High Throughput In vivo Imaging; A Pipeline for Automated Analysis of Longitudinal Preclinical Studies

Abstract

Fast acquisition times, multi‐animal beds, and improved accessibility of preclinical imaging systems have enabled high‐throughput imaging, significantly accelerating in vivo experimentation. However, image analysis remains largely manual, labor‐intensive, and inefficient, with significant inter‐ and intra‐user variability, adding to the already lengthy timeline of preclinical validation. This article outlines an end‐to‐end pipeline for image acquisition and subject preprocessing, followed by deep learning‐based segmentation and reporting of metrics relevant for validation. This pipeline is applicable to imaging studies investigating biodistribution, dosimetry, disease progression, and treatment effect over time. A longitudinal positron emission tomography–computed tomography (PET‐CT) case study is used to demonstrate the protocol in a practical animal research context. The protocol encompasses image acquisition and reconstruction, fusion, cropping, sorting, metadata addition, deep learning‐based and interactive organ and tumor segmentation, 3‐dimensional visualization, and extraction of quantitative uptake curves. © 2026 The Author(s). Current Protocols published by Wiley Periodicals LLC.

Basic Protocol 1: Image Acquisition and Reconstruction

Basic Protocol 2: Multi‐Subject Cropping and Sorting

Basic Protocol 3: Atlas Alignment and Segmentation

Basic Protocol 4: Quantitation and Reporting

INTRODUCTION

Longitudinal preclinical in vivo imaging enables quantitative assessment of disease progression, treatment efficacy, and biodistribution (Cao et al., 2006; Ferrini et al., 2025). Many workflows, including biodistribution and toxicological assessment, have historically been conducted ex vivo, requiring the sacrifice of subjects at multiple time points (Imlimthan et al., 2025). With recent changes in FDA guidelines and a global push toward the 3Rs (replacement, reduction, and refinement) in animal research, there is a shift towards longitudinal quantification via in vivo imaging (Williams, 2024). Evaluating the same subject at multiple time points enables more accurate tracking of biological processes, while reducing animal use, conserving experimental resources, and minimizing overall study time.

In the context of radiopharmaceutical development, assessing tracer biodistribution across multiple time points and regions is essential for determining binding specificity and quantifying the absorbed radiation dose delivered to tumors relative to organs. The multi‐subject and multi‐timepoint nature of such studies contributes to the accumulation of substantial amounts of imaging data. This accumulation of data increases study complexity and overwhelms manual analysis (Fig. 1).

Figure 1.

Study structure and data accumulation. Eight mice with subcutaneous tumors received intravenous injection of a Zirconium (Zr)‐89‐labeled PET tracer. Half of the mice received anti‐tumor treatment. All mice were imaged on a Mediso nanoScan PET‐CT at 4, 24, 48, 72, 96, and 120 hr after injection. This results in 48 CT scans and 48 PET scans, with a total of 96 3D images to be processed. (Created with BioRender.com).

The most time‐consuming and labor‐intensive step in manual analysis is segmentation: labeling specific volumes, such as organs, tumors, and other regions of interest (ROIs). Across datasets, differences in image resolution, contrast, and acquisition parameters alter the visual definition of anatomical structures, leading to variability in segmentation boundary placement. ROI labeling also varies between annotators, influenced by differences in training, experience, and individual interpretation of anatomical or pathological features. Together, these sources of variability introduce significant inconsistencies that can substantially affect the accuracy, reproducibility, and reliability of quantitative measurements (Jensen et al., 2024; Rosenhain et al., 2018).

There is a significant lack of standardization of segmentation criteria across preclinical studies, which affects the comparability of findings. A recent multicenter preclinical PET‐CT study reported statistically significant differences in mean standard uptake value (SUV) for multiple organs and tumors (Kuntner et al., 2024). These findings emphasize that quantitative PET measurements are highly susceptible to site‐specific variations and reinforce the need for standardized processing and analysis pipelines.

Efforts to develop new segmentation tools have aimed to accelerate the analysis of imaging data. Atlas‐based approaches have proven effective in neuroimaging research, given the brain's rigid structure within the skull (Presotto et al., 2022). Other ROIs, such as the bladder, kidneys, and liver, lack consistent shapes and positions across subjects, making rigid atlas‐based methods ineffective.

