Skip to main content
2024 BMES Annual Meeting Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.
Back

Biomedical Imaging and Instrumentation

Biomedical Imaging and Instrumentation - Poster Session D

Poster S2 - AUTOMATED 4D MYOCARDIAL DEFORMATION ANALYSIS USING SELF ADAPTIVE NNUNET

Friday, October 25, 2024
3:30 PM – 4:30 PM EST
Location: Exhibit Hall E, F & G

Introduction:  Light Sheet Microscopy (LSM) in conjunction with embryonic zebrafish, is rapidly advancing in vivo, 4D characterization of preclinical myocardial contractility. In this regard, rapid imaging speed, centimeter Field-of-View (FOV) and cellular resolution in 4D (3d+time) LSM, have enabled high throughput characterization of dynamic cardio genesis within zebrafish model. However, manual labelling of differentiated cells for real time tracking is a time-consuming task, requiring considerable domain expertise. To overcome this, we present application of nnUnet with LSM platform, for automated cell segmentation across different datasets without any pretraining or hyper parameter tuning.

Materials and

Methods:  The network takes the input of size: A × A × m {A = 2k|k∈N} array, where “m” is the image depth. Owing to single color channel, n =1. 3x3 kernel sizes were utilized for convolutions during down sampling.Each computational block consisted of sequences of convolution, instance realization and leaky Rectified Linear Unit. Loss was calculated as a sum of cross entropy and Dice loss, and stochastic gradient descent optimizer was utilized with a high initial learning rate of 0.01. A custom, non- gated 4D light sheet setup reconstructed ventricular myocardial nuclei, across 4 – 6 days post fertilization (dpf) zebrafish heart maturation stages. 4D greyscale nuclei volumes were binarized using scale space segmentation and frame – frame linking of nuclei trajectories was performed using linear assignment problem framework. Centroids of triangles (polygons) were plotted between global coordinate nuclei vector triad at each time step to establish origin for local coordinate vector transformation. Area of triangle was quantifies for all time points to establish deformation. We employed one-way ANOVA to detect any significant differences. Where differences were found, Tukey’s test was used for pairwise mean comparison (alpha value = 0.05).

Results, Conclusions, and Discussions: We quantified area ratios (AR) of 1.61 +/- 0.09, 1.58 +/- 0.12 and 1.50 +/- 0.07 in the inner ventricular bulge at 4-,5- and 6 dpf for ground truth, manually labelled nuclei. In contrast, ground truth outer ventricular bulge trajectories experienced higher ARs of 1.97 +/- 0.2, 1.86 +/- 0.12 and 1.83 +/- 0.19 at 4,5- and 6-dpf respectively. No significant differences were observed in AR’s calculated from 3D nnUnet nuclei trajectories of inner ventricle at 1.59 +/- 0.13 (p = 0.09), 1.56 +/- 0.09 (p = 0.19) and 1.48 +/- 0.09 (p = 0.18) across 4-6 dpf. Furthermore, no significant differences were observed for 3D nnUnet outer ventricle trajectories of 1.95 +/- 0.2 (p = 0.18), 1.84 +/- 0.19 (p = 0.19) and 1.81 +/- 0.16 (p = 0.065). Hence, we achieved automated segmentation of fused nuclei using a single model across different embryonic zebrafish datasets, quantifying anisotropic and declining ARs of 4-,5- and 6 dpf zebrafish ventricular myocardium indicating cardiac muscle maturity Furthermore. dice scores of 0.954, 0.974 and 0.974 were quantified across across 4-,5- and 6 days post fertilization to establish model accuracy.

Acknowledgements (Optional):