Artificial intelligence-assisted identification and quantification of osteoclasts

Impact of functional electrical stimulation on nerve-damaged muscles by quantifying fat infiltration using deep learning

Quantitative imaging in life sciences has evolved into a powerful approach combining advanced microscopy acquisition and automated analysis of image data. The focus of the present study is on the imaging-based evaluation of the posterior cricoarytenoid muscle (PCA) influenced by long-term functional electrical stimulation (FES), which may assist the inspiration of patients with bilateral vocal fold paresis. To this end, muscle cross-sections of the PCA of sheep were examined by quantitative image analysis. Previous investigations of the muscle fibers and the collagen amount have not revealed signs of atrophy and fibrosis due to FES by a laryngeal pacemaker. It was therefore hypothesized that regardless of the stimulation parameters the fat in the muscle cross-sections would not be significantly altered. We here extending our previous investigations using quantitative imaging of intramuscular fat in cross-sections. In order to perform this analysis both reliably and faster than a qualitative evaluation and time-consuming manual annotation, the selection of the automated method was of crucial importance. To this end, our recently established deep neural network IMFSegNet, which provides more accurate results compared to standard machine learning approaches, was applied to more than 300 H&E stained muscle cross-sections from 22 sheep. It was found that there were no significant differences in the amount of intramuscular fat between the PCA with and without long-term FES, nor were any significant differences found between the low and high duty cycle stimulated groups. This study on a human-like animal model not only confirms the hypothesis that FES with the selected parameters has no negative impact on the PCA, but also demonstrates that objective and automated deep learning-based quantitative imaging is a powerful tool for such a challenging analysis.

A Machine Learning-Based Image Segmentation Method to Quantify In Vitro Osteoclast Culture Endpoints

Quantification of in vitro osteoclast cultures (e.g. cell number) often relies on manual counting methods. These approaches are labour intensive, time consuming and result in substantial inter- and intra-user variability. This study aimed to develop and validate an automated workflow to robustly quantify in vitro osteoclast cultures. Using ilastik, a machine learning-based image analysis software, images of tartrate resistant acid phosphatase-stained mouse osteoclasts cultured on dentine discs were used to train the ilastik-based algorithm. Assessment of algorithm training showed that osteoclast numbers strongly correlated between manual- and automatically quantified values (r = 0.87). Osteoclasts were consistently faithfully segmented by the model when visually compared to the original reflective light images. The ability of this method to detect changes in osteoclast number in response to different treatments was validated using zoledronate, ticagrelor, and co-culture with MCF7 breast cancer cells. Manual and automated counting methods detected a 70% reduction (p < 0.05) in osteoclast number, when cultured with 10 nM zoledronate and a dose-dependent decrease with 1-10 μM ticagrelor (p < 0.05). Co-culture with MCF7 cells increased osteoclast number by ≥ 50% irrespective of quantification method. Overall, an automated image segmentation and analysis workflow, which consistently and sensitively identified in vitro osteoclasts, was developed. Advantages of this workflow are (1) significantly reduction in user variability of endpoint measurements (93%) and analysis time (80%); (2) detection of osteoclasts cultured on different substrates from different species; and (3) easy to use and freely available to use along with tutorial resources.

Contemporary Advances in Computer-Assisted Bone Histomorphometry and Identification of Bone Cells in Culture

Static and dynamic bone histomorphometry and identification of bone cells in culture are labor-intensive and highly repetitive tasks. Several computer-assisted methods have been proposed to ease these tasks and to take advantage of the increased computational power available today. The present review aimed to provide an overview of contemporary methods utilizing specialized computer software to perform bone histomorphometry or identification of bone cells in culture. In addition, a brief historical perspective on bone histomorphometry is included. We identified ten publications using five different computer-assisted approaches (1) ImageJ and BoneJ; (2) Histomorph: OsteoidHisto, CalceinHisto, and TrapHisto; (3) Fiji/ImageJ2 and Trainable Weka Segmentation (TWS); (4) Visiopharm and artificial intelligence (AI); and (5) Osteoclast identification using deep learning with Single Shot Detection (SSD) architecture, Darknet and You Only Look Once (YOLO), or watershed algorithm (OC_Finder). The review also highlighted a substantial need for more validation studies that evaluate the accuracy of the new computational methods to the manual and conventional analyses of histological bone specimens and cells in culture using microscopy. However, a substantial evolution has occurred during the last decade to identify and separate bone cells and structures of interest. Most early studies have used simple image segmentation to separate structures of interest, whereas the most recent studies have utilized AI and deep learning. AI has been proposed to substantially decrease the amount of time needed for analyses and enable unbiased assessments. Despite the clear advantages of highly sophisticated computational methods, the limited nature of existing validation studies, particularly those that assess the accuracy of the third-generation methods compared to the second-generation methods, appears to be an important reason that these techniques have failed to gain wide acceptance.

