Ebook: Healthcare Transformation with Informatics and Artificial Intelligence

This volume contains the papers accepted for the ICIMTH (International Conference on Informatics, Management, and Technology in Healthcare) for the year 2023. The Scientific Programme Committee presents to the academic and professional community of biomedical and health informatics, the scientific outcomes of the ICIMTH 2023 Conference, held from 1–3 July 2023 in Athens, Greece.

The ICIMTH 2023 Conference is the 21st annual conference in this series of scientific events, which gathers scientists working in the field of biomedical and health informatics from all continents. As last year, because the situation as regards the COVID-19 pandemic has improved and restrictions were lifted, the conference was held as a live event. To encourage all presenters to attend the conference, virtual sessions by means of teleconferencing were not offered.

The field of biomedical and health informatics is studied at this conference from a very broad perspective, with the participants presenting research and application outcomes of informatics from cell to populations, including several technologies such as imaging, sensors, biomedical equipment and management and organisational aspects, including legal and social issues. Essentially, data science, informatics, and technology inspire health professionals and informaticians to improve healthcare for the benefit of patients. As was expected, a significant number of papers were still related to the COVID-19 pandemic, but this year we saw an increased influx of submissions in AI studies and applications in healthcare. We wanted this to be reflected in the theme of the conference and the title of the proceedings.

It should be noted that the proceedings are published as an open access eBook with e-access for ease of use and browsing without the loss of any of the advantages of indexing and citation in the biggest scientific literature databases, such as PubMed/Medline and Scopus, that the series of Studies in Health Technology and Informatics (HTI) of IOS Press provides.

At the end of the deadline we had more than 252 submissions, from which, after review, we accepted 149 as full papers, 13 as short communications, and 14 as poster papers.

The organisers would like to thank Dr. Spyros Zogas and Prof. John Mantas for the design of the cover illustration and of the entire conference series. This year the cover illustration signifies the reflection of AI applications in healthcare, as Hygeia faces her mirror automaton image, indicating the possible dangers to healthcare which exist during this extremely fast digital-transformation era of the healthcare system.

The Editors would like to thank the Members of the Scientific Programme Committee, the Organising Committee, and all Reviewers, who performed a very professional, thorough and objective refereeing of the scientific work to achieve a high-quality publishing achievement for a successful scientific event.

Athens, 26.05.2023

The Editors,

John Mantas, Parisis Gallos, Emmanouil Zoulias, Arie Hasman, Mowafa S. Househ, Martha Charalampidou and Andriana Magdalinou.

Automatic document classification is a common problem that has successfully been addressed with machine learning methods. However, these methods require extensive training data, which is not always readily available. Additionally, in privacy-sensitive settings, transfer and reuse of trained machine learning models is not an option because sensitive information could potentially be reconstructed from the model. Therefore, we propose a transfer learning method that uses ontologies to normalize the feature space of text classifiers to create a controlled vocabulary. This ensures that the trained models do not contain personal data, and can be widely reused without violating the GDPR. Furthermore, the ontologies can be enriched so that the classifiers can be transferred to contexts with different terminology without additional training. Applying classifiers trained on medical documents to medical texts written in colloquial language shows promising results and highlights the potential of the approach. The compliance with GDPR by design opens many further application domains for transfer learning based solutions.

Artificial intelligence (AI) tends to emerge as a relevant component of medical care, previously reserved for medical experts. A key factor for the utilization of AI is the user’s trust in the AI itself, respectively the AIt’s decision process, but AI-models are lacking information about this process, the so-called Black Box, potentially affecting usert’s trust in AI. This analysis’ objective is the description of trust-related research regarding AI-models and the relevance of trust in comparison to other AI-related research topics in healthcare. For this purpose, a bibliometric analysis relying on 12985 article abstracts was conducted to derive a co-occurrence network which can be used to show former and current scientific endeavors in the field of healthcare based AI research and to provide insight into underrepresented research fields. Our results indicate that perceptual factors such as “trust” are still underrepresented in the scientific literature compared to other research fields.

