Label-free, High-Resolution 3D Imaging and Machine Learning Analysis of Intestinal Organoids via Low-Coherence Holotomography

Summary

We present a step-by-step protocol for high-resolution, label-free, and three-dimensional imaging of organoids using low-coherence holotomography. This protocol details organoid culture preparation, imaging acquisition, and computational image analysis, enabling real-time visualization of structural dynamics and drug responses in living organoids.

Abstract

Accurate, label-free imaging of intestinal organoids is crucial for studying their morphology, growth dynamics, and responses to environmental stimuli. Holotomography (HT) provides high-resolution, three-dimensional (3D) visualization of live organoids without the need for fluorescent markers, thereby minimizing phototoxicity and preserving sample integrity. Real-time phase-based imaging allows continuous, label-free tracking of structural and functional changes. By using the refractive index as an intrinsic imaging contrast, this method enables quantification of biophysical properties such as volume, protein density, and protein content. The imaging data are further processed through machine learning-driven segmentation and feature extraction to support consistent, high-throughput analysis. This protocol details the complete experimental workflow for employing low-coherence HT in organoid research, covering organoid preparation, imaging acquisition, and machine learning-based data analysis. By integrating computational segmentation and quantitative assessments, this approach enables unbiased evaluation of key organoid properties, including viability, structural organization, and drug response. The ability to capture real-time morphological changes at subcellular resolution makes this protocol highly applicable to organoid-based studies in regenerative medicine, disease modeling, and pharmaceutical screening. The step-by-step methodology outlined here facilitates reproducibility and broad adaptation across different organoid systems.

Introduction

Organoids are three-dimensional (3D) miniaturized versions of organs grown from stem cells, capable of recapitulating key structural and functional features of real organs in vitro¹. Researchers have developed organoid systems for a wide range of tissues - including the intestine, brain, lung, liver, and kidney - each demonstrating the self-organizing architecture and cell-type diversity reminiscent of its in vivo counterpart. By mirroring native organ architecture and multicellular complexity, organoids serve as powerful model systems for studying human development, tissue regeneration, and disease mechanisms¹. They have become pivotal tools for disease modeling and drug discovery, enabling researchers to explore disease pathways and test new therapeutics on patient-specific 3D tissues in the lab². Notably, organoids also hold promise in regenerative medicine³: organoid-derived tissues can be engrafted in vivo to repair damage, as demonstrated with intestinal and liver organoids, underscoring their potential for restoring organ function⁴. This versatility makes organoids indispensable in modern biomedical research and personalized medicine.

Given their multicellular composition and 3D structural complexity, capturing and analyzing organoid dynamics is crucial. However, conventional imaging techniques face significant limitations in resolving organoid structures and dynamics, often producing 2D snapshots that fail to represent their full physiological state⁵.

Brightfield microscopy is a widely used, non-invasive method for capturing organoids. It conveniently monitors growth and morphology but provides limited contrast in unstained 3D samples. Additionally, it lacks optical sectioning, resulting in poor resolution of internal structures. Organoids appear largely transparent, hindering the detection of fine features beneath the surface⁶. Phase contrast enhances visibility by converting phase shifts to intensity differences, revealing general architecture (e.g., a central lumen) without staining. Despite its utility for real-time observation, phase contrast suffers from halo artifacts, low resolution, and limited penetration depth, making it less effective for thicker organoids⁷.

Serial sectioning followed by hematoxylin and eosin (H&E) staining and antibody-based immunostaining is widely used to analyze organoid microstructure and cellular composition⁸. Improved tissue processing techniques, such as formalin-fixed paraffin embedding (FFPE) and cryosectioning, enhance morphology preservation, with FFPE offering superior results⁹. Embedding organoids in hydrogels like agarose and centrifugation align multiple samples in a single section, improving analysis efficiency. H&E-stained sections of patient-derived tumor organoids reflect original tumor histology, while immunofluorescence staining highlights markers like CK8 for cell identification¹⁰. However, these techniques cannot monitor live organoids and may introduce artifacts such as tissue wrinkling or deformation during sectioning.

