Skip to main content
NCBI home page
ACS Omega logoLink to ACS Omega
. 2025 Aug 20;10(34):39203–39211. doi: 10.1021/acsomega.5c06107

Ultrasound Imaging and Machine Learning for Nondestructive Sensing in Bioreactors

Mary Serpe 1, Caleb Lee 1, Reid Povinelli 1, Richard Thyden 1, Luke Perreault 1, Keshi He 1, Glenn R Gaudette 1, Antonio C F dos Santos 1, Bryan J Ranger 1,*
PMCID: PMC12409521  PMID: 40918368

Abstract

The advancement of cell therapy and cellular agriculture underscores the need for noninvasive, cost-effective methods for continuous monitoring of large-scale cell production. Bioreactors, designed to mimic physiological conditions to facilitate cell growth, require reliable quality control measures. This study investigates the potential of ultrasound technology to characterize cellular growth and decellularization in spinach scaffolds. Ultrasound images were acquired of fresh spinach leaves, decellularized leaves, and leaves with cells seeded in defined cloning well regions on decellularized leaves. Machine learning tools were employed to assist in the identification and segmentation of images as part of ultrasound data analysis. Statistical analysis revealed significant differences in mean pixel intensity values between fresh and decellularized leaves (p < 0.001) as well as between regions with and without cells seeded on decellularized leaves (p = 0.001). These findings were confirmed by histological analysis. The results indicate that ultrasound imaging could serve as a viable, nondestructive tool for quality assessment of plant scaffolds in cellular agriculture applications.

graphic file with name ao5c06107_0009.jpg

graphic file with name ao5c06107_0007.jpg

1. Introduction

Bioreactors are vessels designed to support cell growth, which are likely to play an essential role in bringing tissue engineered constructs to the clinic. These systems create an environment conducive to cell proliferation through the careful regulation of media contents, physical stimulus, and other parameters within the bioreactor. They are particularly important in the pharmaceutical and food production industries, and have become a standard in the field of biomanufacturing. Additionally, bioreactors have applications in tissue engineering, as they can simulate various physiological conditions that enhance cell proliferation and performance. They ensure cell survival throughout tissue-engineered constructs by facilitating the delivery of nutrients in a three-dimensional space, while the conditions and structure of the bioreactor can influence tissue structure and proliferation. For bioreactors to be effectively utilized in commercial tissue production, continuous monitoring of bioreactor products throughout the cell growth process is essential. Such monitoring could also help address potential challenges posed by FDA requirements. An ideal monitoring method should not interfere with cell growth and should minimize the need for sampling.

Most mammalian cells require not only a medium in which to grow, but also a base to support their growth. In laboratory and industrial settings, cell culture-treated plastic and scaffolds are used for cell adsorption and cell growth. Recently, decellularized plant scaffolds for cell culture have emerged as a sustainable and edible alternative to synthetic scaffolds. , These decellularized plant-based biomaterials offer a valuable scaffold for cell culture, providing an accessible and cost-effective solution that addresses the challenges of material availability traditionally associated with tissue engineering. Decellularized plant scaffolds offer a novel bioengineered foundation for cell culture. Specifically, the use of the plant’s preexisting vascular structure has been shown to improve cell differentiation and proliferation. Decellularized spinach, in particular, serves as a suitable scaffold for numerous cell types, with the potential to advance lab-grown meat technologies, reduce the environmental impacts of mass farming, and contribute to addressing the global hunger crisis.

Decellularization involves removing cellular contents, such as DNA, from tissues to leave behind an extracellular matrix (ECM) scaffold. The ECM has numerous biomedical applications, including potential use in lab-grown meat, which could help mitigate the environmental impact of traditional meat production. However, the decellularization of plant scaffolds is not yet widely used in cell culture. Key challenges include standardizing the decellularization process and accurately measuring the remaining plant cells. Current methods, such as CyQUANT (Thermo Fisher, Waltham, MA), can quantify cell proliferation, providing a potential solution to understanding the cellular contents of decellularized scaffolds. However, these methods typically require the removal and destruction of scaffold samples with accurate cell counts, a process that is both costly and time-consuming. While bioreactors can utilize current methods to monitor the formation of plant scaffolds for cell growth, there is a need for nondestructive, automatic monitoring of the scaffolds that reduce system stress and overall costs.

Ultrasound imaging is widely used across various fields to visualize, test, and measure parameters that are often difficult to assess with the naked eye. It is particularly notable for being a nonintrusive and nondestructive technique that can analyze and characterize a diverse range of substances, including inorganic materials such as steel and concrete, biological human tissues and bones, and tissue-engineered constructs such as heart valves, cartilage, and scaffolds. For example, ultrasound can diagnose bone fragility by evaluating the attenuation of ultrasonic waves as they propagate through the bone. It can also monitor fatigue microdamage using nonlinear resonant ultrasound spectroscopy (NRUS), or more broadly assess material properties and detect defects using quantitative laser ultrasound visualization (QLUV). These techniques are particularly valuable since they are noninvasive and nondestructive, ensuring that the bone is not damaged during an examination. Ultrasound imaging offers several advantages over other modalities, including rapid generation of images, cost-effectiveness, and high precision. In addition, ultrasound is highly effective for monitoring both inorganic and biological materials without making direct contact, thereby preventing sample destruction and providing opportunities for noninvasiveness.

