Abstract This study reviews the role of artificial intelligence (AI) in enhancing healthcare through an analysis of physician-to-physician communication. It seeks to identify the best practices for extracting value from professional medical chats (PMCs) and assess the impact of AI on patient outcomes and healthcare systems, emphasizing the integration of ethical and responsible AI practices. We conducted an extensive systematic literature review using the Web of Science Core Collection. Searches encompassed English-language articles published between January 2019 and July 2023 using keywords related to AI, machine learning, natural language processing, and physician communication. Of the 247 articles screened, 13 met the inclusion criteria given their in-depth analysis of AI in healthcare communication, methodological soundness, and relevance to clinical outcomes. The review provides insights into interprofessional communication dynamics, the advancement of NLP and deep learning in medical dialogues, and strategies for effective human-machine collaboration. Ethical considerations and the need for transparency in AI applications are key to these central findings. This study highlights the untapped potential of physician-generated real-world data in creating value for healthcare ecosystems. It advocates for a multidisciplinary strategy encompassing communication, education, and collaboration to advance AI in healthcare responsibly. Moreover, it suggests that by combining existing techniques in the AI discipline, including neural networks, generative AI, and genetic algorithms, as well as keeping a “physician in the loop” when building AI systems, we can have a significant impact on healthcare delivery and medical research.

Integrating advanced data processing technologies into healthcare has shifted the medical studies paradigm. These evolve from data collection into management and analysis of Electronic Health Records (EHR) data. This change improved patient care and expanded the scope of clinical research through the secondary usage of existing data. Even though this problem was already solved in other initiatives, it raised new challenges, namely regarding cohort definition, data discovery, and evaluating the study feasibility. There are database catalogues to help in those tasks, but these fail in some cases due to insufficient information. Therefore, in this paper, we address this challenge by proposing a baseline method for information retrieval, including a synthetic dataset for further research. The information present in the dataset was generated from metadata extracted from real-world databases, which represents real problems that do not yet have a solution. The source code of this work is available at http://github.com/bioinformatics-ua/HealthDBFinder.

Over the last two years, the COVID-19 pandemic has affected hundreds of millions of people around the world. As in many crises, people turn to social media platforms, like Twitter, to communicate and share information. Twitter datasets have been used over the years in many research studies to extract valuable information. Therefore, several large COVID-19 Twitter datasets have been released over the last two years. However, none of these datasets contains only Portuguese Tweets, despite the Portuguese Language being reported as one of the top five languages used on Twitter. In this paper, we present the first large-scale Portuguese COVID-19 Twitter dataset. The dataset contains over 19 million Tweets spanning 2020 and 2021, allowing the entire pandemic to be analyzed. We also conducted a sentiment analysis on the dataset and correlated the various spikes in Tweet count and sentiment scores to various news articles and government announcements in Portugal and Brazil. The dataset is available at: https://github.com/bioinformatics-ua/Portuguese-Covid19-Dataset

BACKGROUND: Electronic Health Records (EHRs) aggregate diverse information at the patient level, holding a trajectory representative of the evolution of the patient health status throughout time. Although this information provides context and can be leveraged by physicians to monitor patient health and make more accurate prognoses/diagnoses, patient records can contain information from very long time spans, which combined with the rapid generation rate of medical data makes clinical decision making more complex. Patient trajectory modelling can assist by exploring existing information in a scalable manner, and can contribute in augmenting health care quality by fostering preventive medicine practices (e.g. earlier disease diagnosis). METHODS: We propose a solution to model patient trajectories that combines different types of information (e.g. clinical text, standard codes) and considers the temporal aspect of clinical data. This solution leverages two different architectures: one supporting flexible sets of input features, to convert patient admissions into dense representations; and a second exploring extracted admission representations in a recurrent-based architecture, where patient trajectories are processed in sub-sequences using a sliding window mechanism. RESULTS: The developed solution was evaluated on two different clinical outcomes, unexpected patient readmission and disease progression, using the publicly available Medical Information Mart for Intensive Care (MIMIC)-III clinical database. The results obtained demonstrate the potential of the first architecture to model readmission and diagnoses prediction using single patient admissions. While information from clinical text did not show the discriminative power observed in other existing works, this may be explained by the need to fine-tune the clinicalBERT model. Finally, we demonstrate the potential of the sequence-based architecture using a sliding window mechanism to represent the input data, attaining comparable performances to other existing solutions. CONCLUSION: Herein, we explored DL-based techniques to model patient trajectories and propose two flexible architectures that explore patient admissions on an individual and sequence basis. The combination of clinical text with other types of information led to positive results, which can be further improved by including a fine-tuned version of clinicalBERT in the architectures. The proposed solution can be publicly accessed at https://github.com/bioinformatics-ua/PatientTM.