Confocal fluorescence microscopy is widely used for organoid imaging, yielding high-resolution (sub-micrometer) optical sections that enable detailed visualization of organoid architecture and cellular composition^11,12. However, confocal imaging of 3D organoids is typically limited to superficial layers (up to 100 µm depth) and can be time-consuming and phototoxic due to point-scanning laser illumination¹³. Light-sheet fluorescence microscopy (LSFM) overcomes some of these limitations by illuminating only the imaging plane, achieving rapid volumetric imaging of entire organoids with minimal photobleaching and phototoxicity^14,15; however, this technique often requires tissue clearing for optimal imaging of thick specimens. Additionally, LSFM imaging depth is restricted by light scattering and refractive index (RI) mismatches, which can be mitigated through adaptive optics, multi-view imaging, and optimized excitation wavelengths¹⁶. Sample mounting also introduces structural deformations, while large 3D datasets pose computational challenges in storage and analysis. Recent advancements, including improved genetic labeling techniques, novel clearing reagents, adaptive optics, and AI-driven image analysis, are addressing these limitations, making LSFM increasingly feasible for organoid research⁷.

Despite these advancements, 3D fluorescence imaging techniques pose several challenges for the long-term monitoring of live organoids. First, the use of fluorescent labeling introduces phototoxic effects and perturbs biological processes, posing challenges for prolonged live imaging studies¹⁷. Second, antibody-based labeling suffers from limited penetration due to diffusion barriers within dense organoid structures, requiring advanced permeabilization and tissue-clearing methods, which take longer than 10 days of incubation for the use of first and secondary antibodies¹⁸. Third, transfection and stable fluorescent protein expression in organoids remain inefficient, as conventional gene delivery approaches struggle with 3D environments¹⁹. Fourth, optical clearing techniques, essential for deep imaging, often lead to fluorescence signal loss and tissue distortion, necessitating optimized clearing protocols that balance transparency and preservation of fluorescence²⁰.

To address these limitations, holotomography (HT) has been introduced as a technique for imaging unlabeled live organoids with high resolution over time. Also known as 3D quantitative phase imaging, HT is a label-free, high-resolution modality that enables real-time, quantitative visualization of organoids while preserving their native physiological state. The principle of HT is analogous to X-ray computed tomography; by capturing multiple 2D optical field images under various illumination conditions, the 3D RI distribution of an unlabeled sample is reconstructed through inverse wave equation solving^21,22,23. HT has been widely applied across various biological fields, including hematology^24,25, phase condensation in cell biology²⁶, microbiology²⁷, and histopathology²⁸. More recently, its high-resolution, label-free 3D imaging capability has been increasingly utilized for organoid research^21,22,23,29, offering a non-invasive approach to studying their structural and functional dynamics.

In this study, we present a detailed protocol for the use of low-coherence HT in long-term, label-free monitoring of mouse small intestinal organoids (sIOs). This method enables real-time visualization of organoid growth and drug responses over 24 h while providing quantitative measurements of organoid volume, protein density, and protein content, facilitating rigorous, data-driven analysis of organoid behavior. To investigate morphological alterations in organoids, we treated the samples with cisplatin, a platinum-based chemotherapeutic agent known to induce apoptosis by forming DNA crosslinks and inhibiting cellular replication³⁰. While 3D RI imaging captures structural details, systematic analysis is essential for quantifying organoid dynamics. To achieve this, we integrate ilastik, a machine learning-based segmentation toolkit³¹ to differentiate organoid regions from the background. This protocol outlines the complete workflow, including organoid culture preparation, 3D HT acquisition, machine learning-based segmentation, and quantitative analysis from RI tomograms. By offering a scalable, reproducible, and non-invasive imaging framework, this approach establishes a new standard for organoid research with broad applications in disease modeling, regenerative medicine, and drug screening.

Protocol

NOTE: Details related to all materials, instruments, and software used in this protocol are provided in the Table of Materials. Equivalent products from different manufacturers may be used as alternatives. A schematic process of the entire protocol is presented in Figure 1.