Advances in machine learning, the availability of convolutional neural network (CNN) frameworks, and improved access to computational resources have enabled the use of artificial intelligence (AI) for biomedical image segmentation (Ronneberger et al., 2015). This approach has shown success in providing automated segmentation tools for image analysis challenges (Dhaliwal et al., 2024; Schoppe et al., 2020). However, AI‐assisted segmentation is often limited by the distribution of data used in model training, particularly across imaging systems, acquisition protocols, and subject subtypes (e.g., lean vs. obese specimens) (Saat et al., 2022; Wang et al., 2022). Beyond these train‐time limitations, most implementations deploy AI inference as a static process, offering no mechanism for users to adjust or refine the generated segmentation. This approach limits user input and complicates the application of AI‐segmentation in research (Küper et al., 2023).

The inconsistency of manual workflows and the time‐intensive nature of image analysis in longitudinal studies motivate the development of a standardized, generalizable automated pipeline to efficiently and reproducibly extract meaningful preclinical imaging data while maintaining user supervision. Imalytics Preclinical software (Gremse‐IT GmbH, Aachen, Germany) is a multi‐modal image analysis tool that streamlines and automates many of the common processes and analysis steps in multi‐subject longitudinal studies, including a pipeline for data management, improved standardization, and automated segmentation (Gremse et al., 2016), and was recently upgraded with batch processing and AI‐based segmentation features suitable for this protocol.

NOTE: All protocols involving animals must be reviewed and approved by the appropriate Animal Care and Use Committee and must follow regulations for the care and use of laboratory animals. Appropriate informed consent is necessary for obtaining and use of human study material.

Basic Protocol 1. IMAGE ACQUISITION AND RECONSTRUCTION

The purpose of this protocol is to ensure consistent image analysis for high‐throughput longitudinal analysis. This section also includes the documentation of essential subject information required for analysis and guidance for organizing exported imaging data. Different imaging systems have varying formats, data classification conventions, and metadata storage. This leads to fragmented data structures and inconsistent metadata representation, hindering standardization and reproducibility across studies (Zullino et al., 2022). Therefore, rigorous data curation is required to ensure data integrity, standardization, and interoperability across imaging studies (Yamoah et al., 2019).

Materials

• Animal(s) of interest

• Coupled radiolabeled compound of interest (i.e., ⁸⁹Zr, ¹⁷⁷Lu)

• Multimodal in vivo imaging system such as PET‐CT, SPECT‐CT or other multimodal system.

• Imalytics Preclinical Software installed on a Windows/Mac/Linux computer. A system with at least 32 GB of RAM is recommended for responsive rendering and fast AI inference.

• Animal weighing scale

• Dose calibrator

• 1Weigh animal(s) involved in study.
• 2Inject tracer and note injected activity and injection time per animal.
• 3Place the animal(s) within the imaging system and acquire PET/SPECT‐CT scan. Perform image reconstruction.

  Multi‐subject imaging is common in high‐throughput PET/SPECT‐CT using multi‐animal beds/hotels, particularly for smaller animals such as mice (Thamm et al., 2022). Note each animal's assigned bed location.

• 4Repeat steps 1–3 according to the specifications of the desired study (i.e., number of subjects and time points).

  In nuclear medicine studies, long‐lived isotopes such as Zirconium‐89, Lutetium‐177, and others enable scanning for 5–10 days post‐injection due to their long half‐lives (Wu et al., 2024).

• 5Export PET and CT image data per individual time point into one folder per PET‐CT scan (Fig. 2).

  Images can be exported in various formats (e.g., DICOM, NIfTI, GFF).

  If modalities are not inherently co‐registered during acquisition or have been obtained from separate imaging systems, Imalytics Preclinical provides several registration options (interactive, marker‐based, or image‐based) within the Fusion menu. These tools enable users to generate and apply affine transformations to implement standardized multimodal image alignment (Thamm et al., 2023).

Figure 2.

Data Architecture and Folder Organization of Longitudinal Imaging Exports. A. Common folder structure of exported longitudinal multi‐mouse PET‐CT files. indicating date, isotope, study group, and time‐point in the folder name. B. Hierarchical DICOM export of imaging data illustrating complex nested structures.

Basic Protocol 2. MULTI‐SUBJECT CROPPING AND SORTING

This protocol provides an automated preprocessing method to implement hierarchical data organization, structured as [study; subject; time point], from longitudinal multi‐subject acquisitions. Each subject's sorted folder contains all necessary elements, including images (in specified voxel units) and required metadata, to ensure clarity and reproducibility (Fig. 3).

Figure 3.

