(596a) Image Processing and Immunofluorescent Cell Imaging: Current Challenges and Opportunities

Immunofluorescent imaging is one of the most common ways to visualize and quantify features of cells of the central nervous system (CNS). While there are many fluorescent imaging methods, including confocal, multiphoton, and light sheet microscopy, and many reviews about optimizing imaging parameters of each microscopy method, there are far less papers analyzing image processing techniques applied after image acquisition. All fluorescent images undergo some type of image processing for presentation in a publication – even a simple brightness adjustment is an image processing technique. However, there is minimal detail about image processing methodologies included in publications using immunofluorescent cell imaging; even when methods are mentioned, it is rare that values for specific variables needed for reproducibility are also included. The field has recognized issues related to reproducibility and repeatability of immunofluorescent images and published multiple perspectives. Often, these perspectives propose methods for improved image acquisition parameter selection or through improved cell thresholding segmentation algorithms, yet, the effect has been nominal on the overall field.

All fields using immunofluorescent imaging will benefit from improved reproducibility and repeatability of image processing methods, but we currently need better methods for standardizing, publishing, and searching for image processing techniques and methodologies. We seek to address immunofluorescent image reproducibility through (1) a high-level look at the prevalence of fluorescent imaging on CNS cells and (2) an in-depth look at publishing practices of image processing methods, with specific emphasis on thresholding and segmentation methods, for quantifying CNS cell morphology. Our high-level look at the prevalence of fluorescent imaging on the major CNS cells - neurons, glia, and vascular cells - with common immunofluorescent imaging techniques shows the widespread application of each imaging method on a every cell of the CNS. The effect of image reproducibility is not cell dependent and addressing this issue would improve image processing methods for all fields studying cell biology. To further identify image processing issues on immunofluorescent imaging of CNS cells, we completed a deeper dive by thoroughly analyzing the papers returned by the earlier search. The deeper dive highlights both the publication practices surrounding image processing methodologies and wide variety of models, software, and quantified features for each cell type. We combine our high-level approach and deeper dive to point out areas for improvement in publication practices behind image processing of immunofluorescent images while also visualizing new connections for researchers to explore new software, models, or feature quantification technique for any cell of the CNS.

Methods

Prevalence of Immunofluorescent Microscopy Methods for Imaging CNS Cells

We identified the major cells of the CNS, neurons, glia, and vascular cells, and then subdivided each category into specific cells or classifications. The neuron group was divided by both functional differences and morphological differences. The glia group was further divided into microglia, astrocytes, and oligodendrocytes. Finally, the vascular cells were divided into endothelial cells, pericytes, and vascular smooth muscle cells. The immunofluorescent microscopy techniques searched in combination with each subgroup of CNS cells are confocal, general fluorescent, fluorescent widefield, multiphoton, light sheet, total internal reflection, and super resolution microscopy. For each CNS cell, we determined the related medical subject heading (MeSH) term, except for the neuronal morphological differences groups which do not have MeSH terms. For the immunofluorescent imaging methods, only confocal microscopy and general fluorescent microscopy have MeSH terms. For all other microscopy techniques, we used the MeSH term “Microscopy, fluorescence” and the type of microscopy as an additional search term. All searches were completed on the same day using PubMed. For the first search, we used the CNS cell word/MeSH term and the microscopy word/MeSH term combined with the join AND. For the second search, we did three searches: a primary search which added the MeSH term “Image Processing, Computer-Assisted” with the word AND, a secondary search looking for either segment OR threshold, and a tertiary search looking for image NOT the MeSH term AND segment OR threshold. Once the search was complete, a CSV file of all paper metadata including title, author, and PubMed ID was saved for deeper analysis.

Deeper Dive Exploration of CNS Cells Imaged with Confocal Microscopy and Tagged with Mesh Term for Image Processing

For all papers returned by our literature search – the search using the MeSH term for confocal microscopy, the CNS cell word or term, and the MeSH term for image processing - we separated all papers published after 2010. For all separated papers, we tagged the CSV metadata of the publication with the following categories: animal or cell model, quantified features, threshold techniques used, segmentation techniques used, and general image processing software used. We then compiled a table of each CNS cell (rows of the table) and under each category (columns of the table) included all tagged features and publication references. After compiling the table, we alphabetized each of the table’s cells and applied a two-color heatmap per column for the number of publications that return that specific feature of the category where higher intensity color correlated with higher number of publications.

Results

Confocal microscopy is the most common fluorescent imaging method with multiphoton microscopy as the second most common. Among the cells, sensory receptor cells and endothelial cells have the highest number of overall publications. Additionally, endothelial cells have the most overall use of different microscopy techniques. However, second to endothelial cells, astrocytes also have quite a wide range of fluorescent microscopy techniques used in multiple publications even though they are not the second most published cell. When specifying the search to include papers with the image processing MeSH term, overall publications decrease by a factor of ten for all CNS cells. The decrease in publication prevalence shows that even though authors are regularly using image processing, they are regularly under reporting image processing as a technique within their publications.

After tagging the confocal and image processing publications for specific features, we found that mice are the most common model used and the most common features quantified are count and intensity. It is likely that count and intensity are the most common measured features as they are the most accessible without advanced techniques or software. We also found that the most common segmentation and threshold methods cited in publications were “method unspecified” – the authors claim thresholding and segmentation occurred, and in some cases the software used, but do not include details as to how the processes were carried out. Finally, the most common software used is ImageJ – the open-source image processing software from the NIH – followed by Imaris, with MATLAB being the most frequently used coding language. Prevalence of software use shows the importance of open-source software on cell image processing, but also the wide variety of tools that are available.

Conclusions

Our literature analysis provides an overview of the state of the immunofluorescent cell imaging field and highlights areas where increased methodology publication or publication tagging would make a positive difference. Additionally, our deep dive into specific features shows that image processing methodologies are under-published with both regards to how often they are tagged as image processing and to the variables used for thresholding and segmentation. Our deep dive table can be used by researchers to explore new quantification techniques and methods to apply to their pre-existing work based on similar publications. Immunofluorescent imaging will likely continue to be one of the most popular methods for visualizing and quantifying CNS cells. However, we need to have more rigorous criteria and standards for publication of image processing methods so that we can standardize cell quantification and improve imaging analysis reproducibility.