1. Sample preparation

1. Preparation of sIO culture medium
   1. Prepare the commercially available culture medium according to the manufacturer's instructions. Sterilize the culture medium by filtering it through a 0.22 µm filter unit.
      NOTE: If the medium is prepared entirely under sterile conditions in a cell culture hood and all components are pre-sterilized (e.g., provided in sterile bottles and vials), then filtration through a 0.22 µm filter may not be strictly necessary.
   2. Aliquot 15 mL of prepared medium into sterile storage tubes. Warm the aliquoted medium to room temperature (15 - 25 °C) before use in organoid culture.
2. Thawing sIO stock
   1. Quickly thaw the frozen organoid stock within 2 min using a 37 °C water bath. Transfer the thawed organoid suspension to a microcentrifuge tube and centrifuge at 150 x g for 3 min to pellet the organoids.
   2. Carefully remove the supernatant to wash out DMSO, minimizing potential cytotoxic effects.
   3. Add media and fresh extracellular matrix (ECM) at a 1:4 ratio and gently pipette to ensure uniform resuspension of the pellet. Dispense 15 µL per dome in each well of a 48-well plate. Keep ECM on ice to prevent premature polymerization.
   4. Place the plate upside-down in a 37 °C, 5% CO₂ incubator for 1 h to allow ECM polymerization. After polymerization, carefully add 200 µL of fresh media to each well.
      NOTE: This protocol is compatible with various commercial imaging dishes and well plates. Adjust the media volume according to the well type to ensure that the ECM dome is fully submerged.
3. Passaging of sIOs
   NOTE: Organoids tend to adhere to pipette tips. To minimize adherence, pre-wet pipette tips with the medium.
   1. Aspirate spent medium. Add 200 µL of cell recovery solution to each well and incubate at 4 °C for 30 min to facilitate ECM degradation and organoid retrieval.
   2. Collect the organoid suspension by gentle pipetting and transfer it to a microcentrifuge tube. Centrifuge at 150 x g for 3 min, then carefully remove the supernatant.
   3. (Optional) To passage as single cells, perform one of the following dissociation methods:
      1. Enzymatic method: Add 100 µL of single-cell dissociation solution to the pellet and incubate at 37 °C, 5% CO₂ incubator for 1-2 min. Add 150 µL of media to neutralize. Perform pipetting or gentle tapping. Centrifuge at 150 x g for 3 min and remove the supernatant, leaving only the pellet.
         NOTE: Avoid over-incubating, as prolonged exposure can be highly detrimental to cells.
      2. Mechanical dissociation: Resuspend the pellet in 200 µL of culture medium. Using a P200 pipette, pipette up and down 20x-30x to mechanically dissociate organoids into smaller fragments or single cells. Centrifuge at 150 x g for 3 min, then aspirate the supernatant and retain the pellet.
   4. Add media and fresh ECM at a 1:4 ratio and gently pipette to ensure uniform resuspension of the pellet. Dispense 15 µL per dome in each well of a 48-well plate.
   5. Place the plate upside down in a 37 °C, 5% CO₂ incubator for 1 h to allow ECM to polymerize.
   6. After polymerization, add 200 µL of fresh culture medium to each well and fill the empty outer wells with PBS to prevent evaporation. Replace the culture medium every 2-3 days and passage the organoids weekly. Adjust the passaging interval according to the desired organoid size.
4. Drug treatment
   1. Prepare a 10 mM stock solution of cisplatin by dissolving 3.00 mg of cisplatin powder (MW ≈ 300.05 g/mol) in 1 mL of DMSO. Vortex briefly and protect the solution from light using amber tubes or by wrapping them with aluminum foil.
   2. To prepare the treatment medium, dilute the 10 mM cisplatin stock 1:1,000 in complete sIO culture medium to obtain a final concentration of 10 µM cisplatin and 0.1% DMSO.
   3. For the vehicle control, prepare 0.1% DMSO in culture medium to match the final DMSO concentration without cisplatin.
   4. Remove the spent medium from sIO cultures and gently replace it with 10 µM cisplatin-containing medium or DMSO-only control medium.
   5. After treatment, transfer the samples to the imaging system incubator.