Acute kidney injury (AKI) is an abrupt decrease in kidney function widespread in intensive care. Many AKI prediction models have been proposed, but only few exploit clinical notes and medical terminologies. Previously, we developed and internally validated a model to predict AKI using clinical notes enriched with single-word concepts from medical knowledge graphs. However, an analysis of the impact of using multi-word concepts is lacking. In this study, we compare the use of only the clinical notes as input to prediction to the use of clinical notes retrofitted with both single-word and multi-word concepts. Our results show that 1) retrofitting single-word concepts improved word representations and improved the performance of the prediction model; 2) retrofitting multi-word concepts further improves both results, albeit slightly. Although the improvement with multi-word concepts was small, due to the small number of multi-word concepts that could be annotated, multi-word concepts have proven to be beneficial.

Digital Pathology is an area that could benefit a lot from the automatic classification of scanned microscopic slides. One of the main problems with this is that the experts need to understand and trust the decisions of the system. This paper is an overview of the current state of the art methods used in histopathological practice for explaining CNN classification useful for histopathological experts and ML engineers that work with histopathological images. This paper is an overview of the current state of the art methods used in the histopathological practice for explain. The search was performed using SCOPUS database and revealed that there are few applications of CNNs for digital pathology. The 4-term search yielded 99 results. This research sheds light on the main methods that can be used for histopathology classification and offers a good starting point for future works.

Health data democratization requires a transparent, protected, and interoperable data-sharing environment. We conducted a co-creation workshop with patients living with chronic diseases and relevant stakeholders to explore their opinion on health data democratization, ownership, and sharing in Austria. Participants showed their willingness to share their health data for clinical and research purposes; provided that appropriate transparency and data protection measures are provided.

Patient-Generated Health Data (PGHD), such as data provided by wearable devices, hold promise to improve health outcomes. However, to improve clinical decision-making, PGHD should be integrated or linked with Electronic Health Records (EHRs). Typically, PGHD data are collected and stored as Personal Health Records (PHRs), outside EHR systems. To address this challenge, we created a conceptual framework for PGHD/EHR interoperability through the Master Patient Index (MPI) and DH-Convener platform. Then, we identified the corresponding Minimum Clinical Data Set (MCDS) of PGHD to be exchanged with EHR. This generic approach can be used as a blueprint in different countries.

Although data quality is well defined, the relationship to data quantity remains unclear. Especially the big data approach promises advantages of volume in comparison with small samples in good quality. Aim of this study was to review this issue. Based on the experiences with six registries within a German funding initiative, the definition of data quality provided by the International Organization for Standardization (ISO) was confronted with several aspects of data quantity. The results of a literature search combining both concepts were considered additionally. Data quantity was identified as an umbrella of some inherent characteristics of data like case and data completeness. The same time, quantity could be regarded as a non inherent characteristic of data beyond the ISO standard focusing on the breadth and depth of metadata, i.e. data elements along with their value sets. The FAIR Guiding Principles take into account the latter solely. Surprisingly, the literature agreed in demanding an increase in data quality with volume, turning the big data approach inside out. A usage of data without context – as it could be the case in data mining or machine learning – is neither covered by the concept of data quality nor of data quantity.

The Leadership in Emergencies learning programme, launched in 2019, was designed to strengthen the competencies of World Health Organization (WHO) and Member State staff in teamwork, decision-making and communication, key skills required to lead effectively in emergencies. While the programme was initially used to train 43 staff in a workshop setting, the COVID-19 pandemic required a new remote approach. An online learning environment was developed using a variety of digital tools including WHO’s open learning platform, OpenWHO.org. The strategic use of these technologies enabled WHO to dramatically expand access to the programme for personnel responding to health emergencies in fragile contexts and increase participation among key groups that were previously underserved.