Physicians use electronic health records to monitor patient health and make more accurate prognoses, diagnoses and clinical decisions. However, with the ever increasing amounts of information stored for each patient, manually processing and digesting all this information becomes increasingly challenging, which opens an opportunity for developing clinical decision support systems such as patient trajectory modelling solutions. Patient trajectory modelling is a topic of growing research interest due to its potential to help improving health care quality by fostering preventive medicine practices, since an earlier disease diagnosis can enable better disease management and earlier intervention, along with an improved resource allocation. In this paper, we review recent approaches for patient trajectory prediction, performing a comparison based on three key aspects: 1) what is the core of the developed approach, 2) what type of data is used in the work, and 3) how is temporal information handled in the proposed solution. The resulting selection of works herein presented illustrates the current paradigm in patient trajectory modelling, and provides an overview on some of the existing challenges in this field.

Transformer-based “behemoths” have grown in popularity, as well as structurally, shattering multiple NLP benchmarks along the way. However, their real-world usability remains a question. In this work, we empirically assess the feasibility of applying transformer-based models in real-world ad-hoc retrieval applications by comparison to a “greener and more sustainable” alternative, comprising only 620 trainable parameters. We present an analysis of their efficacy and efficiency and show that considering limited computational resources, the lighter model running on the CPU achieves a 3 to 20 times speedup in training and 7 to 47 times in inference while maintaining a comparable retrieval performance. Code to reproduce the efficiency experiments is available on “https://github.com/bioinformatics-ua/EACL2021-reproducibility/“.

Cardiovascular diseases are the leading cause of death worldwide. These diseases are related with a broad range of factors but usually show high correlation with diminished left ventricle function, which can be evaluated by measuring the ventricular ejection fraction through transthoracic echocardiography (TTE), a cost-effective and highly portable first-line diagnosing technique. Ejection fraction (EF) is currently determined through a semi-automatic process that requires manual delineation of the left ventricle area both in a diastolic and systolic frame of the patient's exam. To remove this manual annotation step, which is both time-consuming and user dependent, automatic Computer-Aided Diagnosis (CAD) systems can be used. Herein, we propose the first steps for such a system that classifies ejection fraction in four classes, based on TTE exams, with the objective of automatically providing valuable information to physicians. Our classification method is based on a 3D-Convolutional Neural Network (3D-CNN) trained on a dataset constructed with exams from a cardiology reference center. The dataset creation consisted of three main steps: firstly, for each exam, cine-loops showing the apical 4 chambers view were manually selected; then, 30 sequential frames were extracted from each cine-loop; finally, each frame was pre-processed to mask burned-in metadata. The neural network was designed to explore concepts such as convolutions using asymmetric filters and residual learning blocks. The model was trained on a dataset with 4000 TTE exams and tested on a separate dataset containing 1600 TTE cases. We obtained an accuracy of 78% and a F1 score of 71.3% for unhealthy EF (below 45%), 63.3% for intermediate EF (45-55%), 72.3% for healthy EF (55-75%) and 54.6% for abnormally high EF (above 75%). These results are promising and show that convolutional neural networks can be applied to this domain. Furthermore, this work will serve as a foundation for future research where other relevant cardiac metrics will be determined.

The annotation of concepts and susceptible interactions has been assuming a key role in the extraction of relevant information from published documents. However, distinct annotation tools generate also different formats, creating a barrier to efficiently combine and exchange this information. The migration of curated information into semantic web format and services provides an additional value to share that knowledge, but data transformation represents here an additional challenge. In this manuscript, we present a unified layer between text-mining tools and semantic web services to reduce the effort of combining different formats. The Ann2RDF is focused on reusing existing curated data from external text-mining tools to improve their availability through an open representation model. This result in a more suitable transition process, in which desired annotations are enriched with the possibility to be shared, compared and reused across semantic Knowledge Bases.