2. Low-coherence holotomography imaging using operating software

1. Sample preparation for imaging
   1. Dispense 15 µL of organoid-ECM dome onto #1.5 coverslip-bottom imaging dish by following the procedure described in steps 1.3.1-1.3.4. Place each dome in the dish as close to the center as possible for optimal imaging.
   2. Prior to polymerization, incubate at room temperature for 1 min to allow the organoids to settle down to the bottom.
   3. Place the plate upside down in a 37 °C, 5% CO₂ incubator for 1 h to polymerize the ECM. Then, gently add enough culture medium to fully submerge the organoids.
      NOTE: Adjust the medium volume to ensure that the organoids are fully submerged according to the size of the coverslip-bottomed imaging dish.
   4. At 5 days post-passage, wash the sample 2x-3x with PBS immediately before imaging to remove debris and reduce optical artifacts.
2. Imaging setup
   1. Turn on the environmental controller to maintain sample viability. The chamber controller unit automatically adjusts the temperature to 37 °C and the CO₂ concentration to 5%.
   2. Press the Door button on the front of the instrument to open the door. Add water to form a thin layer inside the chamber well to prevent the culture media from drying out.
   3. Place the imaging dish in a proper vessel holder, insert it into the imaging chamber, and secure it using a pin to prevent movement. Make sure that the dish is firmly attached to the vessel holder without tilting.
   4. Close the door of the instrument using the Door button to block external light interference.
   5. Launch TomoStudio X software and log in. Click Start to open the main window.
   6. Click Add Project in the top-left corner and assign experimental parameters (e.g., project name, medium type, sample type, etc.). The specimen panel will be created automatically. Ensure the correct medium type is assigned for the use of the appropriate RI in the experiment.
   7. Click on the Desired well in the panel and click Create at the top to register the wells as specimens.
   8. Click ROI Setup in the top-right corner to define the region of interest (ROI) in the dish. Once all parameters are set, click Run Experiment in the bottom-right corner to open the image acquisition window.
   9. Click Load Vessel in the top-right corner. A brightfield image will appear. Adjust the Z-position using the +Z and -Z buttons to focus.
   10. In the Single imaging tab, adjust the ROI size. Capture organoids within a 160 µm x 160 µm field of view and acquire axial stacks up to 140 µm in depth. For larger organoids, acquire multiple RI tomograms and stitch images by selecting the Tile Imaging checkbox at the bottom.
   11. Navigate to the Time Lapse Imaging tab to configure long-term imaging. Set the duration and interval time as needed. In this experiment, time-lapse imaging was performed at 10 min intervals over a 24 h period, and analysis was conducted at 2 h intervals throughout the same duration.
   12. Click the Scan icon to capture the current ROI location. Manually center the ECM dome using brightfield imaging. Click the BF button to adjust the intensity and exposure values for brightfield imaging.
   13. Select ROI by moving the ROI box that appears in the Preview panel. Once the desired ROI is found, click Add Point at the bottom. The imaging point list will appear.
   14. Click Acquire to start imaging. The raw image data will be acquired.
3. Image acquisition
   NOTE: This process generates a Tomocube Common File (TCF) from raw image data. The raw image data must be processed into a TCF file, which contains multiple types of images, such as an RI tomogram and a maximum intensity projection. The RI tomogram represents the sample's 3D RI distribution, and the maximum intensity projection shows the highest RI value for each lateral position in the XY plane.
   1. Drag and drop raw image files to the HTX processing server. Click Process to generate a TCF file from the raw image file.
   2. To view the RI tomograms generated in the TCF file, use TomoAnalysis Viewer, MATLAB, or Python. To view a TCF file in MATLAB, use functions like h5info and h5read. Likewise, in Python, use the h5py package for reading H5 files.