The YOLO series of object detection algorithms, including YOLOv4 and YOLOv5, have shown superior performance in various medical diagnostic tasks, surpassing human ability in some cases. However, their black-box nature has limited their adoption in medical applications that require trust and explainability of model decisions. To address this issue, visual explanations for AI models, known as visual XAI, have been proposed in the form of heatmaps that highlight regions in the input that contributed most to a particular decision. Gradient-based approaches, such as Grad-CAM [1], and non-gradient-based approaches, such as Eigen-CAM [2], are applicable to YOLO models and do not require new layer implementation. This paper evaluates the performance of Grad-CAM and Eigen-CAM on the VinDrCXR Chest X-ray Abnormalities Detection dataset [3] and discusses the limitations of these methods for explaining model decisions to data scientists.

Predicting waiting times in A&E is a critical tool for controlling the flow of patients in the department. The most used method (rolling average) does not account for the complex context of the A&E. Using retrospective data of patients visiting an A&E service from 2017 to 2019 (pre-pandemic). An AI-enabled method is used to predict waiting times in this study. A random forest and XGBoost regression methods were trained and tested to predict the time to discharge before the patient arrived at the hospital. When applying the final models to the 68,321 observations and using the complete set of features, the random forest algorithm’s performance measurements are RMSE=85.31 and MAE=66.71. The XGBoost model obtained a performance of RMSE=82.66 and MAE=64.31. The approach might be a more dynamic method to predict waiting times.

In this study, we attempted to classify categorical emotional states using Electrodermal Activity (EDA) signals and a configurable Convolutional Neural Network (cCNN). The EDA signals from the publicly available, Continuously Annotated Signals of Emotion dataset were down-sampled and decomposed into phasic components using the cvxEDA algorithm. The phasic component of EDA was subjected to Short-Time Fourier Transform-based time-frequency representation to obtain spectrograms. These spectrograms were input to the proposed cCNN to automatically learn the prominent features and discriminate varied emotions such as amusing, boring, relaxing, and scary. Nested k-Fold cross-validation was used to evaluate the robustness of the model. The results indicated that the proposed pipeline could discriminate the considered emotional states with a high average classification accuracy, recall, specificity, precision, and F-measure scores of 80.20%, 60.41%, 86.8%, 60.05%, and 58.61%, respectively. Thus, the proposed pipeline could be valuable in examining diverse emotional states in normal and clinical conditions.

During the COVID-19 pandemic the field of infodemic management has grown significantly. Social listening is the first step in managing the infodemic but little is known of the experience of public health professionals using social media analysis tools for health. Our survey sought the views of infodemic managers. Participants (n=417) had an average of 4.4 years’ experience in social media analysis for health. Results reveal gaps in technical capabilities of tools, data sources, and languages covered. For future planning for infodemic preparednessand preventi on it is vital to understand and deliver for analysis needs of those working in the field.

Trust in authorities is important during health emergencies, and there are many factors that influence this. The infodemic has resulted in overwhelming amounts of information being shared on digital media during the COVID-19 pandemic, and this research looked at trust-related narratives during a one-year period. We identified three key findings related to trust and distrust narratives, and a country-level comparison showed less mistrust narratives in a country with a higher level of trust in government. Trust is a complex construct and the findings of this study present results that warrant further exploration.

Gout is a systemic disease that is caused by the deposition of monosodium urate crystals in various tissues which leads to inflammation in them. This disease is often misdiagnosed. It leads to the lack of adequate medical care and development of serious complications, such as urate nephropathy and disability. The current situation can be improved by optimizing the medical care provided to patients, which requires searching for new strategies in terms of diagnosis. One of these strategies is the development of an expert system for providing information assistance to medical specialists which was a purpose of this study. The developed prototype expert system for gout diagnosis has knowledge base including 1144 medical concepts and 5 640 522 links, intelligent knowledge base editor and software which helps practitioner make the final decision. It has sensitivity of 91,3% [95% CI, 89,1%-93,1%], specificity of 85,4% [95% CI, 82,9%-87,6%] and AUROC 0,954 [95% CI, 0,944-0,963].

