A Hybrid AI-Based Method for ICD Classification of Medical Documents.

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.

An empirical evaluation of a novel domain-specific language - modelling vehicle routing problems with Athos.

Domain-specific languages (DSLs) are a popular approach among software engineers who demand for a tailored development interface. A DSL-based approach allows to encapsulate the intricacies of the target platform in transformations that turn DSL models into executable software code. Often, DSLs are even claimed to reduce development complexity to a level that allows them to be successfully applied by domain-experts with limited programming knowledge. Recent research has produced some scientifically backed insights on the benefits and limitations of DSLs. Further empirical studies are required to build a sufficient body of knowledge from which support for different claims related to DSLs can be derived. In this research study, we adopt current DSL evaluation approaches to investigate potential gains in terms of effectiveness and efficiency, through the application of our DSL Athos, a language developed for the domain of traffic and transportation simulation and optimisation. We compare Athos to the alternative of using an application library defined within a general-purpose language (GPL). We specified two sets of structurally identical tasks from the domain of vehicle routing problems and asked study groups with differing levels of programming knowledge to solve the tasks with the two approaches. The results show that inexperienced participants achieved considerable gains in effectiveness and efficiency with the usage of Athos DSL. Though hinting at Athos being the more efficient approach, the results were less distinct for more experienced programmers. The vast majority of participants stated to prefer working with Athos over the usage of the presented GPL's API. Supplementary Information: The online version contains supplementary material available at 10.1007/s10664-022-10210-whttps://doi.org/10.1007/s10664-022-10210-w.

The Disruption of Trust in the Digital Transformation Leading to Health 4.0.

The specification and application of policies and guidelines for public health, medical education and training, and screening programmes for preventative medicine are all predicated on trust relationships between medical authorities, health practitioners and patients. These relationships are in turn predicated on a verbal contract that is over two thousand years old. The impact of information and communication technology (ICT), underpinning Health 4.0, has the potential to disrupt this analog relationship in several dimensions; but it also presents an opportunity to strengthen it, and so to increase the take-up and effectiveness of new policies. This paper develops an analytic framework for the trust relationships in Health 4.0, and through three use cases, assesses a medical policy, the introduction of a new technology, and the implications of that technology for the trust relationships. We integrate this assessment in a set of actionable recommendations, in particular that the trust framework should be part of the design methodology for developing and deploying medical applications. In a concluding discussion, we advocate that, in a post-pandemic world, IT to support policies and programmes to address widespread socio-medical problems with mental health, long Covid, physical inactivity and vaccine misinformation will be essential, and for that, strong trust relationships between all the stakeholders are absolutely critical.

Detecting myocardial scar using electrocardiogram data and deep neural networks.

Ischaemic heart disease is among the most frequent causes of death. Early detection of myocardial pathologies can increase the benefit of therapy and reduce the number of lethal cases. Presence of myocardial scar is an indicator for developing ischaemic heart disease and can be detected with high diagnostic precision by magnetic resonance imaging. However, magnetic resonance imaging scanners are expensive and of limited availability. It is known that presence of myocardial scar has an impact on the well-established, reasonably low cost, and almost ubiquitously available electrocardiogram. However, this impact is non-specific and often hard to detect by a physician. We present an artificial intelligence based approach - namely a deep learning model - for the prediction of myocardial scar based on an electrocardiogram and additional clinical parameters. The model was trained and evaluated by applying 6-fold cross-validation to a dataset of 12-lead electrocardiogram time series together with clinical parameters. The proposed model for predicting the presence of scar tissue achieved an area under the curve score, sensitivity, specificity, and accuracy of 0.89, 70.0, 84.3, and 78.0%, respectively. This promisingly high diagnostic precision of our electrocardiogram-based deep learning models for myocardial scar detection may support a novel, comprehensible screening method.

Identifying Heart Failure in ECG Data With Artificial Intelligence-A Meta-Analysis.

Introduction: Electrocardiography (ECG) is a quick and easily accessible method for diagnosis and screening of cardiovascular diseases including heart failure (HF). Artificial intelligence (AI) can be used for semi-automated ECG analysis. The aim of this evaluation was to provide an overview of AI use in HF detection from ECG signals and to perform a meta-analysis of available studies. Methods and Results: An independent comprehensive search of the PubMed and Google Scholar database was conducted for articles dealing with the ability of AI to predict HF based on ECG signals. Only original articles published in peer-reviewed journals were considered. A total of five reports including 57,027 patients and 579,134 ECG datasets were identified including two sets of patient-level data and three with ECG-based datasets. The AI-processed ECG data yielded areas under the receiver operator characteristics curves between 0.92 and 0.99 to identify HF with higher values in ECG-based datasets. Applying a random-effects model, an sROC of 0.987 was calculated. Using the contingency tables led to diagnostic odds ratios ranging from 3.44 [95% confidence interval (CI) = 3.12-3.76] to 13.61 (95% CI = 13.14-14.08) also with lower values in patient-level datasets. The meta-analysis diagnostic odds ratio was 7.59 (95% CI = 5.85-9.34). Conclusions: The present meta-analysis confirms the ability of AI to predict HF from standard 12-lead ECG signals underlining the potential of such an approach. The observed overestimation of the diagnostic ability in artificial ECG databases compared to patient-level data stipulate the need for robust prospective studies.