Objectives: Smart hospitals involve the application of recent information and communications technology (ICT) innovations to medical services; however, the concept of a smart hospital has not been rigorously defined. In this study, we aimed to derive the definition and service types of smart hospitals and investigate cases of each type. Methods: A literature review was conducted regarding the background and technical characteristics of smart hospitals. On this basis, we conducted a focus group interview with experts in hospital information systems, and ultimately derived eight smart hospital service types.Results: Smart hospital services can be classified into the following types: services based on location recognition and tracking technology that measures and monitors the location information of an object based on short-range communication technology; high-speed communication network-based services based on new wireless communication technology; Internet of Things-based services that connect objects embedded with sensors and communication functions to the internet; mobile health services such as mobile phones, tablets, and wearables; artificial intelligence-based services for the diagnosis and prediction of diseases; robot services provided on behalf of humans in various medical fields; extended reality services that apply hyper-realistic immersive technology to medical practice; and telehealth using ICT. Conclusions: Smart hospitals can influence health and medical policies and create new medical value by defining and quantitatively measuring detailed indicators based on data collected from existing hospitals. Simultaneously, appropriate government incentives, consolidated interdisciplinary research, and active participation by industry are required to foster and facilitate smart hospitals.

The white blood cell differential count of the bone marrow provides information concerning the distribution of immature and mature cells within maturation stages. The results of such examinations are important for the diagnosis of various diseases and for follow-up care after chemotherapy. However, manual, labor-intensive methods to determine the differential count lead to inter- and intra-variations among the results obtained by hematologists. Therefore, an automated system to conduct the white blood cell differential count is highly desirable, but several difficulties hinder progress. There are variations in the white blood cells of each maturation stage, small inter-class differences within each stage, and variations in images because of the different acquisition and staining processes. Moreover, a large number of classes need to be classified for bone marrow smear analysis, and the high density of touching cells in bone marrow smears renders difficult the segmentation of single cells, which is crucial to traditional image processing and machine learning. Few studies have attempted to discriminate bone marrow cells, and even these have either discriminated only a few classes or yielded insufficient performance. In this study, we propose an automated white blood cell differential counting system from bone marrow smear images using a dual-stage convolutional neural network (CNN). A total of 2,174 patch images were collected for training and testing. The dual-stage CNN classified images into 10 classes of the myeloid and erythroid maturation series, and achieved an accuracy of 97.06%, a precision of 97.13%, a recall of 97.06%, and an F-1 score of 97.1%. The proposed method not only showed high classification performance, but also successfully classified raw images without single cell segmentation and manual feature extraction by implementing CNN. Moreover, it demonstrated rotation and location invariance. These results highlight the promise of the proposed method as an automated white blood cell differential count system.

As the number of robotic surgery procedures has increased, so has the importance of evaluating surgical skills in these techniques. It is difficult, however, to automatically and quantitatively evaluate surgical skills during robotic surgery, as these skills are primarily associated with the movement of surgical instruments. This study proposes a deep learning-based surgical instrument tracking algorithm to evaluate surgeons’ skills in performing procedures by robotic surgery. This method overcame two main drawbacks: occlusion and maintenance of the identity of the surgical instruments. In addition, surgical skill prediction models were developed using motion metrics calculated from the motion of the instruments. The tracking method was applied to 54 video segments and evaluated by root mean squared error (RMSE), area under the curve (AUC), and Pearson correlation analysis. The RMSE was 3.52 mm, the AUC of 1 mm, 2 mm, and 5 mm were 0.7, 0.78, and 0.86, respectively, and Pearson’s correlation coefficients were 0.9 on the x-axis and 0.87 on the y-axis. The surgical skill prediction models showed an accuracy of 83% with Objective Structured Assessment of Technical Skill (OSATS) and Global Evaluative Assessment of Robotic Surgery (GEARS). The proposed method was able to track instruments during robotic surgery, suggesting that the current method of surgical skill assessment by surgeons can be replaced by the proposed automatic and quantitative evaluation method.

Background: Federated learning is a decentralized approach to machine learning; it is a training strategy that overcomes medical data privacy regulations and generalizes deep learning algorithms. Federated learning mitigates many systemic privacy risks by sharing only the model and parameters for training, without the need to export existing medical data sets. In this study, we performed ultrasound image analysis using federated learning to predict whether thyroid nodules were benign or malignant. Objective: The goal of this study was to evaluate whether the performance of federated learning was comparable with that of conventional deep learning. Methods: A total of 8457 (5375 malignant, 3082 benign) ultrasound images were collected from 6 institutions and used for federated learning and conventional deep learning. Five deep learning networks (VGG19, ResNet50, ResNext50, SE-ResNet50, and SE-ResNext50) were used. Using stratified random sampling, we selected 20% (1075 malignant, 616 benign) of the total images for internal validation. For external validation, we used 100 ultrasound images (50 malignant, 50 benign) from another institution. Results: For internal validation, the area under the receiver operating characteristic (AUROC) curve for federated learning was between 78.88% and 87.56%, and the AUROC for conventional deep learning was between 82.61% and 91.57%. For external validation, the AUROC for federated learning was between 75.20% and 86.72%, and the AUROC curve for conventional deep learning was between 73.04% and 91.04%. Conclusions: We demonstrated that the performance of federated learning using decentralized data was comparable to that of conventional deep learning using pooled data. Federated learning might be potentially useful for analyzing medical images while protecting patients’ personal information.