In this study, a new method for detecting emotions using Blood Volume Pulse (BVP) signals and machine learning was presented. The BVP of 30 subjects from the publicly available CASE dataset was pre-processed, and 39 features were extracted from various emotional states, such as amusing, boring, relaxing, and scary. The features were categorized into time, frequency, and time-frequency domains and used to build an emotion detection model with XGBoost. The model achieved the highest classification accuracy of 71.88% using the top 10 features. The most significant features of the model were computed from time (5 features), time-frequency (4 features), and frequency (1 feature) domains. The skewness calculated from the time-frequency representation of the BVP was ranked highest and played a crucial role in the classification. Our study suggests the potential of using BVP recorded from wearable devices to detect emotions in healthcare applications.

Medicines are important for well-being. Thus, medication errors can have severe consequences, even death. Transfers between professionals and levels of care are a challenge in terms of medicines management. Norwegian governmental strategies encourage communication and collaboration between levels of care, and several initiatives are invested in to improve digital medicine management. In the project Electronic Medicines Management (eMM), we established an arena for interprofessional discussions about medicines management. This paper provides an example of how the eMM arena contributed to knowledge sharing and development in current medicines management practices at a nursing home. Building on communities of practice as a method we carried out the first of several sessions, with nine interprofessional participants. The results illustrate how discussion and agreement were reached on a common practice across different levels of care, and how the knowledge required bringing this knowledge back to the local practices.

Our study used functional magnetic resonance imaging and fractal functional connectivity (FC) methods to analyze the brain networks of Autism Spectrum Disorder (ASD) and typically developing participants using data available on ABIDE databases. Blood-Oxygen-Level-Dependent time series were extracted from 236 regions of interest of cortical, subcortical, and cerebellar regions using Gordon’s, Harvard Oxford, and Diedrichsen atlases respectively. We computed the fractal FC matrices which resulted in 27,730 features, ranked using XGBoost feature ranking. Logistic regression classifiers were used to analyze the performance of the top 0.1%, 0.3%, 0.5%, 0.7%, 1%, 2%, and 3% of FC metrics. Results showed that 0.5% percentile features performed better, with average 5-fold accuracy of 94%. The study identified significant contributions from dorsal attention (14.75%), cingulo-opercular task control (14.39%), and visual networks (12.59%). This study could be used as an essential brain FC method to diagnose ASD.

In this study, we examined the structural connectivity (SC) of autism spectrum disorder (ASD) and typical development using the distance correlation and machine learning algorithm. We preprocessed diffusion tensor images using a standard pipeline and parcellated the brain into 48 regions using atlas. We derived diffusion measures in white matter tracts, such as fractional anisotropy, radial diffusivity, axial diffusivity, mean diffusivity, and mode of anisotropy. Additionally, SC is determined by the Euclidean distance between these features. The SC were ranked using XGBoost and significant features were fed as the input to the logistic regression classifier. We obtained an average 10-fold cross-validation classification accuracy of 81% for the top 20 features. The SC computed from the anterior limb of internal capsule L to superior corona radiata R regions significantly contributed to the classification models. Our study shows the potential utility of adopting SC changes as the biomarker for the diagnosis of ASD.

In this study, we classify the seizure types using feature extraction and machine learning algorithms. Initially, we pre-processed the electroencephalogram (EEG) of focal non-specific seizure (FNSZ), generalized seizure (GNSZ), tonic-clonic seizure (TCSZ), complex partial seizure (CPSZ) and absence seizure (ABSZ). Further, 21 features from time (9) and frequency (12) domain were computed from the EEG signals of different seizure types. XGBoost classifier model was built for individual domain features and combination of time and frequency features and validated the results using 10-fold cross-validation. Our results revealed that the classifier model with combination of time and frequency features performed well followed by the time and frequency domain features. We obtained a highest multi-class accuracy of 79.72% for the classification of five types of seizure while using all the 21 features. The band power between 11-13 Hz was found to be the top feature in our study. The proposed study can be used for the seizure type classification in clinical applications.