Ultrasound has proven effective in bioprocess monitoring and control, including chemical analysis. Tissue engineering and biomaterials are particularly relevant fields for ultrasound monitoring, since traditional methods are often “destructive, time-and-cost inefficient, and do not allow time-lapse measurements.” In contrast, ultrasound offers rapid, noninvasive, and nondestructive evaluation and it can be easily integrated into sterile environments required for tissue engineering. , Recent advances in machine learning have expanded the analytical capabilities of ultrasound imaging by enabling automated and semiautomated image segmentation, feature extraction, and pattern recognition. Such approaches can facilitate more objective and reproducible interpretation of image data.

Our focus on the use of ultrasound imaging for the nondestructive characterization of plant and tissue-engineered scaffolds aligns closely with bioreactor monitoring and scaffold characterization. Ultrasound could be applied at various stages along the bioreactor process, including tracking cell concentration, verifying decellularization in plant scaffolds, and quantifying cell proliferation. Other studies have discussed the challenges of noninvasive monitoring cell proliferation, highlighting the potential of ultrasound as a viable solution. ,

The purpose of this study is to use ultrasound technology and machine learning tools to quantify cellular presence in plant scaffolds during both decellularization and cell seeding. Our methodology involves using the mean pixel intensity of the ultrasound images of the scaffold to quantify cellular levels. This paper tracks the scaffolds through the decellularization process, sterilization, and cell seeding, with ultrasound images acquired at the conclusion of each stage. To validate the ultrasound images, histology is used in conjunction with the ultrasound images to confirm cell growth. The mean pixel intensity at each stage (fresh, decellularized, recellularized within a cloning well) is analyzed using ANOVA to determine whether these values differ significantly. The overarching objective of this study is to explore ultrasound technology as a nondestructive, efficient alternative for quantifying cellular levels in plant-based scaffolds.

2. Materials and Methods

2.1. Decellularization Process

Spinach leaves (Nature’s Promise Organic, Landover, MD, USA) were cut into 10 rectangles of 1 cm × 3 cm to fit the ultrasound plate appropriately and allow space for cloning well attachment. The samples were then imaged using ultrasound according to the procedures described below on Day 0, as seen in Figure A. The spinach leaves were subsequently placed in 40 mL of 1% sodium dodecyl sulfate (SDS) solution (Fisher Scientific, Pittsburgh PA) in 50 mL conical tubes and continuously agitated on an analog tube roller (Scilogex, Rocky Hill, CT, USA) for 5 days. The SDS solution was replaced every 24 h with 40 mL of new 1% SDS solution. After 5 days of 1% SDS solution treatment, the spinach leaves were thoroughly rinsed with deionized (DI) water. The spinach leaves were then immersed in 40 mL of 1% Tween 20 (Sigma-Aldrich, St. Louis, MO, USA), 3% concentrated bleach solution (The Clorox Co., Oakland, CA, USA) for 2 days. The Tween/bleach solution was replaced every 24 h to ensure the solution remained fresh. , On Day 7, following 2 days of Tween/bleach treatment, the spinach leaves were imaged a second time using the ultrasound procedure, as shown in Figure B.

1.

1

Open in a new tab

(A) Fresh spinach leaf sample, (B) ultrasound image of fresh spinach leaf, (C) spinach leaf sample after decellularization protocol, (D) ultrasound probe images of full decellularized spinach scaffold, (E) recellularized spinach leaf with NIH 3T33 cells seeded on cloning well regions, and (F) ultrasound probe images of spinach leaf with NIH 3T3 cells seeded.

2.2. Recellularization Process

After decellularization, the leaf scaffold samples (n = 10) were seeded with mouse embryonic fibroblast cells (NIH/3T3), acquired from ATCC (American Type Culture Collection). These cells were chosen for their rapid growth ability to adhere and proliferate on decellularized plant scaffolds.

Before seeding cells onto the decellularized scaffolds, the leaves were sterilized. The sterilization process involved placing the scaffolds in a 6-well plate (plate model) and soaking them in 70% ethanol solution for 30 min. After the ethanol treatment, the leaves were rinsed three times with sterile PBS for 5 min each. Following sterilization, the scaffolds were soaked in growth media (Ham’s DMEM/F12, 10% heat-inactivated fetal bovine serum (FBS) (Thermo Fisher Scientific, Waltham, MA, USA), 1% P/S, 4 ng/mL recombinant human fibroblast growth factor-2 (FGF2), 2.5 ng/mL recombinant human hepatocyte growth factor (HGF), 10 ng/mL recombinant human epidermal growth factor (EGF), and 5 ng/mL recombinant human insulin-like growth factor-1 (IGF)) (Thermo Fisher, Waltham, MA, USA) for 1 h. After 1 h, the media was aspirated from the wells.