3. Image analysis

1. Machine learning-based image segmentation
   1. Export TCF image file to HDF5 format (Supplementary Coding File 1) to ensure the data is in a multi-dimensional format compatible with ilastik.
   2. Open ilastik and navigate to File > New Project. Select Pixel Classification under the Segmentation Workflows section. Save the project file in a designated directory.
   3. Load the HDF5 file by navigating to 1. Input Data tab. Click Add New File, select the Appropriate H5 Dataset, and ensure that the correct image channels are assigned.
   4. Navigate to 2. Feature Selection tab to select relevant features (Color/Intensity, Edge,Texture) to optimize segmentation.
      NOTE: Feature selection may vary depending on the dataset and image characteristics. For the current analysis, the selected features included: Color/Intensity (σ = 1.60), Edge (σ = 3.50), and Texture (σ = 1.60). Optimization based on a heuristic approach may be required depending on image complexity. Alternatively, the Suggest Features > Run Feature Selection function in ilastik can be used to identify an optimal set of parameters.
   5. In 3. Training tab, label organoid and non-organoid regions by applying brush strokes of distinct colors. To improve segmentation fidelity, carefully annotate regions around boundaries.
   6. Click Live Update to preview segmentation results and make refinements as necessary.
   7. Navigate to 4. Prediction Export. In the Source Selection, choose Simple Segmentation to export labeled predictions. Click Export All, ensuring the output format is set to .h5 or .tiff, depending on downstream analysis requirements.
2. Quantitative feature analysis
   NOTE: This protocol step is executed in MATLAB, with each procedure detailed in Supplementary Coding File 2.
   1. By combining the RI tomogram with the mask image, create a masked RI image that includes only the organoid region.
   2. Using the masked RI image, calculate quantitative parameters such as organoid volume, protein density, and protein content.
      1. Volume: Compute the total volume of the organoid by multiplying the number of pixels in the segmented mask by the pixel resolution for the X, Y, and Z. Calculate the number of pixels by counting all voxel values in the organoid-labeled mask array, using the built-in MATLAB function Sum.
         NOTE: The pixel resolution is determined by the imaging system and can be found under Data > 3D attributes group within the HDF5 structure.
      2. Protein Density: Calculate the protein density by subtracting the medium RI value from the mean organoid RI and dividing it by the Protein Refractive Index Increment.
         1. Medium RI: Measure the RI of the surrounding medium using a refractometer prior to analysis.
         2. Mean RI Value: Determine the mean RI of the organoid by averaging the RI values of voxels located within the organoid mask.
         3. Protein Refractive Index Increment (α): Use a previously reported empirical constant derived from refractive index increment studies.
      3. Protein Content: Calculate the total protein content by multiplying the volume determined in Step 3.2.2.1 with the protein density obtained in Step 3.2.2.2.

Results

Figure 1 illustrates the schematic workflow for 4D imaging and quantitative analysis of sIOs. This protocol enables continuous live imaging and comprehensive assessment of sIOs through a four-step process: Culture, 4D Imaging, Image Processing, and Quantitative Analysis (Figure 1D).

After a 5-day incubation period, mature sIOs were transferred to coverslip-bottomed imaging dishes for visualization using low-coherence holotomography (H...

Discussion

Holotomography presents a powerful approach for label-free, high-resolution imaging of live organoids, overcoming many challenges associated with conventional imaging techniques. In this protocol, we describe a comprehensive methodology for applying low-coherence HT to sIO research, enabling real-time, quantitative visualization of organoid morphology, growth dynamics, and responses to drug treatment. The integration of machine learning-based segmentation and quantitative analysis further enhances the analytical capabili...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
48-well plate	Corning	3548	Culture dish
Cell recovery solution	Corning	354253	Matrigel degradation
Cisplatin	supelco	PHR1624	N/A
Dimethyl sulfoxide (DMSO)	Sigma-Aldrich	D2650	Vehicle
Eppendorf tube	Corning	MCT-175-C-S	N/A
HTX processing server	Tomocube Inc., Korea	N/A	Data processing server
HT-X1	Tomocube Inc., Korea	N/A	Operating Software
Ilastik	N/A	N/A	Machine learning-based segmentation toolkit
Imaging dish vessel holder	Tomocube Inc., Korea	N/A	Imaging dish holder
IntestiCult Organoid Growth Medium (Mouse)	STEMCELL Technologies	#06005	Commercially available pre-made culture medium
Matlab	MathWorks	Matlab 2024a	Software for image processing
Matrigel	Corning	356231	ECM for 3D organoids
Millipore Steritop Vacuum Bottle Top Filter	Merck	S2GPT05RE	0.22 μm filter unit for medium sterilization
Mouse Intestinal Organoids	STEMCELL Technologies	#70931	N/A
Refractometer	ATAGO	R-5000	Refractive index calculator
TomoDish	Tomocube Inc., Korea	901002-02	#1.5 coverslip-bottomed imaging dish
TomoStudio X	Tomocube Inc., Korea	N/A	Imaging software
TrypLE	Gibco	12604	Single cell dissociation