A Multi-Task Deep Learning Framework for Characterizing Beating Behavior and Synchrony in Cardiomyocyte Clusters

Figures

• 

  Figure 1. A cardiomyocyte image and its ground truth. (a) Original image; (b) Annotated mask shown in green.

• 

  Figure 2. CardioSegNet architecture.

• 

  Figure 3. Visual Comparison of Segmentation Performance Among the Three Models.Yellow indicates cardiomyocyte regions (foreground); purple indicates background.

• 

  Figure 4. Comparison before and after baseline correction. (a) Beating output curve of Instance 1 before correction; (b) Beating output curve of Instance 2 before correction; (c) Instance 1 after correction, with an inappropriate threshold leading to peak misidentification; (d) Instance 2 after correction with a properly adjusted threshold, showing no misjudgment.

• 

  Figure 5. Example of cardiomyocyte beating performance characterization using CardioSegNet and the pixel difference method. (a) Beating curves of three cell clusters. (b) Beating performance analysis of a single cycle. (c) Performance characterization of the first three beating cycles of Sample 3.

• 

  Figure 6. Beating synchrony analysis based on the PeriodAwareNAPTDij matrix for a representative acquired video. (See Video S3 in the Supplementary Material for the original video.) (a) PeriodAwareNAPTDij matrix; (b–d) Comparisons of beating curves within subgroups area 1, area 2, and area 3, respectively; (e) Schematic of spatial positions of the three subgroups.

• 

  Figure 7. Relationship among physical distance, cluster BRI, and beating synchrony degree of cardiomyocyte clusters. (a,b) Relationship between physical distance and period-aware synchrony metric for Sample 1 and Sample 2, respectively; (c,d) Synchrony networks for Sample 1 and Sample 2. Cell clusters with the same color belong to the same subgroup. Line color indicates the synchrony degree between two clusters. Node color represents the BRI value of each cluster, which reflects the regularity of its beating rhythm.