Hierarchical structure of a sorted data folder processed through the pipeline. A. The images and metadata are organized by study → group → subject → time‐point. (Created with BioRender.com) B. The folder structure is stored as such: “Radiopharmaceutical study\Control group\CTRL1\004h”.

While multi‐subject imaging increases throughput, commonly used image file formats cannot store metadata, such as injected activity, across multiple subjects. Traditionally, images are cropped one at a time, and metadata required for analysis is added retroactively. This further increases the time and effort required for analysis. To address this, the protocol introduces an automated pre‐processing method that crops multi‐subject scans into individual subject regions based on fixed bed dimensions. The process also automatically assigns metadata to each subject and uses this information to perform unit conversions to subject‐specific normalized units (e.g., SUV, percent injected dose per milliliter (%ID/mL)). Finally, the resulting data are stored in a nested hierarchy (subject → time point), ensuring that all time points for a given subject are grouped together, rather than organizing data as multi‐subject scans per time point. While the multi‐subject cropping workflow is specific to mouse imaging, this step can also be used to standardize image size, add metadata, and exclude regions outside the subject body for larger animals, including rats and non‐human primates.

• 1Load any 3D image in which you can visualize the complete (multi)‐mouse bed in Imalytics Preclinical (Gremse‐IT).
• 2Use the marker tool to place indicators in the approximate center of each subject. The red cross hair can help visualize these locations.

  Opening multiple slice views allows for more precision in indicating the subject center. The marker name translates into the subject name in the following steps.

• 3Use the ruler tool to measure the Height, Width, and Depth of one subject region, naming those measurements as such (Fig. 4).

Figure 4.

Demonstration of marker positions and cropping dimensions. Height is denoted along the y‐axis, Width along the x‐axis, and Depth along the z‐axis, in the context of a subject in the supine position.

• 4In the Menu, open Modules → Nuclear → Multisubject Cropping.
• 5Click “Use markers” to populate the subject information for cropping and sorting.
• 6If required, adjust each subject's name, weight, injected dose, and associated measurement time.

  Subject names can be labeled flexibly. In this study, the control and treatment groups are labeled as CTRLX and TREATX, respectively, where X denotes the subject number within each group.

• 7Indicate the position (if any) of subcutaneous tumors (i.e., Flank/Shoulder and associated side (Left/Right)).
• 8Save the template as an .xlsx or .csv file, as this contains the necessary information for any future study using this specific animal bed (Table 1).

  Users can crop this individual scan/time point and generate converted images in commonly used radiopharmaceutical units (SUV, %ID/ml, etc.). This is useful to ensure proper subject splitting.

  The cropping file (typically named Cropping.xlsx) is required for each multi‐mouse PET/SPECT‐CT scan. It contains one row per mouse, specifying subject ID, injection details, cropping parameters, and an optional tumor location. The same cropping file can be reused for subsequent scans. When scanning the same mice at a new time point, only the first column needs updating to reflect the new time point. When scanning different mice, subject ID, weight, injection details, and tumor location must be updated accordingly. However, the cropping parameters will remain consistent across all images for a given imaging system‐bed combination.

Table 1.

Template Structure for the Cropping File Used in Preprocessing

Subject Name	Injection Dose [MBq]	Weight [g]	Injection Dose Measurement Time	Mouse World Center [x]	Mouse World Center [y]	Mouse World Center [z]	Crop Width [x]	Crop Height [y]	Crop Depth [z]	Tumor Position
TREAT1/004hr	5.43	23	2025‐04‐01T10:50:00	14.77	14.52	−57.8	32.24	24.04	98.19	RightShoulder
TREAT2/004hr	5.80	20	2025‐04‐01T10:54:00	−15.5	14.31	−57.8	32.24	24.04	98.19	RightShoulder
TREAT3/004hr	5.31	22	2025‐04‐01T10:57:00	13.44	−23.9	−57.8	32.24	24.04	98.19	RightShoulder
TREAT4/004hr	5.66	21	2025‐04‐01T10:59:00	−17.05	−22.63	−57.8	32.24	24.04	98.19	RightShoulder

• 9For each timepoint or scan, store a cropping file within the corresponding timepoint folder. These files should derive from the same template file; only the subject names should be adjusted following the nested folder structure.
• 10Open Modules → Nuclear → Multi‐Subject Cropping (batch) and select the ‘Create pipeline file’ option.
• 11Indicate the Underlay/Overlay file path, which allows the software to recognize the imaging system's output folder structure. A PipelineSettings.ini file must be saved for each scan, although it is typically identical across scans and can therefore be copied and pasted.

