Automated Quantification of Human Osteoclasts Using Object Detection

Abstract

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.

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

“…Similar reductions in user variability upon automation of histomorphometric analyses have been reported [ 10 , 39 – 41 ]. In contrast, the recent AI-based models quantifying in vitro osteoclasts on plastic did not measure improvements in operator variability from manual counting methods [ 22 – 25 ]. The ilastik model presented in this study requires limited operator input of defined parameters (as defined in Supp.…”

“…It is, therefore, likely that more complex models, such as deep learning (DL), will be required to fully automate the simultaneous quantification of both osteoclast number and resorptive activity. DL has already successfully quantified osteoclast and nuclei numbers [ 22 , 24 , 25 ], but not resorption events. Due to greater processing layers, DL could discover complicated feature patterns in large datasets that better delimit the resorption pit-dentine disc boundary for osteoclast activity analysis.…”

“…Historically, automatically quantifying osteoclasts in vitro has proven challenging due to the non-uniform cell shape, size, and considerable spacing between nuclei and the cytoplasm of single osteoclasts [ 8 , 21 ]. Four recent reports have built complex AI-based models to quantify TRAP+ or fluorescently labelled osteoclasts cultured on plastic but not bone or dentine [ 22 – 25 ]. Resorption parameters were not quantified in any of these models [ 22 – 25 ].…”

“…Four recent reports have built complex AI-based models to quantify TRAP+ or fluorescently labelled osteoclasts cultured on plastic but not bone or dentine [ 22 – 25 ]. Resorption parameters were not quantified in any of these models [ 22 – 25 ]. To date, ML, specifically ilastik, has not been applied to simultaneously measure osteoclast culture endpoints such as osteoclast number and resorption area for cells grown on physiologically relevant substrates.…”

Sensor extended imaging workflow for creating fit for purpose models in basic and applied cell biology

“…By these means, imaging became standard in screening assays as well as in the observation of rare events 27 . By the broad implementation of image analysis using deep learning algorithms the value of images as a read-out in cell biology has increased tremendously 28 . One recent example of deep learning in combination with imaging is sensor-extended imaging flow cytometry, which enables high-throughput single-cell analysis and compensates for the technical loss of resolution with a virtual high-resolution image generator 29 .…”