Figure 4.

Image segmentation approach to evaluate time-lapse series of moving structures. (1) A threshold is applied to the raw CFP time series to generate a dynamic binary mask. The raw YFP series can also be used, but only one mask (from CFP or from YFP) is required. Numbers within the raw images represent virtual pixel intensities; in this example, the threshold was set to an intensity of 50. Now, target structures with intensities above the threshold receive the value “1” (white), while dim/non-fluorescent structures below the threshold receive the value “NaN” for “not a number” (blue), that does not equal “0.” (2) Image segmentation is performed by multiplication of the same dynamic binary mask with the original raw CFP and YFP time series. Within segmented time series, pixels representing (bright) target structures retain their original intensities, while pixels representing dim/non-fluorescent structures are now set to non-numerical values (“NaN,” blue). (3) CFP and YFP intensity changes are extracted from segmented time series in a region of interest (ROI, red) that contains the moving vessel wall throughout the whole time series. “NaN” pixels do not contribute to the ROI's mean intensities. (4) Segmented CFP and YFP time series are used to calculate pixel-by-pixel CFP/YFP ratio series. As background is set to “NaN,” ratio series do not suffer from artifacts occurring when noisy images are divided by each other. The dynamic binary mask can also be used to estimate changes in vessel diameter (see Figure 3D).