To create distinct cell seeded areas on the leaves, two 10 mm diameter (8 mm inner diameter) sterilized cloning wells (Corning Life Sciences, Tewksbury, MA, USA) were placed onto the scaffolds using forceps. The cloning wells were gently positioned on the scaffold and then were tapped down with the forceps to create a seal. Two cloning wells were added to each of the sterilized scaffolds. One well was seeded with 500,000 cells, while only growth media was added to the other well. A hemocytometer (Hausser Scientific, Horsham, PA, USA) was used to estimate the total cells present in the cell solution. Excess cell culture media was added to each of the wells in the plate to ensure cell growth. The seeded scaffolds were incubated at 37 °C with 5% CO2 for 48 h.

At the end of the 2-day growth period, the media in the wells was aspirated, and recellularized scaffolds were imaged using the ultrasound procedure described in Section After imaging, macroscopic images were taken (Figure C), and the samples were returned to the empty wells in preparation for fixing, staining, and additional imaging.

2.3. Ultrasound Method

Images were captured using a Clarius L20 ultrasound probe (Clarius, Vancouver, BC, Canada). The ultrasound probe operates at a frequency up to 20 MHz and is specifically designed for high-resolution imaging of superficial structures. In this frequency range, the axial resolution is approximately 0.1 mm and the lateral resolution is around 0.2 mm. These resolution values are well suited for imaging of thin biological samples, such as the spinach leaves used in our study which have a typical thickness ranging from 0.2 to 0.6 mm. Given the resolution of the ultrasound transducer is finer than the thickness of a typical spinach leaf, it is sufficient to detect and resolve relevant structural features of the spinach leaves. The musculoskeletal (MSK) setting was selected as it provided clear images of the spinach leaf while effectively reducing background noise.

To minimize variability between images and ensure the probe did not directly make contact with the sample, the Clarius L20 ultrasound probe was mounted 3.5 cm above the benchtop and submerged approximately 1 mm into a clear basin containing 290 mL of 1% phosphate-buffered saline (PBS) solution diluted from 10% phosphate-buffered saline (PBS) solution (Fisher Scientific, Pittsburgh, PA, USA), positioned 0.5 cm above the samples, as shown in Figure . Images were captured every 5 s. The images were then exported through the Clarius cloud (Clarius, Vancouver, BC, Canada) and stored as DICOM files for subsequent image analysis. Ultrasound images along the midline of scaffold (prerecellularization) and the midline of the recellularized area (postrecellularization) were selected for further analysis. One representative cross-section ultrasound image of the same spinach scaffolds as it progressed from the fresh leaf stage to the decellularized stage, and finally to the recellularized stage are shown in Figure B,D, and F, respectively.

2.

2

Open in a new tab

(A) Schematic of ultrasound probe setup. Clarius L20 ultrasound probe mounted above the basin of PBS with samples clamped in place. (B) Image of ultrasound setup: Clarius L20 ultrasound probe mounted to ring clamp submerged in PBS solution to image sample.

2.4. Ultrasound Image Processing

The ultrasound image processing is summarized in Figure . All images were individually processed using ImageJ, further data extraction and analysis was done using a custom algorithm in Matlab R2023b (Mathworks Inc., Natick, MA). DICOM images were imported into ImageJ, cropped to focus on the region surrounding the leaf, and converted into PNG files (Figure A). Each image (n = 10) was subsequently analyzed using the Trainable Weka Segmentation tool, an ImageJ plugin that combines machine learning algorithms with a set of selected image features to produce pixel-based segmentations. The segmented images were manually classified into three groups: leaf, background, and cloning well. Figure B illustrates the manual classification with the background in red, the leaf is green, and the cloning well is purple. The individual classification was then automated using the same tool. Adjustments were made, if needed, to the automated classifier on a leaf-to-leaf basis up to three times. After classifying all images, a combined automated classifier was trained using data of all leaves (n = 10) using the Fast Random Forest classifier method. This approach utilizes features such as Gaussian blur, Sobel filter, Hessian, difference of gaussians and membrane projections of the identified classified regions. Figure C shows the result of applying the combined trained classifier, with the background in black, the leaf in white and the cloning well as a gap within the leaf.

3.

3

Open in a new tab

Image processing sequence example: (A) Unprocessed spinach leaf ultrasound image, (B) manually classified spinach leaf ultrasound image, (C) segmented image of a spinach leaf ultrasound image, analyzed by the trained classifier, leaf segment probability shown, (D) processed leaf segment probability map, (E) segmented image of the spinach leaf ultrasound image, analyzed by the trained classifier, cloning well segment probability show, (F) processed cloning well segment probability map, and (G) unprocessed spinach leaf ultrasound image overlaid with classification selection.

