectors corresponding to the cardiomyocyte long-axis in whole heart. v _1,DT and v _3,ST show excellent correspondence. Tracks are coloured by orientation: apico-basal (red), anterior-posterior (green) and lateral-septal (blue)

Fig. 6

Tracking of principal eigenvectors in a 0.5 mm thick mid-ventricular short-axis slice. Based on STSRI data reconstructed at 3.6 μm isotropic resolution, v _3,ST, v _1,ST and v _2,ST generally corresponds with v _1,DT, v _2,DT and v _3,DT. Tracks are coloured by orientation: apico-basal (red), anterior-posterior (green) and lateral-septal (blue)

Fig. 7

a Representative short-axis slice through the LV myocardial wall. b Simulated simplified 3D microstructure in a 0.9 × 0.9 × 0.1 mm transmural region in the myocardium. Helix angles (HA) ranged from −88° (subepicardium) to 88° (subendocardium). c Magnified 0.1 mm isotropic region showing the simulated cell long-axis, sheetlet and sheetlet-normal directions. d, e Structure tensors reconstructed at 0.1 mm isotropic resolution with kernel sizes of 7³ (K7) and 11³ voxels (K11). Tensors are displayed as superquadric glyphs and coloured by helix angle. In both cases, v _3,ST is oriented along the voxel-averaged cell long-axis. In contrast, v _1,ST is aligned to the sheetlet-normal direction for K7, and the sheetlet direction for K11. f Illustration of diffusion tensors

Fig. 8

Helix angle (HA), transverse angle (TA), sheetlet elevation (SE) and sheetlet azimuth (SA) in LV myocardium reconstructed with DTI and STSRI using different reconstruction parameters. Here, high-resolution STSRI data with isotropic pixel size of 1.1 μm were analysed using centre frequencies (CF) of $\frac{2\pi }{3},\frac{\sqrt{2}\pi }{3},\frac{\pi }{3},\frac{\pi }{3\sqrt{2}},\frac{\pi }{6}$ and corresponding kernel sizes of 7³, 7³, 9³, 11³, 17³ voxels. The STSRI image corresponding to this region is given in Fig. 9

Fig. 9

a Pilot SRI data in the LV myocardium reconstructed at 1.1 μm isotropic resolution shows improved definition of cells, sheetlets, arterioles, venules and capillaries. Scale bar is 500 μm. b Magnified SRI image. c Normalised maximum-intensity-projection (MIP) of SRI data through a 100 μm slice, corresponding to a single-voxel thickness in the DTI data, shows bright vascular structures. d Normalised MIP of SRI data with vascular component set to 0. e Helix angle (HA), transverse angle (TA), sheet elevation (SE) and sheet azimuth (SA) based on DTI, STSRI and STSRI data with vessels excluded. Here, the ST kernel size = 7³ voxels, centre frequency = $\frac{2\pi }{3}$, and SE and SA were calculated based on v _3,DT and v _1,ST. Excluding vessels from the ST reconstruction results in modest reduction in heterogeneity due to vessel contribution (see arrows)