  The PipelineSettings.ini file can also contain adjustments to general scan metadata (e.g., isotope name, scan start time, voxel unit, etc.).

• 12For the ‘Input root folder’, select the directory containing scans prepared in Basic Protocol 1.
• 13Select an output folder in which to store the sorted subject folders.
• 14Check the boxes to create SUV and %ID/ml maps.

  The “Dry run” option allows users to validate file definitions and directory structures prior to execution. It automatically reports missing files or incorrectly specified inputs to ensure compliance with the protocol specifications.

• 15Select Crop to generate the subject preprocessing pipeline and inspect progress. The Imalytics software loads each multi‐subject PET‐CT scan, performs cropping, applies metadata, generates %ID/ml and SUV maps, and saves the cropped PET‐CT data to the sorted folder.
• 16Using the batch mode, navigate through the sorted folder and check if the cropping and fusion worked properly. If the cropping is too tight for some mice, increase the cropping sizes for some or all scans, then rerun the batch multi‐subject cropping.

Basic Protocol 3. ATLAS ALIGNMENT AND SEGMENTATION

This section outlines the development of the machine learning models integrated within the software and utilized in this protocol. To date, the organ segmentation AI model has been trained on 935 µCT (micro‐computed tomography) scans of mice acquired from multiple sites, encompassing data from six different imaging system vendors, including MILabs, Mediso, CT Imaging, Molecubes, Inviscan, and Sofie (Fig. 5). Given the large training dataset and the diversity of acquisition parameters, the spatial positioning of the subject within the scanner represents a critical factor influencing AI inference performance. When an image presents anatomical structures in positions and orientations that deviate from those encountered during training, the AI model's feature recognition accuracy can deteriorate substantially. To mitigate this effect, an image alignment step is implemented to standardize subject positioning prior to model inference, thereby improving consistency across datasets and enhancing segmentation robustness. The relevant models integrated into the software include full‐body (facilitating background and bed removal), organ (liver, kidneys, lungs, heart, bladder, and more), and subcutaneous tumor segmentation models. After AI‐based segmentation, the user should inspect each segmentation in 2D and 3D views and, if required, correct it using interactive segmentation tools.

Figure 5.

Distribution of µCT training data across imaging system vendors used for the Imalytics mouse organ segmentation model.

• 1Load a single subject's underlay (i.e., CT) generated in the cropping steps in Basic Protocol 2.
• 2Select the AI mouse mode and select the model of your choice.
• 3Open the three slice views and align the cyan‐colored mouse template to the anatomical data (Fig. 6).

  This rough alignment is done by zooming, dragging, and rotating. Important landmarks are the rib cage, sternum, and pelvis, shown in the overlaid template.

Figure 6.

Aligned template used for mouse alignment.

• 4Within the batch menu, select “Manage Batch Settings”. Update the settings and enable the nested batch option to ensure the software correctly identifies all scans within the study (including all time points and subjects).

  In the biodistribution study, the body‐segmentation, organ‐segmentation, and tumor‐segmentation mouse models will be used.

• 5Save the alignment file, e.g., as atlas.v2d, in the scan folder.
• 6Go through all the subjects and time points, adjust the alignment, and save the alignment file.

  For batch analysis, transform files must be named consistently. The save () button can be used to automatically name and save transform files as “Atlas.v2d” when created.

• 7Run batch segmentation with the selected AI model by clicking on the batch inference button. Select the viewing transform file name and click “Run” to generate the segmentation for the selected model.

  Instead of using an alignment for each folder, a global alignment file can be selected, e.g., the alignment of the first scan. This may be sufficient if the subjects are aligned and cropped very consistently.

• 8Repeat step 7 for an additional AI model (e.g., tumor segmentation, body segmentation) and add the result to the current batch (Fig. 7).

  Users can also run custom models within the AI module. The protocol for model implementation is described in the Imalytics Manual. The same alignment file can be used if the AI models share their atlas.

Figure 7.

3D renderings presenting the AI‐generated segmentation for the body (A), organs (B), and tumor (C), followed by the combined transparent rendering of all regions.

• 9Review all batch results and inspect segmentations for missing or incorrect regions and correct any errors.

  Users can adjust all segmentation results using various manual and semi‐automated tools available in Imalytics [Help → Show Manual].

• 10Generate 3D rendering for each timepoint in Batch → Save 3D screenshot series.