The images were then converted into binary matrices based on the probability map for each segment (background, leaf and cloning well). The matrices are then smoothed using a rolling matrix smoothing algorithm to remove noncontiguous regions, eliminating any remaining misclassified pixels. Figure D shows leaf segment probability matrix and Figure F shows the same leaf’s cloning well probability matrix. The matrix was then used to generate traces that delineated the different regions. Figure G shows the region traces overlaid with the original image. Pixel intensity in each region was then analyzed.

2.5. Histology

To confirm that recellularization could be visualized in the ultrasound images, the recellularized scaffolds were stained and imaged. First, the recellularized scaffolds were fixed in 2 mL of 4% paraformaldehyde (Sigma-Aldrich, St. Louis, MO, USA) for 10 min. After fixation, the scaffolds were rinsed twice with 10 mL of 1× PBS. Next, the scaffolds were placed in 2 mL of a 0.25% Triton X-100 (Sigma-Aldrich, St. Louis, MO, USA) solution for 10 min, followed by two additional rinses with 1× PBS. The scaffolds were then blocked at 4 °C in 2 mL of 5% goat serum (Gibco, Grand Island, NY, USA) in 1% BSA (Fisher Scientific, Pittsburgh, PA, USA) solution for 30 min.

The blocking solution was aspirated, and then the scaffolds were soaked in 2 mL of 0.25% Alexa Fluor 488 Phalloidin A12379 (Invitrogen, Waltham, MA, USA) and 0.0167% Hoechst dye (Invitrogen, Waltham, MA) with 1% BSA in 1× PBS solution for 60 min in the dark. Following the staining, the scaffolds were rinsed three times again with 2 mL of 1× PBS in the dark. The stained scaffolds were stored in the dark in a 6-well plate in 1× PBS at 4 °C. Finally, the scaffolds were imaged using the BioTek Cytation 1 (Agilent Technologies, Inc., Santa Clara, CA).

2.6. Histological Analysis

Histology images of each sample were exported as PNG files and imported into ImageJ. The images were manually rotated and cropped to align with the horizontal orientation of the ultrasound images. The images were then split into individual channels (red, green, and blue). The green channel corresponded to the Phalloidin stain fluorescence, while the blue stain represented the Hoechst stain fluorescence. The red channel was discarded. Each new image was saved as a TIFF file for further analysis.

The results of the ultrasound classification were compared to histology images of the corresponding samples. For each classified ultrasound image, the average pixel intensity of nonbackground regions along the width of the leaf was calculated. The background threshold, previously identified with the Weka training tool, was applied to isolate the sample pixels. For the histology images, the blue channel was used, and pixel intensities along the width of the leaf were averaged across all pixels. To enable direct comparison between the two image types, the center of each image was aligned.

For the ultrasound images, data located more than 15 mm from the center of the leaf were excluded from analysis. These peripheral regions are larger than the sample areas and correspond to the clamps used to hold the leaves during the ultrasound imaging process. Similarly, the histology images were cropped to exclude any data beyond 13 mm away from the center of the leaf. This threshold was chosen to avoid any potential interference with the well walls, which would affect image quality and introduce additional noise into the analysis. By limiting both types of images to these defined regions, we ensured that only relevant portions of the samples were considered for comparison.

2.7. Statistics

ANOVA statistical analysis was conducted on the mean pixel intensity values for the different stages of spinach scaffolds (fresh, decellularized, recellularized) using Prism (GraphPad, Boston, MA). A p-value threshold of 0.05 was considered statistically significant. Unless specified otherwise, all data is expressed as mean ± standard deviation.

3. Results

3.1. Comparison of Fresh and Decellularized Spinach Scaffold Ultrasound Images

Analysis of the ultrasound images of the n = 10 leaf samples before and after decellularization revealed statistically significant differences in pixel intensities (p < 0.001), as shown in Figure (first two columns). The fresh leaves exhibited an average pixel intensity of 232.25 ± 3.57, and the decellularized leaves had an average intensity of 86.13 ± 16.63.

4.

4

Open in a new tab

Comparison of mean pixel intensity value of the n = 10 leaf samples and corresponding regions throughout the process. All images were obtained using ultrasound imaging of the sample midline. Fresh and decellularized leaves were analyzed as a whole while recellularized leaves were segmented into cloning well regions (with cells) and noncloning well regions. * = p < 0.05.

3.2. Cloning Well Cell Region Compared to Decellularized Portion of Scaffold

The same leaf segments were then analyzed after recellularization with NIH 3T3. The average pixel intensity of the cell clusters within the cloning well was significantly higher (121.46 ± 28.82) compared to the decellularized regions of the scaffold (81.95 ± 13.02), with a p-value of 0.001, as shown in Figure . Additionally, the mean pixel intensity values of the decellularized scaffolds versus the recellularized scaffolds outside the cloning well cell seeded areas were not statistically significant, indicating no change due to the incubation process, growth media and temperature changes. This result confirms the absence of cells outside of the cloning wells.

