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.

Biomedical Imaging and Instrumentation

Biomedical Imaging and Instrumentation - Poster Session A

Poster F6 - A Pipeline for Hessian-Based Frangi Filtering of Two-Photon Imaged Cerebral Blood Vessels

Thursday, October 24, 2024
10:00 AM - 11:00 AM EST
Location: Exhibit Hall E, F & G

Introduction:  Two-photon microscopy is a fluorescence imaging technique that enables deep tissue visualization with minimal photobleaching and high spatial resolution. By utilizing near-infrared light, this imaging technique facilitates the excitation of fluorescent molecules within biological tissues, providing detailed images of cellular structures and blood vessels. However, two-photon microscopy images can result in significant noise and image artifacts, complicating efforts to segment the vessels for 3D quantification. Various modalities of image filtering have been developed to combat these issues more generally in biomedical images. One such modality, used for vessel imaging, is Hessian-based Frangi Filtering, which enhances vascular structures based on their “vesselness”, allowing researchers to better distinguish between vessel and non-vessel pixels in images. By leveraging the Hessian matrix, the filter accentuates areas with tubular shape features and suppresses areas with other shape features, significantly improving the clarity and contrast of vessel structures. Developing a robust pipeline for applying Frangi vessel filtering to two-photon images allows for more efficient segmentation of vessels to enable quantitative analysis. We developed and tested a Hessian-based Frangi filter pipeline on 5-minute time series 2D images in the brains of normal BalbC mice.

Materials and

Methods:  The pipeline consisted of preprocessing the images in FIJI, followed by uploading the preprocessed TIF image files to a MATLAB script which applied the hessian-based Frangi filtering, binarized the image, and output images of the paravascular space and parenchyma. Preprocessing of the images was conducted by , first, performing time series alignment to correct any motion artifacts. Then, we enhanced the brightness and contrast to ensure the vasculature was visualized. An average image was then generated from all images in the time series to reduce noise and improve contrast. From there we refined a MATLAB script to run the preprocessed image through a Frangi Vesselness Filter. The output is converted to a binarized image using thresholding. To ensure accurate vessel segmentation, the binarized mask was further processed in FIJI, where we overlaid the mask onto the original averaged time-series image and manually refined the mask to separate capillaries from larger vessels. This involved cleaning up the mask to remove noise and non-vascular structures, ensuring only the relevant vasculature was included.
Finally, we wrote an additional MATLAB script to automatically segment the paravascular and parenchymal space using refined vessel masks and a series of dilation and subtraction commands. The final segmentation masks were used to extract metrics of interest including average intensity calculations and timecourse plots for different vessel types, for paravascular spaces, and for the parenchymal space, to be used for further analysis of dye transport rates in our experiments.

Results, Conclusions, and Discussions: Optimization of this pipeline to best capture capillaries in two-photon images required systematic adjustment of the parameters of the Frangi filter including the values for the Frangi scale range, scale ratio, β1, and β2. The Frangi scale range controls the range of sigma values used in the Gaussian filter during the vesselness computation. These sigma values determine the size of the structures that the filter can detect, with lower sigma values highlighting finer structures like small capillaries and higher sigma values enhancing larger vessels. The Frangi scale ratio, on the other hand, defines the step size between successive sigma values within the specified range. The optimal value for the Frangi scale range, which enhanced the visibility of smaller capillaries, was determined to be [2, 6], with a Frangi scale ratio of 5. We also increased the β1 value from the default of .5 to 1 and significantly increased the β2 values from the default of 15 to 1000. Ultimately a collection of optimized parameters were determined and we found that the β2 value had the most prominent effect on reducing the enhancement of nonvascular blob-like structures that appeared. We were able to create accurate masks from the resulting filtered and binarized images that depicted the location of capillaries in the mice brains.

Acknowledgements (Optional):