Automated Quantification of Human Osteoclasts Using Object Detection

A balanced skeletal remodeling process is paramount to staying healthy. The remodeling process can be studied by analyzing osteoclasts differentiated in vitro from mononuclear cells isolated from peripheral blood or from buffy coats. Osteoclasts are highly specialized, multinucleated cells that break down bone tissue. Identifying and correctly quantifying osteoclasts in culture are usually done by trained personnel using light microscopy, which is time-consuming and susceptible to operator biases. Using machine learning with 307 different well images from seven human PBMC donors containing a total of 94,974 marked osteoclasts, we present an efficient and reliable method to quantify human osteoclasts from microscopic images. An open-source, deep learning-based object detection framework called Darknet (YOLOv4) was used to train and test several models to analyze the applicability and generalizability of the proposed method. The trained model achieved a mean average precision of 85.26% with a correlation coefficient of 0.99 with human annotators on an independent test set and counted on average 2.1% more osteoclasts per culture than the humans. Additionally, the trained models agreed more than two independent human annotators, supporting a more reliable and less biased approach to quantifying osteoclasts while saving time and resources. We invite interested researchers to test their datasets on our models to further strengthen and validate the results.

OC_Finder: Osteoclast Segmentation, Counting, and Classification Using Watershed and Deep Learning

Osteoclasts are multinucleated cells that exclusively resorb bone matrix proteins and minerals on the bone surface. They differentiate from monocyte/macrophage lineage cells in the presence of osteoclastogenic cytokines such as the receptor activator of nuclear factor-κB ligand (RANKL) and are stained positive for tartrate-resistant acid phosphatase (TRAP). In vitro osteoclast formation assays are commonly used to assess the capacity of osteoclast precursor cells for differentiating into osteoclasts wherein the number of TRAP-positive multinucleated cells is counted as osteoclasts. Osteoclasts are manually identified on cell culture dishes by human eyes, which is a labor-intensive process. Moreover, the manual procedure is not objective and results in lack of reproducibility. To accelerate the process and reduce the workload for counting the number of osteoclasts, we developed OC_Finder, a fully automated system for identifying osteoclasts in microscopic images. OC_Finder consists of cell image segmentation with a watershed algorithm and cell classification using deep learning. OC_Finder detected osteoclasts differentiated from wild-type and Sh3bp2KI/+ precursor cells at a 99.4% accuracy for segmentation and at a 98.1% accuracy for classification. The number of osteoclasts classified by OC_Finder was at the same accuracy level with manual counting by a human expert. OC_Finder also showed consistent performance on additional datasets collected with different microscopes with different settings by different operators. Together, successful development of OC_Finder suggests that deep learning is a useful tool to perform prompt and accurate unbiased classification and detection of specific cell types in microscopic images.

Quantification of Immunohistochemically Stained Cells in Skin Biopsies

Immunohistochemical quantification of inflammatory cells in skin biopsies is a valuable tool for diagnosing skin diseases and assessing treatment response. The quantification of individual cells in biopsies is time-consuming, tedious, and difficult. In this study, we presented and compared two methods for the quantification of CD8⁺ T cells in skin biopsies from patients with psoriasis using both commercial software (Adobe Photoshop) and open-source software (Qupath). In addition, we provided a detailed, step-by-step description of both methods. The methods are scalable by replacing the CD8 antibody with other antibodies to target different cells. Moreover, we investigated the correlation between quantifying CD8⁺ cells normalized to area or epidermal length and cell classifications, compared cell classifications in QuPath with threshold classifications in Photoshop, and analyzed the impact of data normalization to epidermal length or area on inflammatory cell densities in skin biopsies from patients with psoriasis. We found a satisfactory correlation between normalizing data to epidermal length and area for psoriasis skin. However, when non-lesional and lesional skin samples were compared, a significant underestimation of inflammatory cell density was found when data were normalized to area instead of epidermal length. Finally, Bland–Altman plots comparing Qupath and Photoshop to quantify inflammatory cell density demonstrated a good agreement between the two methods.