5.

5

Open in a new tab

Example average pixel intensities comparison between ultrasound and histology data. (A) True positive example. Red circle indicates the region that was inoculated and cells identified. (B) False positive example compared to the histology image’s blue channel. The red arrow identifies the origin of the false positive signal. (C) Same false positive example compared to the histology image’s green channel. The red arrow identifies the origin of the false positive signal. Black line: Ultrasound data; Blue line: histology data using only blue channels; Green line: histology data using only green channels.

3.3. Orthogonal Confirmation of Ultrasound Classification

The presence of living adhered cells were confirmed in each scaffold using histology and visual inspection. The images showed that cell presence was limited to the region corresponding to the inoculated well location. Cells were present in 7 samples. No cells were present in 2 samples. One sample could not be imaged and removed from analysis pertaining to the recellularization process.

Ultrasound and histology data were compared using the average pixel intensity of each column along the x-axis of the images. The locations of high pixel intensities which indicate cell presence were found to be located in the same regions of the scaffold when blue channels of histology images (corresponding to nuclei stained with Hoestch) and ultrasound images were compared. The ultrasound monitoring method identified the recellularized region in 7 samples along with 2 false positives. An example of a true positive is shown in Figure A. These results indicate that the ultrasound images, in combination with a trained machine learning-based image segmentation tool, are effective at identifying regions of cellular growth on decellularized spinach leaf scaffolds. Figure B shows an example of an ultrasound false positive. The peak at approximately 6 mm away from the center of the scaffold has no corresponding peak in the blue channel curve. The peak corresponds to a secondary vein of the spinach leaf present on the scaffold (Figure B, red arrow). A similar peak is seen on the green channel of the histology image (Figure C).

4. Discussion

With FDA approval of cell therapy, the use of bioreactor systems is likely to continue to grow. While the bioreactor field has matured regarding production of biotherapeutics, it is still developing for production of cells and tissues. While both therapies use bioreactors, the biotherapeutics product is produced by cells, whereas in cell/tissue therapy cells and their substrate are the products, requiring different monitoring systems.

Studies have similarly focused on continuous monitoring of extracellular matrix scaffold development. Kim and Wagner summarized works focused on scaffolds prepared by decellularization protocols for recellularization. These studies agree that traditional biological assays and histology are challenging to implement due to their time-consuming nature and destructive impact, while ultrasound offers a more ideal solution. Although the tissues being decellularized, the extracellular matrix scaffolds, the characteristics being monitored, and the recellularization purposes may differ, these findings align with the goals of this study.

Our results, as a proof-of-concept, demonstrate that ultrasound technology is a promising tool for monitoring cell presence on decellularized plant scaffolds. Significant differences in mean pixel intensity were observed between fresh and decellularized leaves. This drop in intensity likely reflects the reduction in leaf density as plant cells are removed during decellularization, leaving behind the cellulose scaffold. This outcome is consistent with previous studies describing the decellularization process and supports the potential of ultrasound to quantitatively track cellular changes in plant-based scaffolds. Ultrasound applicability and accuracy on other scaffolds currently used in cell/tissue production remains to be determined.

Further analysis of the plant scaffold by region revealed that recellularized areas had significantly higher mean pixel intensity compared to the surrounding decellularized regions. This result aligns with expectations, as the density of the scaffold should increase only within the confined area of the cloning well, where cells are present. The surrounding decellularized area, devoid of cells, exhibited pixel intensities similar to those of the decellularized leaf. Naturally occurring high density regions can lead to false positives results. Image screening and more comprehensive analysis of the whole scaffold could diminish the effects of such disturbances.

The use of decellularized plant-based scaffolds has shown great promise in the rapidly evolving field of cellular agriculture. These scaffolds have diverse applications, ranging from organ regeneration to lab grown meat production. However, to meet the growing demand in cellular agriculture, there is a need to scale up production methods to make this process more cost-effective and widely applicable. A potential solution to this challenge is the use of bioreactors, which can simulate physiological conditions conducive to cellular proliferation on decellularized scaffolds, thereby facilitating large-scale cellular production. Despite this progress, the decellularization and cellular production processes require continuous monitoring to ensure scaffold quality and safety. Currently, existing methods for measuring cellular levels are typically destructive, which poses a barrier to efficient, noninvasive monitoring. The overarching aim of this study was to explore ultrasound technology as a potential solution for continuous, nondestructive monitoring of the decellularization and recellularization of scaffolds, as depicted in Figure .

6.

6

Open in a new tab

Schematic representation of ultrasound probe continuously monitoring the decellularization and recellularization of spinach scaffolds.

