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 D

Poster F9 - OrganoSeg 2: Organoid Segmentation for Inter- and Intra-Tumor Heterogeneity Analysis

Friday, October 25, 2024
3:30 PM - 4:30 PM EST
Location: Exhibit Hall E, F & G

Introduction:  Variability in patient responses to cancer treatment poses a significant obstacle to formulate more effective treatment plans, necessitating more accurate models and a thorough understanding of the underlying biology. Organoids, which are three-dimensionally cultured cellular structures, represent one of the most physiologically accurate approaches in cancer research for studying tumor behavior and testing treatment responses ex vivo. Organoids present characteristics similar to the tissue from which they were derived, making them useful for studying heterogeneities between patients. The three-dimensional environment that organoids are cultured in allows them to mimic in vivo structural organization. However, this environment presents challenges for automated image processing, which is essential for generating data on organoid cultures. Previously, our lab developed OrganoSeg, which uses adaptive thresholding to segment organoids and calculate morphological data, and presents an easily understandable workflow. Since then, machine learning alternatives such as OrgaQuant, MOrgAna, and OrganoID have been published. While these are accurate, they show limitations with respect to runtime, morphological calculations, processing images with multiple organoids, and a reliance on manually generated training data. OrganoSeg remains a useful tool for organoid segmentation but presents several areas for improvement, including poor segmentation in unevenly lit regions, and inefficient software and post-segmentation modifications. Additionally, there are no options in the original app to match individual organoids across images or expand analysis to fluorescent images, limiting it to population level morphological data. We aimed to facilitate more efficient and accurate organoid analysis and expand the information that can be generated from organoid cultures.

Materials and

Methods:  The updated OrganoSeg GUI was built in the MATLAB App Designer environment. The segmentation pipeline uses the same adaptive thresholding technique, but includes an additional step for edge correction. The new pipeline thresholds the image gradient based on a user-defined threshold, with a correction for background intensity. It then combines this binary mask with the original segmentation and fills holes, using a watershed transformation of the original segmentation to maintain distinct organoids.

Additional parameters were incorporated to filter out non-organoid objects from segmentation. Post-segmentation modifications such as organoid splitting, combining, removing, and edge-correcting were added or updated.

Code was modified by eliminating redundant or excessive function calls and using the minimal required array size for computations. Important processes in the software were functionalized to facilitate continued development.

For organoid tracking between timepoints, images are aligned through non-rigid registration using MATLAB’s imregdemons function. Organoids are matched through distance between centroids. Organoids are iteratively matched through all previous images, so that organoids can be matched across all images even if they are absent from some. A user defined parameter determines the maximum distance allowing organoids to be matched.

For fluorescence analysis, users select fluorescent images to pair with the brightfield images, using the segmentation masks to select the regions of interest and calculate a user-specified percentile pixel intensity for each organoid. The distribution of intensity values is displayed, allowing visualization and selection of a cutoff for fluorescent marking. Tracked images can be analyzed for fluorescence.

Results, Conclusions, and Discussions: The updated OrganoSeg platform offers more flexibility for users and improves segmentation quality. Additional parameters for pixel intensity and circularity help eliminate non-organoids, such as dye, debris, and plastic-attached organoids. Individual organoid splitting and edge correction were incorporated, allowing users to modify observably incorrect organoid segmentations (multiple at the same time) without affecting accurate segmentations. The combine tool now enforces that combined organoids are represented by a connected region, ensuring accurate morphological calculations. Incorporating these modifications enhances the segmentation quality, allowing users to efficiently adjust and verify the results for a more accurate representation of the true morphological data (Figure 1).

Organoids in dimly lit regions that previously would not be fully segmented are more accurately represented. Well-lit regions with accurate segmentations are resistant to over segmentation and have boundaries retained (Figure 2). This adjustment makes organoid imaging more generalizable across brightfield images, regardless of lighting quality.

The runtimes of essential processes in the software were significantly reduced. Across 78 images under the same settings, segmentation time was reduced by 51% on average (0.89 to 0.43 seconds per image) and exporting time to an Excel sheet was reduced by 95% on average (4.6 to 0.21 seconds per image).

Organoid tracking enables the analysis of changes in organoid behavior at the individual or subpopulation level. While non-rigid registration using segmentation masks is effective in most images, it requires improvement for images with low organoid density. Despite this, organoids are successfully tracked across images (Figure 3). Fluorescence analysis integrates molecular information, such as apoptosis level, with morphological data. Fluorescence analysis on tracked images allows users to correlate morphological and molecular changes longitudinally. Tracking individual organoids enhances the ability to detect inter-tumor heterogeneities, and facilitates the observation of intra-tumor variations.

OrganoSeg will continue to be developed to address limitations. Currently, organoid splitting relies on geometric transformations which are less effective on irregularly shaped organoids. Splitting based on changes in intensity or texture would be more generalizable. Additional options for localizing organoid analysis would be beneficial, such as allowing users to sort images into experimental groups and visualize differences.

Acknowledgements (Optional):  Thank you to Dr. Cameron Griffiths, Dr. Bishal Paudel, and Dr. Piotr Przanowski for their suggestions and guidance throughout this project

Thank you to Dr. Lakeshia Taite and the UVA Dean’s Undergraduate Engineering Summer Fellowship Program for supporting this research