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 D

Poster G8 - Software Development For Fast and Accurate Cell Identification Across Multiple Imaging Modalities

Friday, October 25, 2024

3:30 PM – 4:30 PM EST

Location: Exhibit Hall E, F & G

Presenting Author(s)

Anastasiia Rudaeva

Undergraduate Research Assistant
University of Rochester
Rochester, New York, United States

Co-Author(s)

Brandon Williams, MS

PhD Candidate in Biomedical Engineering
Boston University, United States

Primary Investigator(s)

JW

John A. White

Chair and Professor of Biomedical Engineering
Boston University, United States

Introduction: Image processing is a critical aspect of biomedical engineering, offering solutions for analyzing and interpreting medical images. In neuroimaging, different microscopes, settings, or scales are often used to collect information from the same group of cells. Image registration is challenging, but necessary to identify the same individual cells across different imaging conditions. Manual cell identification is time consuming, leaves significant room for errors, and is ineffective for high cell density images. This study presents the development of a MATLAB GUI program aimed at simplifying the process of image registration between two microscopes with different magnification and spatial resolution. The application integrates various image processing techniques, including Maximally Stable Extremal Regions (MSER) feature detection, image piecewise adjustment, spatial filtering, and one-by-one automated object labeling to facilitate accurate and efficient analysis. The primary objective is to provide a user-friendly interface that enhances the precision and speed of image analysis to achieve faster and more accurate cell identification.

Materials and

Methods: A targeted illumination, line-scan confocal (TICO) microscope is used to take a primary reference image of neurons expressing the fluorescent voltage indicator, Voltron2-JF585. After voltage imaging experiments are completed, the acute mouse brain slices are fixed and stained using immunohistochemistry for the identification of Parvalbumin (PV+) expressing cells. A point-scanning confocal microscope is used to take a secondary post-hoc image of Voltron2-JF585 and a spectrally separate image of PV+ cells in approximately the same field of view. The MATLAB GUI application was developed using MATLAB's App Designer. First, post-hoc images were rotated and scaled to the TICO image using point pairs for linear, affine, and piecewise 2D geometric transformations. Then, the images were filtered using a special band pass filter, intensity adjustment, area/diameter thresholding, and automated splitting of adjacent cells. Proper filtering allows the use of MSER detection to find individual cells and their centroids. Next, using one identifiable cell on both images, visible neurons on the confocal image were labeled according to x-axes and y-axes distance to the centroids. To find the same positions on the second image, the same manually identifiable cell was used. Program checks the presence of objects in the approximate regions to see if the same cells can be found. After all cells are labeled, we can identify target cells. All the above described steps can be adjusted by the user to accommodate possible differences between images as well as correct possible mistakes in labeling outcome.

Results, Conclusions, and Discussions: Results
The MATLAB GUI application was used to analyze a series of neuroimaging datasets that consisted of TICO and post-hoc confocal Voltron2-JF585 expression images, and post-hoc confocal PV+ cell images. The implemented filtering techniques effectively isolated individual cells, even in high-density images, which allowed it to accurately label cells. The interface was successfully tested on images with cell density starting from 5-7 neurons per image and increasing to densities greater than 50 neurons per image . The automated splitting and thresholding processes significantly reduced manual intervention, thus minimizing errors. The interface allowed parameter adjustment for optimal results, ensuring flexibility in handling diverse image sets. The ability to cross-reference cell positions between images under different settings provided reliable identification of cells across multiple conditions.

Conclusion
The developed MATLAB GUI application demonstrates significant potential in automating and streamlining the analysis of neuroimaging data. By integrating these image processing techniques with a user-friendly interface, the application enhances both the accuracy and efficiency of identifying and labeling individual cells. The simplicity of the interactive interface increases access to the application of image registration to users without requiring understanding of MATLAB coding or processing methods. This tool offers a substantial improvement over manual analysis, especially for high cell density images and large data sets. The application can be a valuable asset for researchers in neuroimaging, facilitating faster and more precise outcomes.

Discussion
Further work can be focused on creating a ground truth dataset with prelabeled cells and intentional changes of quality made to the images to collect statistical data on the accuracy of our image registration program. Another area of work can be finding new ways to improve the automatic object identification and reducing the amount of required manual adjustments. By continuing to develop and improve this application, we aim to provide a comprehensive solution for a common challenge in neuroimaging analysis.

Acknowledgements (Optional):