Basic Protocol 4. QUANTITATION AND REPORTING

This protocol details how to gather meaningful quantitative data from both anatomical and functional scans and generate the figures required for biodistribution and other longitudinal studies. Segmentation data can be used to report the number of voxels within each ROI, as well as the volume, mean, sum, total, standard deviation, and minimum and maximum values in specified units (for both underlay and overlay). 2D images and videos of subjects can be generated in batch mode for standardized visualization and interpretation. For PET‐SPECT/CT studies, time‐activity curves are generated to indicate the rate and magnitude of tracer accumulation and clearance within a specific ROI.

• 1Within the sorted data, load a subject's anatomical (underlay.gff), functional (PercentID.gff), and segmentation data (OrgansAndTumor.segff) into Imalytics Preclinical.
• 2Set the Batch Settings and activate the Nested Option.

  Users should inspect the AI‐generated segmentation results for all subjects in their study and make manual adjustments as needed.

• 3Select the specific layer for which quantitative data is desired (Underlay/Overlay).
• 4Under the batch menu, select the ‘Batch class statistics’ function. This function generates a spreadsheet containing one row per segmented region and time point. Each row reports the region name, time after injection [h], volume [mm³], mean [%ID/ml], total [m%ID], and additional statistical metrics. Longitudinal curves of volume and mean activity are also generated for each labeled region.

  This provides all necessary information for each subject, including the volume profile and time activity curves, with the option to save a report as .xlsx and/or .csv. Volume curves can also help identify subjects for which the AI‐generated segmentation results may require adjustment.

Commentary

Critical Parameters

Acquisition and reconstruction settings influence image quality and downstream segmentation accuracy. Parameters such as reconstruction method and voxel size are important in producing interpretable imaging data. Selecting imaging time points that align with the tracer's uptake, clearance, and physical half‐life ensures that key phases of biodistribution are captured.

During preprocessing, it is important to precisely define cropping dimensions to include all relevant regions while avoiding unnecessary background data. Adjusting the cropping buffer can help avoid data loss.

Consistent naming of subjects, time points, and relevant metadata (i.e., tumor position) facilitates automated data parsing and integration into the hierarchical structure presented in this protocol.

The atlas alignment step enhances the reliability of AI‐powered segmentation. Variations in positioning can significantly influence model inference. Therefore, implementing a standardized alignment procedure improves segmentation robustness and minimizes the need for manual correction.

Several scanner‐specific factors can affect the performance of this pipeline. Body segmentation depends on the mouse bed geometry being consistent with that represented in the training data. A bed whose shape or dimensions differ substantially may introduce errors in the body segmentation. Bore size is also a practical constraint, as scanners with a smaller bore can accommodate only a two or three‐mouse bed, reducing the number of subjects that can be imaged per acquisition. Additionally, the CT data should achieve a spatial resolution of 200 µm or better to support reliable analysis. Therefore, clinical scanners are insufficient for mouse imaging.

Troubleshooting

 

Problem	Possible Cause	Solution
Low nuclear imaging signal	Scanner malfunction or incorrect usage. Failed or incomplete tracer administration. Scanning duration is too short.	Verify scanner performance using a calibration phantom and confirm acquisition settings. Check if total activity in whole segmented body at early time point is close to injected activity. Increase acquisition duration.
Certain subject regions are missing in cropped images	Cropping parameters incorrectly defined in Cropping.xlsx file	Verify mouse center coordinates and increase cropping width, height, and length as needed
Non‐subject structures visible in cropped images or 3D visualizations	Inaccurate cropping centers or sizes. Subjects positioned too closely in the bed. High‐density bed components produce obscuring CT signal.	Apply body segmentation and batch image cleaning to remove voxels outside the target subject, including adjacent subjects and imaging bed components
Quantitative values outside experimentally expected ranges	Incorrect voxel intensity unit (e.g., mCi/mL instead of kBq/mL). Incorrect or missing calibration factor. Incorrect decay correction due to wrong scan start and injection times.	Verify that voxel units are correctly defined. Acquire phantom scan with known activity, determine the correct calibration factor, and apply it in the PipelineSettings.ini file. Double‐check metadata to achieve correct decay correction and %ID/mL conversion.
Inaccurate AI Segmentation	Artifacts (i.e., metal, rings, cupping in CT). The subject exhibited a low body fat percentage, resulting in minimal intrinsic contrast between internal organs. Noisy µCT images. Incorrect HU calibration.	Use advanced µCT reconstruction with artifact correction. Use older mouse models with a higher fat percentage. Adjust µCT scanning protocol to a higher dose or better dose efficiency. Apply correction factor and offset to CT intensities in PipelineSettings.ini file. Manually adjust segmentations.
Missing or inconsistent metadata	Metadata is missing or not entered correctly	Enter metadata for subjects and scan into Excel or PipelineSettings.ini file, respectively
Misalignment between imaging modalities	Intense use of scanner or outdated calibration requiring maintenance	Integrate quality‐control protocols (e.g., phantoms). Correct registration using fusion tools in Imalytics [Menu → Fusion] for individual datasets or generate a correction transform and apply globally via PipelineSettings.ini file.