Ultrasound technology offers several advantages over traditional methods of analysis, such as CyQUANT DNA assays and histology. The CyQUANT assay requires the destruction of the scaffold to release DNA content before analysis. Histology requires long incubation times and cell death. Whereas ultrasound provides a nondestructive alternative that allows for repeated measurements without compromising sample integrity. Additionally, ultrasound does not require single-use materials, which reduces costs and waste. Spectroscopy has been another promising alternative for in-line monitoring of decellularization, but this process relies on the analysis of the effluent as an indirect measurement of decellularization.

Machine learning demonstrates significant potential for classifying the decellularization process and subsequent cellular presence on scaffolds, as well as performing efficient quality control checks of scaffolds in bioreactors without the need to remove scaffold samples for testing. In this study, machine learning techniques were utilized as a supportive tool to assist in identifying and localizing cell presence on recellularized spinach leaf samples. This approach illustrates the ability to monitor decellularization, cell adhesion, and cell proliferation in real time. Similar machine learning techniques have been successfully utilized to classify cell lines and cell states from image data, and deep convolutional neural networks have been employed to classify decellularization states based on spectroscopy measurements. By integrating machine learning with ultrasound technology, quality control checks could be made not only nondestructive, but also automated, allowing for continuous monitoring of the scaffolds within bioreactors. Numerous studies have demonstrated the promising potential of machine learning for ultrasound image analysis. These advances highlight the capacity of machine learning to enhance both the efficiency and accuracy of quality control in cellular agriculture applications.

The ultrasound technology in this application still needs to be refined. One challenge is the presence of background noise in the images, which could potentially influence pixel intensity measurements. Moreover, bubbles that form around the scaffold following decellularization may interfere with the image quality and affect the mean pixel intensity value. Despite these challenges, our results support the initial hypothesis that ultrasound images of the scaffold exhibit significant differences at various stages of the decellularization and recellularization process. These findings highlight the potential of ultrasound probes for continuous monitoring in bioreactors, where they could be used to track scaffold development in real time.

In future work, we aim to expand this approach to include a wider variety of scaffold types, such as animal-derived extracellular matrix (ECM) materials, and explore the feasibility of monitoring scaffolds in motion or tracking multiple scaffolds simultaneously within bioreactors. Additionally, we plan to develop a fully automated system for analyzing ultrasound images, which would eliminate the need for manual input, as was required in this study. Further research will also focus on optimizing the method’s sensitivity. Factors such as ultrasound depth and frequency may impact background noise, amplitude and throughput, and these aspects will be explored in greater detail. While a commercial probe was used in this study, the development of a specialized instrument specifically tailored for continuous monitoring of decellularization and cell growth may be needed. Further investigation is needed to better understand the potential effects of ultrasound waves on scaffold properties and cellular growth rates, ensuring that ultrasound interference does not adversely affect the integrity or growth of the scaffolds.

5. Conclusion

Ultrasound analysis effectively distinguishes between natural leaves, decellularized leaves and recellularized regions on plant scaffolds, demonstrating its potential as a powerful, noninvasive tool for monitoring cellular growth. Ultrasound technology shows considerable promise as a nondestructive, real-time monitoring tool for decellularized plant scaffolds used in cellular agriculture and cell/tissue engineering. Further research is needed to integrate ultrasound technology with bioreactor systems and refine image analysis techniques. These advancements will be key for optimizing this approach and enabling its successful application in large-scale cellular agriculture processes.

Acknowledgments

This research was supported in part by the Boston College Undergraduate Research Fellowship (URF) Program, which provided funding for the participating undergraduate coauthors. We also thank members of our broader research teams for their valuable input and support throughout this project.

The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript.

Boston College Undergraduate Research Fellowship Program.

The authors declare no competing financial interest.

References

  1. Bioreactor - an overview | ScienceDirect Topics. https://www.sciencedirect.com/topics/immunology-and-microbiology/bioreactor (Accessed 10 July 2024).
  2. Mitra S., Murthy G. S.. Bioreactor Control Systems in the Biopharmaceutical Industry: A Critical Perspective. Syst. Microbiol. BioManuf. 2022;2(1):91–112. doi: 10.1007/s43393-021-00048-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Selden C., Fuller B.. Role of Bioreactor Technology in Tissue Engineering for Clinical Use and Therapeutic Target Design. Bioengineering. 2018;5(2):32. doi: 10.3390/bioengineering5020032. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Blose, K. J. ; Krawiec, J. T. ; Weinbaum, J. S. ; Vorp, D. A. . Chapter 13 - Bioreactors for Tissue Engineering Purposes. In Regenerative Medicine Applications in Organ Transplantation; Academic Press; Boston: 2014. pp. 177–185. DOI: 10.1016/B978-0-12-398523-1.00013-6 [DOI] [Google Scholar]
  5. Hansmann J., Groeber F., Kahlig A., Kleinhans C., Walles H.. Bioreactors in Tissue EngineeringPrinciples, Applications and Commercial Constraints. Biotechnol. J. 2013;8(3):298–307. doi: 10.1002/biot.201200162. [DOI] [PubMed] [Google Scholar]
  6. Lourenço N. D., Lopes J. A., Almeida C. F., Sarraguça M. C., Pinheiro H. M.. Bioreactor Monitoring with Spectroscopy and Chemometrics: A Review. Anal. Bioanal. Chem. 2012;404(4):1211–1237. doi: 10.1007/s00216-012-6073-9. [DOI] [PubMed] [Google Scholar]
  7. Gershlak J. R., Hernandez S., Fontana G., Perreault L. R., Hansen K. J., Larson S. A., Binder B. Y. K., Dolivo D. M., Yang T., Dominko T., Rolle M. W., Weathers P. J., Medina-Bolivar F., Cramer C. L., Murphy W. L., Gaudette G. R.. Crossing Kingdoms: Using Decellularized Plants as Perfusable Tissue Engineering Scaffolds. Biomaterials. 2017;125:13–22. doi: 10.1016/j.biomaterials.2017.02.011. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Modulevsky D. J., Lefebvre C., Haase K., Al-Rekabi Z., Pelling A. E.. Apple Derived Cellulose Scaffolds for 3D Mammalian Cell Culture. PLoS One. 2014;9(5):e97835. doi: 10.1371/journal.pone.0097835. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Harris A. F., Lacombe J., Zenhausern F.. The Emerging Role of Decellularized Plant-Based Scaffolds as a New Biomaterial. Int. J. Mol. Sci. 2021;22(22):12347. doi: 10.3390/ijms222212347. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Jones J. D., Rebello A. S., Gaudette G. R.. Decellularized Spinach: An Edible Scaffold for Laboratory-Grown Meat. Food Biosci. 2021;41:100986. doi: 10.1016/j.fbio.2021.100986. [DOI] [Google Scholar]
  11. Saldin L. T., Cramer M. C., Velankar S. S., White L. J., Badylak S. F.. Extracellular Matrix Hydrogels from Decellularized Tissues: Structure and Function. Acta Biomater. 2017;49:1–15. doi: 10.1016/j.actbio.2016.11.068. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Ahmad K., Lim J.-H., Lee E.-J., Chun H.-J., Ali S., Ahmad S. S., Shaikh S., Choi I.. Extracellular Matrix and the Production of Cultured Meat. Foods. 2021;10(12):3116. doi: 10.3390/foods10123116. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Zhu Y., Zhang Q., Wang S., Zhang J., Fan S., Lin X.. Current Advances in the Development of Decellularized Plant Extracellular Matrix. Front. Bioeng. Biotechnol. 2021;9:9. doi: 10.3389/fbioe.2021.712262. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Dhandapani V., Vermette P.. Decellularized Bladder as Scaffold to Support Proliferation and Functionality of Insulin-Secreting Pancreatic Cells. J. Biomed Mater. Res. B Appl. Biomater. 2023;111(11):1890–1902. doi: 10.1002/jbm.b.35292. [DOI] [PubMed] [Google Scholar]
  15. Dalecki D.. Mechanical Bioeffects of Ultrasound. Annu. Rev. Biomed. Eng. 2004;6:229–248. doi: 10.1146/annurev.bioeng.6.040803.140126. [DOI] [PubMed] [Google Scholar]
  16. Haupert S., Guerard S., Peyrin F., Mitton D., Laugier P.. Non Destructive Characterization of Cortical Bone Micro-Damage by Nonlinear Resonant Ultrasound Spectroscopy: E83599. PLoS One. 2014;9(1):e83599. doi: 10.1371/journal.pone.0083599. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Yang O., Qadan M., Ierapetritou M.. Economic Analysis of Batch and Continuous Biopharmaceutical Antibody Production: A Review. J. Pharm. Innov. 2020;15:182–200. doi: 10.1007/s12247-018-09370-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Chung C.-Y., Heebner J., Baskaran H., Welter J. F., Mansour J. M.. Ultrasound Elastography for Estimation of Regional Strain of Multilayered Hydrogels and Tissue-Engineered Cartilage. Ann. Biomed. Eng. 2015;43(12):2991–3003. doi: 10.1007/s10439-015-1356-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Fatakdawala, H. ; Griffiths, L. G. ; Wong, M. L. ; Julien, M. ; Marcu, L. . Multimodal Imaging of Vascular Grafts Using Time-Resolved Fluorescence and Ultrasound. In Photonic Therapeutics and Diagnostics XI; SPIE, 2015, Vol. 9303, pp. 370–376. 10.1117/12.2078278. [DOI] [Google Scholar]