Domestic violence affects people of all socioeconomic backgrounds and education levels and can happen to anyone. It is a public health issue that needs to be addressed with health and social care professionals playing an essential role in prevention and early intervention. These professionals need to be prepared through proper education. A European funded project developed “DOMINO - Stop domestic violence” educational mobile application which was piloted among 99 social and/or health care students and professionals. Most of the participants (n= 59, 59.6%) indicated that the DOMINO mobile application was easy to install and over half of them (n=61, 61.6%) would recommend the app. They found it easy to use, and quick access to useful materials and tools. Participants found case studies and the checklist good and useful tools for them. The DOMINO educational mobile application is available open access, in English, Finnish, Greek, Latvian, Portuguese and Swedish, for any stakeholder worldwide who is interested to learn more about domestic violence prevention and intervention.

The aim of the paper is to conduct a formative evaluation and assess the implementation of a nursing app using the qualitative TPOM framework to outline how different socio-technical aspects of the process influence digital maturity. The research question is: what are the main socio-technical preconditions for improving digital maturity in a healthcare organization? We conducted 22 interviews and used the TPOM framework for analyzing the empirical data. Exploiting the potential of lightweight technology demands a mature healthcare organization motivated actors’ extensive collaboration, and good coordination of the complex ICT infrastructures. The TPOM categories are used to show the digital maturity of the nursing app implementation in relation to technology, human factors, organization, and the wider macro environment.

In this study, we analyzed the utility of electromyogram (EMG) signals recorded from the zygomaticus major (zEMG), the trapezius (tEMG), and the corrugator supercilii (cEMG) for emotion detection. We computed eleven-time domain features from the EMG signals to classify the emotions such as amusing, boring, relaxing, and scary. The features were fed to the logistic regression, support vector machine, and multilayer perceptron classifiers, and model performance was evaluated. We achieved an average 10-fold cross-validation classification accuracy of 67.29%. 67.92% and 64.58% by LR using the features extracted from the EMG signals recorded from the zEMG, tEMG, and cEMG, respectively. The classification accuracy improved to 70.6% while combining features from the zEMG and cEMG for the LR model. However, the performance dropped while including the features of EMG from all three locations. Our study shows the importance of utilizing the zEMG and cEMG combination for emotion recognition.

University students are experiencing a mental health crisis across the world. COVID-19 has exacerbated this situation. We have conducted a survey among university students in two universities in Lebanon to gauge mental health challenges experienced by students. We constructed a machine learning approach to predict anxiety symptoms among the sample of 329 respondents based on student survey items including demographics and self-rated health. Five algorithms including logistic regression, multi-layer perceptron (MLP) neural network, support vector machine (SVM), random forest (RF) and XGBoost were used to predict anxiety. Multi-Layer Perceptron (MLP) provided the highest performing model AUC score (AUC=80.70%) and self-rated health was found to be the top ranked feature to predict anxiety. Future work will focus on using data augmentation approaches and extending to multi-class anxiety predictions. Multidisciplinary research is crucial in this emerging field.

Physician shortage is a major concern in many health care systems globally, while healthcare leadership constitutes one of the most vital factors within human resource management. Our study examined the relationship between managers’ leadership styles and physicians’ intent to leave their current position. In this cross-sectional national survey, questionnaires were distributed to all physicians working in the public health sector of Cyprus. Most demographic characteristics evaluated by chi-square or Mann-Whitney test, were statistically significantly different between those who intended to leave their job and those who did not. The results of our study demonstrated that transformational leadership has a positive influence on retention of physicians in public hospitals, while non leadership infers a negative influence. Developing leadership skills in physician supervisors is of a great importance for organizations to make a large impact on health professionals’ retention and overall performance.