SUMMARY: The continuous growth of the biomedical scientific literature has been motivating the development of text-mining tools able to efficiently process all this information. Although numerous domain-specific solutions are available, there is no web-based concept-recognition system that combines the ability to select multiple concept types to annotate, to reference external databases and to automatically annotate nested and intercepted concepts. BeCAS, the Biomedical Concept Annotation System, is an API for biomedical concept identification and a web-based tool that addresses these limitations. MEDLINE abstracts or free text can be annotated directly in the web interface, where identified concepts are enriched with links to reference databases. Using its customizable widget, it can also be used to augment external web pages with concept highlighting features. Furthermore, all text-processing and annotation features are made available through an HTTP REST API, allowing integration in any text-processing pipeline. AVAILABILITY: BeCAS is freely available for non-commercial use at http://bioinformatics.ua.pt/becas. CONTACTS: [email protected] or [email protected].

BACKGROUND: Automatic recognition of biomedical names is an essential task in biomedical information extraction, presenting several complex and unsolved challenges. In recent years, various solutions have been implemented to tackle this problem. However, limitations regarding system characteristics, customization and usability still hinder their wider application outside text mining research. RESULTS: We present Gimli, an open-source, state-of-the-art tool for automatic recognition of biomedical names. Gimli includes an extended set of implemented and user-selectable features, such as orthographic, morphological, linguistic-based, conjunctions and dictionary-based. A simple and fast method to combine different trained models is also provided. Gimli achieves an F-measure of 87.17% on GENETAG and 72.23% on JNLPBA corpus, significantly outperforming existing open-source solutions. CONCLUSIONS: Gimli is an off-the-shelf, ready to use tool for named-entity recognition, providing trained and optimized models for recognition of biomedical entities from scientific text. It can be used as a command line tool, offering full functionality, including training of new models and customization of the feature set and model parameters through a configuration file. Advanced users can integrate Gimli in their text mining workflows through the provided library, and extend or adapt its functionalities. Based on the underlying system characteristics and functionality, both for final users and developers, and on the reported performance results, we believe that Gimli is a state-of-the-art solution for biomedical NER, contributing to faster and better research in the field. Gimli is freely available at http://bioinformatics.ua.pt/gimli.

BACKGROUND: Concept recognition is an essential task in biomedical information extraction, presenting several complex and unsolved challenges. The development of such solutions is typically performed in an ad-hoc manner or using general information extraction frameworks, which are not optimized for the biomedical domain and normally require the integration of complex external libraries and/or the development of custom tools. RESULTS: This article presents Neji, an open source framework optimized for biomedical concept recognition built around four key characteristics: modularity, scalability, speed, and usability. It integrates modules for biomedical natural language processing, such as sentence splitting, tokenization, lemmatization, part-of-speech tagging, chunking and dependency parsing. Concept recognition is provided through dictionary matching and machine learning with normalization methods. Neji also integrates an innovative concept tree implementation, supporting overlapped concept names and respective disambiguation techniques. The most popular input and output formats, namely Pubmed XML, IeXML, CoNLL and A1, are also supported. On top of the built-in functionalities, developers and researchers can implement new processing modules or pipelines, or use the provided command-line interface tool to build their own solutions, applying the most appropriate techniques to identify heterogeneous biomedical concepts. Neji was evaluated against three gold standard corpora with heterogeneous biomedical concepts (CRAFT, AnEM and NCBI disease corpus), achieving high performance results on named entity recognition (F1-measure for overlap matching: species 95%, cell 92%, cellular components 83%, gene and proteins 76%, chemicals 65%, biological processes and molecular functions 63%, disorders 85%, and anatomical entities 82%) and on entity normalization (F1-measure for overlap name matching and correct identifier included in the returned list of identifiers: species 88%, cell 71%, cellular components 72%, gene and proteins 64%, chemicals 53%, and biological processes and molecular functions 40%). Neji provides fast and multi-threaded data processing, annotating up to 1200 sentences/second when using dictionary-based concept identification. CONCLUSIONS: Considering the provided features and underlying characteristics, we believe that Neji is an important contribution to the biomedical community, streamlining the development of complex concept recognition solutions. Neji is freely available at http://bioinformatics.ua.pt/neji.

MOTIVATION: The recognition of named entities (NER) is an elementary task in biomedical text mining. A number of NER solutions have been proposed in recent years, taking advantage of available annotated corpora, terminological resources and machine-learning techniques. Currently, the best performing solutions combine the outputs from selected annotation solutions measured against a single corpus. However, little effort has been spent on a systematic analysis of methods harmonizing the annotation results and measuring against a combination of Gold Standard Corpora (GSCs). RESULTS: We present Totum, a machine learning solution that harmonizes gene/protein annotations provided by heterogeneous NER solutions. It has been optimized and measured against a combination of manually curated GSCs. The performed experiments show that our approach improves the F-measure of state-of-the-art solutions by up to 10% (achieving ