A Semiautomatic Cell Counting Tool for Quantitative Imaging of Tissue Engineering Scaffolds

Deep Learning-Based TEM Image Analysis for Fully Automated Detection of Gold Nanoparticles Internalized Within Tumor Cell

Transmission electron microscopy (TEM) imaging can be used for detection/localization of gold nanoparticles (GNPs) within tumor cells. However, quantitative analysis of GNP-containing cellular TEM images typically relies on conventional/thresholding-based methods, which are manual, time-consuming, and prone to human errors. In this study, therefore, deep learning (DL)-based methods were developed for fully automated detection of GNPs from cellular TEM images. Several models of "you only look once (YOLO)" v5 were implemented, with a few adjustments to enhance the model's performance by applying the transfer learning approach, adjusting the size of the input image, and choosing the best optimization algorithm. Seventy-eight original (12,040 augmented) TEM images of GNP-laden tumor cells were used for model implementation and validation. A maximum F1 score (harmonic mean of the precision and recall) of 0.982 was achieved by the best-trained models, while mean average precision was 0.989 and 0.843 at 0.50 and 0.50-0.95 intersection over union threshold, respectively. These results suggested the developed DL-based approach was capable of precisely estimating the number/position of internalized GNPs from cellular TEM images. A novel DL-based TEM image analysis tool from this study will benefit research/development efforts on GNP-based cancer therapeutics, for example, by enabling the modeling of GNP-laden tumor cells using nanometer-resolution TEM images.

Object detection for automatic cancer cell counting in zebrafish xenografts

Cell counting is a frequent task in medical research studies. However, it is often performed manually; thus, it is time-consuming and prone to human error. Even so, cell counting automation can be challenging to achieve, especially when dealing with crowded scenes and overlapping cells, assuming different shapes and sizes. In this paper, we introduce a deep learning-based cell detection and quantification methodology to automate the cell counting process in the zebrafish xenograft cancer model, an innovative technique for studying tumor biology and for personalizing medicine. First, we implemented a fine-tuned architecture based on the Faster R-CNN using the Inception ResNet V2 feature extractor. Second, we performed several adjustments to optimize the process, paying attention to constraints such as the presence of overlapped cells, the high number of objects to detect, the heterogeneity of the cells' size and shape, and the small size of the data set. This method resulted in a median error of approximately 1% of the total number of cell units. These results demonstrate the potential of our novel approach for quantifying cells in poorly labeled images. Compared to traditional Faster R-CNN, our method improved the average precision from 71% to 85% on the studied data set.

An Image-Guided Intrascaffold Cell Assembly Technique for Accurate Printing of Heterogeneous Tissue Constructs

For tissue engineering and regenerative medicine, creating thick and heterogeneous scaffold-based tissue constructs requires deep and precise multicellular deposition. Traditional cell seeding strategies lack the ability to create multicellular tissue constructs with high cell penetration and distribution, while emerging strategies aim to simultaneously combine cell-laden tissue segments with scaffold fabrication. Here we describe a technique that allows for three-dimensional (3D) intrascaffold cell assembly in which scaffolds are prefabricated and pretreated, followed by accurate cell distribution within the scaffold using an image-guided technique. This two-step process yields less limitation in scaffold material choice as well as additional treatments, provides accurate cell distribution, and has less potential to harm cells. The image processing technique captures a 2D geometric image of the scaffold, followed by a series of processes, mainly including grayscale transformation, threshold segmentation, and boundary extraction, to ultimately locate scaffold macropore centroids. Coupled with camera calibration data, accurate 3D cell assembly pathway plans can be made. Intrascaffold assembly parameter optimization and complex intrascaffold gradient, multidirectional, and vascular structure assembly were studied. Demonstration was also made with path planning and cell assembly experiments using NIH3T3-cell-laden hydrogels and collagen-coated poly(lactic-co-glycolic acid) (PLGA) scaffolds. Experiments with CellTracker fluorescent monitoring, live/dead staining, and phalloidin-F-actin/DAPI immunostaining and comparison with two control groups (bioink manual injection and cell suspension static surface pipetting) showed accurate cell distribution and positioning and high cell viability (>93%). The PrestoBlue assay showed obvious cell proliferation over seven culture days in vitro. This technique provides an accurate method to aid simple and complex cell colonization with variant depth within 3D-scaffold-based constructs using multiple cells. The modular method can be used with any existing printing platform and shows potential in facilitating direct spatial organization and hierarchal 3D assembly of multiple cells and/or drugs within scaffolds for further tissue engineering studies and clinical applications.

Improving reliability of live/dead cell counting through automated image mosaicing

Cell counting is one of the basic needs of most biological experiments. Numerous methods and systems have been studied to improve the reliability of counting. However, at present, manual cell counting performed with a hemocytometer still represents the gold standard, despite several problems limiting reproducibility and repeatability of the counts and, at the end, jeopardizing their reliability in general. We present our own approach based on image processing techniques to improve counting reliability. It works in two stages: first building a high-resolution image of the hemocytometer's grid, then counting the live and dead cells by tagging the image with flags of different colours. In particular, we introduce GridMos (http://sourceforge.net/p/gridmos), a fully-automated mosaicing method to obtain a mosaic representing the whole hemocytometer's grid. In addition to offering more significant statistics, the mosaic "freezes" the culture status, thus permitting analysis by more than one operator. Finally, the mosaic achieved can thus be tagged by using an image editor, thus markedly improving counting reliability. The experiments performed confirm the improvements brought about by the proposed counting approach in terms of both reproducibility and repeatability, also suggesting the use of a mosaic of an entire hemocytometer's grid, then labelled trough an image editor, as the best likely candidate for the new gold standard method in cell counting.

Semi-quantitative monitoring of confluence of adherent mesenchymal stromal cells on calcium-phosphate granules by using widefield microscopy images

The analysis of cell confluence and proliferation is essential to design biomaterials and scaffolds to use as bone substitutes in clinical applications. Accordingly, several approaches have been proposed in the literature to estimate the area of the scaffold covered by cells. Nevertheless, most of the approaches rely on sophisticated equipment not employed for routine analyses, while the rest of them usually do not provide significant statistics about the cell distribution. This research aims at studying confluence and proliferation of mesenchymal stromal cells (MSC) adherent on OSPROLIFE(®), a commercial biomaterial in the form of granules. In particular, we propose a Computer Vision approach that can routinely be employed to monitor the surface of the single granules covered by cells because only a standard widefield fluorescent microscope is required. In order to acquire significant statistics data, we analyse wide-area images built by using MicroMos v2.0, an updated version of a previously published software specific for stitching brightfield and phase-contrast images manually acquired via a widefield microscope. In particular, MicroMos v2.0 permits to build accurate "mosaics" of fluorescent images, after correcting vignetting and photo-bleaching effects, providing a consistent representation of a sample region containing numerous granules. Then, our method allows to make automatically a statistically significant estimate of the percentage of the area of the single granules covered by cells. Finally, by analysing hundreds of granules at different time intervals we also obtained reliable data regarding cell proliferation, confirming that not only MSC adhere onto the OSPROLIFE(®) granules, but even proliferate over time.