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.

Facing the FACS-Using AI to Evaluate and Control Facial Action Units in Humanoid Robot Face Development.

This paper presents a new approach for evaluating and controlling expressive humanoid robotic faces using open-source computer vision and machine learning methods. Existing research in Human-Robot Interaction lacks flexible and simple tools that are scalable for evaluating and controlling various robotic faces; thus, our goal is to demonstrate the use of readily available AI-based solutions to support the process. We use a newly developed humanoid robot prototype intended for medical training applications as a case example. The approach automatically captures the robot's facial action units through a webcam during random motion, which are components traditionally used to describe facial muscle movements in humans. Instead of manipulating the actuators individually or training the robot to express specific emotions, we propose using action units as a means for controlling the robotic face, which enables a multitude of ways to generate dynamic motion, expressions, and behavior. The range of action units achieved by the robot is thus analyzed to discover its expressive capabilities and limitations and to develop a control model by correlating action units to actuation parameters. Because the approach is not dependent on specific facial attributes or actuation capabilities, it can be used for different designs and continuously inform the development process. In healthcare training applications, our goal is to establish a prerequisite of expressive capabilities of humanoid robots bounded by industrial and medical design constraints. Furthermore, to mediate human interpretation and thus enable decision-making based on observed cognitive, emotional, and expressive cues, our approach aims to find the minimum viable expressive capabilities of the robot without having to optimize for realism. The results from our case example demonstrate the flexibility and efficiency of the presented AI-based solutions to support the development of humanoid facial robots.

Detecting glass and metal in consumer trash bags during waste collection using convolutional neural networks.

We present a proof-of-concept method to classify the presence of glass and metal in consumer trash bags. With the prevalent utilization of waste collection trucks in municipal solid waste management, the aim of this method is to help pinpoint the locations where waste sorting quality is below accepted standards, making it possible and more efficient to develop tailored procedures that can improve the waste sorting quality in areas with the most urgent needs. Using trash bags containing various amounts of glass and metal, in addition to common waste found in households, we use a combination of sound recording and a beat-frequency oscillation metal detector as inputs to a machine learning algorithm to identify the occurrence of glass and metal in trash bags. A custom-built test rig was developed to mimic a real waste collection truck, which was used to test different sensors and build the datasets. Convolutional neural networks were trained for the classification task, achieving accuracies of up to 98%. These promising results support this method's potential implementation in real waste collection trucks, enabling location-specific and long-term monitoring of consumer waste sorting quality, which can provide decision support for waste management systems, and research on consumer behavior.