  20. Dutta, D. ; Lee, K. W. ; Allen, R. ; Yu, J. ; Wang, Y. ; Kim, K. . Non-Invasive Assessment of Mechanical Property Changes with Cell Growth in a Polymer-Based Vascular Graft Using Ultrasound Elasticity Imaging. In 2011 IEEE International Ultrasonics Symposium; IEEE, 2011; pp 2245–2248. DOI: 10.1109/ULTSYM.2011.0557. [DOI] [Google Scholar]
  21. Sebastian J. A., Strohm E. M., Baranger J., Villemain O., Kolios M. C., Simmons C. A.. Assessing Engineered Tissues and Biomaterials Using Ultrasound Imaging: In Vitro and in Vivo Applications. Biomaterials. 2023;296:122054. doi: 10.1016/j.biomaterials.2023.122054. [DOI] [PubMed] [Google Scholar]
  22. Brattain L. J., Telfer B. A., Dhyani M., Grajo J. R., Samir A. E.. Machine Learning for Medical Ultrasound: Status, Methods, and Future Opportunities. Abdom. Radiol. 2018;43(4):786–799. doi: 10.1007/s00261-018-1517-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Kreitz S., Dohmen G., Hasken S., Schmitz-Rode T., Mela P., Jockenhoevel S.. Nondestructive Method to Evaluate the Collagen Content of Fibrin-Based Tissue Engineered Structures Via Ultrasound. Tissue Eng., Part C. 2011;17(10):1021–1026. doi: 10.1089/ten.tec.2010.0669. [DOI] [PubMed] [Google Scholar]
  24. Akbari S., Anderson P., Zang H., Ganjian A., Balke R., Kwon T., Pollard D.. Non-Invasive Real-Time Monitoring of Cell Concentration and Viability Using Doppler Ultrasound. SLAS Technol. 2022;27(6):368–375. doi: 10.1016/j.slast.2022.09.003. [DOI] [PubMed] [Google Scholar]
  25. Mercado K. P., Helguera M., Hocking D. C., Dalecki D.. Estimating Cell Concentration in Three-Dimensional Engineered Tissues Using High Frequency Quantitative Ultrasound. Ann. Biomed. Eng. 2014;42(6):1292–1304. doi: 10.1007/s10439-014-0994-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Jones J. D., Thyden R., Perreault L. R., Varieur B. M., Patmanidis A. A., Daley L., Gaudette G. R., Dominko T.. Decellularization: Leveraging a Tissue Engineering Technique for Food Production. ACS Biomater. Sci. Eng. 2023;9(5):2292–2300. doi: 10.1021/acsbiomaterials.2c01421. [DOI] [PubMed] [Google Scholar]
  27. Jing L., Sun J., Liu H., Wang X., Huang D.. Using Plant Proteins to Develop Composite Scaffolds for Cell Culture Applications. Int. J. Bioprinting. 2021;7(1):298. doi: 10.18063/ijb.v7i1.298. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Schneider C. A., Rasband W. S., Eliceiri K. W.. NIH Image to ImageJ: 25 Years of Image Analysis. Nat. Methods. 2012;9(7):671–675. doi: 10.1038/nmeth.2089. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Arganda-Carreras I., Kaynig V., Rueden C., Eliceiri K. W., Schindelin J., Cardona A., Sebastian Seung H.. Trainable Weka Segmentation: A Machine Learning Tool for Microscopy Pixel Classification. Bioinformatics. 2017;33(15):2424–2426. doi: 10.1093/bioinformatics/btx180. [DOI] [PubMed] [Google Scholar]
  30. Kim K., Wagner W. R.. Non-Invasive and Non-Destructive Characterization of Tissue Engineered Constructs Using Ultrasound Imaging Technologies: A Review. Ann. Biomed. Eng. 2016;44(3):621–635. doi: 10.1007/s10439-015-1495-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Mora-Navarro C., Garcia M. E., Sarker P., Ozpinar E. W., Enders J. R., Khan S., Branski R. C., Freytes D. O.. Monitoring Decellularization via Absorbance Spectroscopy during the Derivation of Extracellular Matrix Scaffolds. Biomed. Mater. 2022;17(1):015008. doi: 10.1088/1748-605X/ac361f. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Belashov A. V., Zhikhoreva A. A., Belyaeva T. N., Salova A. V., Kornilova E. S., Semenova I. V., Vasyutinskii O. S.. Machine Learning Assisted Classification of Cell Lines and Cell States on Quantitative Phase Images. Cells. 2021;10(10):2587. doi: 10.3390/cells10102587. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Barbulescu G. I., Buica T. P., Goje I. D., Bojin F. M., Ordodi V. L., Olteanu G. E., Heredea R. E., Paunescu V.. Optimization of Complete Rat Heart Decellularization Using Artificial Neural Networks. Micromachines. 2022;13(1):79. doi: 10.3390/mi13010079. [DOI] [PMC free article] [PubMed] [Google Scholar]

Articles from ACS Omega are provided here courtesy of American Chemical Society

RESOURCES