Automatic detection of adult cardiomyocyte for high throughput measurements of calcium and contractility

Figures

• 

  Fig 1. Illustration of the MultiCell high throughput system. A 35mm dish is placed in the microscope, MultiCell user interface showing the automatic cell finding feedback and the IonWizard acquisition window with sarcomere length traces.

• 

  Fig 2. Structure of automatic cell finding scheme. Frame layer finds valid cells in the FOV. Filter layer finds the best subregion for sarcomere length measurement. It includes three main steps: fine autofocus, angle correction and region selection. Once the subregion is found, contractility measurement is conducted in this region, top trace shows sarcomere length, bottom trace shows cell length trace.

• 

  Fig 3. Pipeline of static and motion cell finding. Static cell detection only finds cells within a predefined minimum/maximum area and width and height ratio. Motion cell detection finds cells with sufficient motion information in it.

• 

  Fig 4. Validation result of 1 dish of rat cardiomyocytes. Image is a stitched and overlay view of 54 videos annotated manually (white dots) and found automatically by the algorithm (blue and green overlay i.e. True Positives). Blue overlay indicates cells which were filtered out for measurement based on a low SNR of the FFT signal. Red box indicates cells which were found by the software, but not annotated manually i.e. False Positives. Yellow boxes indicate cells that were annotated manually but not found by the algorithm i.e. False Negatives.

• 

  Fig 5. Results isoprenaline (30nM) experiment. A. Average contractility trace before and after isoprenaline (ISO, n = 21) shaded area indicates the 95% confidence interval. B. Average calcium trace before and after ISO (n = 21). C. Fractional shortening of each cell before and after ISO. D. Calcium amplitude change of each cell before and after ISO treatment.

• 

  Fig 6. Data distribution of 5 dishes measured manually and automatically. A. Baseline sarcomere length (SL) distribution tends to shift to the left with automatically acquired data. B. Fractional shortening is shifted to the left in most dishes with automatically acquired data.