Understanding Results

Qualitative inspection of longitudinal PET/CT images provides an intuitive understanding of tracer distribution and treatment response. Visual inspection highlights the location and intensity of radiotracer uptake, with warmer regions generally indicating higher activity concentration. By comparing images across time points, researchers can observe trends such as increased tumor uptake and changes in off‐target signals. These visual patterns highlight overall biodistribution and can provide a rapid assessment of treatment effect (Fig. 8).

Figure 8.

Qualitative demonstration of tumor uptake, with associated %ID/mL color map, for a subject in the treated study.

Quantitative imaging metrics and trends, such as specific radiotracer uptake, provide critical information about the biological response, while changes in tumor volume reflect macroscopic treatment effects. In the dataset presented in this protocol, a significant difference in tumor volume was observed between the treatment and control groups, consistent with the expected response (Fig. 9).

Figure 9.

Difference in tumor volume over the time of the study, averaged over subjects within each group. Error bars denote the standard error of the mean (SEM).

In addition to tumor volume, mean tracer uptake in the subcutaneous tumor can, depending on the tracer and research objectives, serve as an indicator of radiotracer targeting efficiency, biological activity, and functional changes within the tumor. The treated group exhibits a pronounced increase in mean tumor uptake (%ID/ml), exceeding a fourfold increase over the control group at the final imaging time point (Fig. 10A).

Figure 10.

A. Difference in mean tumor uptake over the time of the study, averaged over subjects within the group. B. Total Uptake in all regions in the control group, averaged over all subjects within the group. Error bars denote the SEM.

This increasing mean uptake in the tumor suggests tumor shrinkage, which leads to greater activity per unit volume. The total uptake (in %ID) serves as an alternative metric that reflects absolute tracer accumulation independent of tumor volume (Fig. 10B). Additionally, the dose delivered to each ROI can be calculated using a voxel‐based dosimetry method, also available in Imalytics, which can help quantify the total projected dose, as well as the residence times required for subsequent absorbed dose calculations (Ho et al., 2024).

Time Considerations

Longitudinal scanning protocols can take weeks if observing long‐term treatment effects. Image analysis with this protocol can be performed in under one hour for a study containing 8 subjects and 6 time points.

Author Contributions

Mark Tchinov: Conceptualization; data curation; formal analysis; investigation; methodology; validation; visualization; writing—original draft. Zachary Gurberg: Methodology; project administration; software; supervision; writing—review and editing. Diana Möckel: Data curation; methodology; resources; software; writing—review and editing. Stefanie Rosenhain: Data curation; methodology; resources; software; writing—review and editing. Joseph Marsilio: Investigation; methodology; resources; visualization. Sarah Schraven: Methodology; resources; writing—review and editing. Kendahl Ott: Conceptualization; investigation; resources; writing—review and editing. Aaron LeBeau: Investigation; resources; supervision; writing—review and editing. Randall J. Kimple: Investigation; resources; writing—review and editing. Felix Gremse: Conceptualization; funding acquisition; software; supervision; writing—review and editing.

Conflict of Interest

Felix Gremse is a shareholder of Gremse‐IT GmbH. Diana Möckel, Stefanie Rosenhain, and Sarah Schraven are employees of Gremse‐IT Gmbh. Mark Tchinov, Joseph Marsilio, and Zachary Gurberg are employees of Gremse‐IT Canada Inc.

Tchinov, M. , Gurberg, Z. , Möckel, D. , Rosenhain, S. , Marsilio, J. , Schraven, S. , Ott, K. , LeBeau, A. , Kimple, R. J. , & Gremse, F. (2026). High throughput in vivo imaging; A pipeline for automated analysis of longitudinal preclinical studies. Current Protocols, 6, e70414. doi: 10.1002/cpz1.70414

Data Availability Statement

Data available upon request from the authors.