Recently, personal healthcare has become an important issue. Artificial intelligence (AI) technology can predict abnormal symptoms and body conditions. The Internet of Things (IoT) can collect human body datasets and check the condition of a person’s body through specific devices. Furthermore, data analysis techniques can help find patterns in healthcare data and present graphs of the results that users can easily understand.
However, healthcare based on such technologies would be difficult to realise. Data features are huge and complicated; thus, service models generally exhibit poor performance. To increase the accuracy of healthcare services, new AI, IoT, and data analysis technologies are necessary to consider different aspects and a variety of theoretical and experimental contributions, including new services, architecture, systems, algorithms, and approaches.
We edited a special issue of Connection Science on this topic. In the special issue, several leading researchers presented their ideas and argued their relevance and significance.
“Detecting susceptible communities and individuals in hospital contact networks: a model based on social network analysis” focuses on identifying vulnerable communities in hospital contact networks, including hospital stakeholders, patients, and healthcare workers, to determine highly influential nodes within the network by analyzing independent contact networks. To identify the communities, the authors use the cliques percolation method (CPM), an overlapping community detection method, and the PageRank algorithm to rank the node influence. The experimental results obtained from an actual hospital contact network over four days of experimentation demonstrated the model's effectiveness.
“Network security situation assessment based on dual attention mechanism and HHO-ResNeXt” focuses on combining ResNeXt with efficient channel attention and a contextual transformer to construct a network state model. Existing convolutional neural networks (CNNs) cannot solve complex network security problems, because determining the importance of a channel is difficult owing to the limited reception area. Accordingly, the Harris Hawks optimisation algorithm is selected to optimise the proposed model. Specifically, the authors calculate the network security situation value of the adopted dataset and demonstrate the model’s superiority in three indicators, namely, accuracy, precision, and the F-score, in the network security situation evaluation.
“Early Prediction of Ransomware API Calls Behaviour based on GRU-TCN in Healthcare IoT” focuses on protecting the medical IoT environment against increasing ransomware attacks by predicting ransomware behaviour to encrypt medical systems or leaked data. Although many methods have been analyzed, they may become infected or encrypted. The study proposes an early prediction technique for ransomware behaviour (EPS-Ran) to reduce the possibility of system infection when analyzing such behaviour. The EPS-Ran analyzes the opcode and API call sequences for over 30 seconds of operation; thus, it can predict ransomware behaviour and has a lower error rate than existing methods, even though the operation was shortened from 120 seconds to 30 seconds.
“Particle-based simulation technique for medical applications” uses existing particle-based techniques to simulate nonlinear elastic objects. However, only tissue accurately cut during incision can be simulated. Therefore, the study proposes a particle-based elastic object simulation technique to model debris. The study also verifies the superiority of the proposed method compared with traditional particle-based methods by accurately representing complex debris and comparing incisions in the same body area.
“Predicting Adolescent Violence in WZT Drawing Images Based on Deep Learning” focuses on the problem of negative behaviour in adolescents owing to changes in mental and physical stress. Among different problematic behaviours, students exposed to violence exhibit various healthcare-related problems. A projective test using pictures is performed on the students to obtain information. The pictures represent direct experiences that may elicit subconscious reactions from the subjects. Few methods can be used to analyze images drawn by adolescents as image data. The study employs image data created by adolescents through the Wartegg–ZeichenTest method to predict adolescent violence. The subjects include students from special schools who received punishment for violence. The study uses a CNN, SoftMax, support vector machine with a style transfer generative adversarial network, and ensemble techniques to analyze drawn images and predict violence through deep learning.