Quantification of Myocyte Disarray in Human Cardiac Tissue

Figures

• 

  Figure 1. Cardiomyocyte orientation by frequency analysis. (A) Cardiomyocyte long axis orientation (dashed line) is perpendicular with Z-bands (Scalebar: 16 μm). (B) Intensity profile of the Power Spectrum Density (PSD) of the selection (red square) in (A) is shown in the XZ plane. (C–E) Myocyte orientation in the frequency space: (C) the 16 × 16 µm selection of the image in (A) is shown with the original pixel size; (D) frequency information contained in (C); (E) position of frequency peaks is highlighted (orange circles), and original myocyte long axis is restored.

• 

  Figure 2. 3D spectrum manipulation pipeline. The scheme shows the frequency analysis approach developed to evaluate the reliability of the Z-bands signal and to extract the 3D orientation. S: frequency spectrum of a 16 × 16 × 16 μm virtual sample chunk; S′: spectrum with isotropic voxel resolution; Sf′: filtered spectrum; Sxxf′ and Sxx′: power spectrum densities.

• 

  Figure 3. Spatial resolutions of local disarray quantification. A regular cardiomyocyte architecture is schematized with the different spatial resolutions involved in the local disarray quantification. The minimum (a 32 μm side cube) and the maximum (a 96 μm side cube) macrovoxel include, respectively, 2 × 2 × 2 and 6 × 6 × 6 orientation vectors evaluated in a sub-volume of 16 μm sides (in yellow). In red is highlighted the average spatial resolution of 64 μm, selected to create the 3D maps of the local disarray, and for the myocyte angular dispersion estimation by virtual sample comparison.

• 

  Figure 4. Virtual sample generation. On the right, a representative 3D portion (base) of a Z-bands reconstruction previously aligned with Y axis. Virtual cardiomyocytes are extracted from the base and inserted in the silico sample (on the left). Base rotations are randomly selected from two normal distributions (here, θ¯=φ¯=0∘ and σθ=σφ=10∘).

• 

  Figure 5. Imaging and segmentation results. Reconstruction and segmentation of the contractile fibers in a human septal tissue strip, labeled with an anti-α-actinin antibody and imaged with a two-photon fluorescence microscope. (A) A representative sagittal plane of the reconstruction (Scale bar: 100 μm, depth: 150 μm, pixel size: 0.44 μm). (E) Sarcomere Z-bands are visible in the whole strip at sub-micrometer resolution. Results of the main steps of the image processing pipeline are shown: deconvolution and equalization (B,F), binarization by means of the pixel classification (C,G), and the final tissue segmentation (D,H) obtained by smoothing the previous result.

• 

  Figure 6. Accuracy of FFT-based orientation analysis. (A,B) Tables show the performance of the FFT-based orientation analysis averaged on 140 portions of 16 × 16 × 16 μm randomly extracted from three IVS strips. Virtual rotations on XY-plane (azimuth φ) and out-of-plane (elevation θ) are applied to each portion of Z-bands signal. The absolute error of the proposed method is shown in degree for both φ and θ. (C) Relative errors are grouped for all the 13 × 13 virtual rotations applied on the 140 portions and shown as distributions.

• 

  Figure 7. Cytoarchitecture reconstruction and disarray mapping. The 3D reconstruction of a representative strip (IVS1) is shown. The volume is virtually dissected in chunks of 16 × 16 × 16 μm. The FFT-based analysis is applied to each portion to extract the local fiber orientation. The entire cytoarchitecture is collected and represented in a vector space. Vectors color represents the alignment degree with the main axis of force production (Y). Then, local cellular disarray d is evaluated with a pre-defined resolution (here of 64 × 64 × 64 μm), and a 3D map of the cell misalignment is created smoothing the results with a gaussian kernel with σ = 10 μm.

• 

  Figure 8. Disarray characterization in control and pathological samples. (A) Two representative tissue disarray maps of a control (IVS2) and a pathological (AF2) cardiac tissue strip are evaluated with a spatial resolution of 32 μm. A virtual cut (gray dashed lines) is applied to both maps to show internal sections (white dashed lines). Maps show clear differences between healthy and pathological tissue. (B) Results of a multi-resolution cellular disarray investigation applied on three septum tissue strips (IVS1, IVS2, and IVS3) and two fibrillating atrium tissue strips (AF1 and AF2) strips are shown. The raw distributions of the local disarray values are shown. A log-normal PDF is used to fit disarray distributions, and results are normalized separately for each spatial resolution. Global statistics are shown as Global Disarray (D), mode, and SD of the distributions.

• 

  Figure 9. Evaluation of cellular angular dispersion by means of virtual samples generation. Artificial samples of 1,300 × 400 × 280 μm were generated with different (σθ, σφ) values to estimate the actual 3D myocyte angular dispersion by the local disarray distributions found in control and pathological samples. (A) Disarray analysis is performed with a fixed resolution of 64 µm on three artificial samples created with σθ=σφ=0∘, 10°, and 20°. For each sample, the 3D map of cellular disarray is shown partially cutted along the image plane, and the distributions of the local disarray values are shown. (B,C) The dependence between disarray distribution and spatial resolution is confirmed only in disordered citoarchitectures using log-normal PDF to fit disarray in two virtual samples with σθ=σφ=0∘ (ordered, B) and 20° (disordered, C). (D) Log-normal PDF of cellular disarray at 64 µm is shown for each cardiac tissue strip (gray curves). Average PDF is calculated for control (blue) and pathological (orange) strips. The cellular angular dispersion in the tissue is extracted comparing the disarray PDF of two artificial samples with σθ=σφ=5∘ (control) and 20° (pathological).