Hypobaric hypoxia deteriorates bone mass and strength in mice

Mountaineers at high altitude are at increased risk of acute mountain sickness as well as high altitude pulmonary and cerebral edema. A densitometric study in mountaineers has suggested that expeditions at high altitude decrease bone mineral density. Surprisingly, the in vivo skeletal effects of hypobaric hypoxia are largely unknown, and have not been studied using advanced contemporary methods to assess bone microstructure. Eighty-four 22-week-old female mice were divided into seven groups with 12 mice in each group: 1. Baseline; 2. Normobaric, 4 weeks; 3. Hypobaric hypoxia, 4 weeks; 4. Normobaric, 8 weeks; 5. Hypobaric hypoxia, 8 weeks; 6. Normobaric, 12 weeks; and 7. Hypobaric hypoxia, 12 weeks. Hypobaric hypoxia mice were housed in hypobaric chambers at an ambient pressure of 500 mbar (5500 m altitude), while normobaric mice were housed at sea level atmospheric pressure for 4, 8, or 12 weeks, respectively. Hypobaric hypoxia had a profound impact on femoral cortical bone and L4 trabecular bone, while the effect on femoral trabecular bone was less pronounced. Hypobaric hypoxia reduced the bone strength of the femoral mid-diaphysis and L4 at all time-points. At femoral cortical bone, hypobaric hypoxia reduced bone formation through fewer mineralizing surfaces and lower bone formation rate after 2 weeks. In addition, bone strength decreased, and C-terminal telopeptide of type I collagen (CTX-I) increased independently of the duration of exposure to simulated high altitude. At L4, hypobaric hypoxia resulted in a substantial reduction in bone volume fraction, trabecular thickness, and trabecular number after 4 weeks of exposure. Hypobaric hypoxia reduced bone strength and femoral bone mass, while femoral trabecular bone was much less affected, indicating the skeletal response to hypobaric hypoxia differ between cortical and trabecular bone. These findings provide initial preclinical support for future clinical studies in mountaineers to assess bone status and bone strength after exposure to prolonged high altitude exposure.

Effect of Acetazolamide and Zoledronate on Simulated High Altitude-Induced Bone Loss

Exposure to hypobaric hypoxia at high altitude puts mountaineers at risk of acute mountain sickness. The carbonic anhydrase inhibitor acetazolamide is used to accelerate acclimatization, when it is not feasible to make a controlled and slow ascend. Studies in rodents have suggested that exposure to hypobaric hypoxia deteriorates bone integrity and reduces bone strength. The study investigated the effect of treatment with acetazolamide and the bisphosphonate, zoledronate, on the skeletal effects of exposure to hypobaric hypoxia. Eighty 16-week-old female RjOrl : SWISS mice were divided into five groups: 1. Baseline; 2. Normobaric; 3. Hypobaric hypoxia; 4. Hypobaric hypoxia + acetazolamide, and 5. Hypobaric hypoxia + zoledronate. Acetazolamide was administered in the drinking water (62 mg/kg/day) for four weeks, and zoledronate (100 μg/kg) was administered as a single subcutaneous injection at study start. Exposure to hypobaric hypoxia significantly increased lung wet weight and decreased femoral cortical thickness. Trabecular bone was spared from the detrimental effects of hypobaric hypoxia, although a trend towards reduced bone volume fraction was found at the L4 vertebral body. Treatment with acetazolamide did not have any negative skeletal effects, but could not mitigate the altitude-induced bone loss. Zoledronate was able to prevent the altitude-induced reduction in cortical thickness. In conclusion, simulated high altitude affected primarily cortical bone, whereas trabecular bone was spared. Only treatment with zoledronate prevented the altitude-induced cortical bone loss. The study provides preclinical support for future studies of zoledronate as a potential pharmacological countermeasure for altitude-related bone loss.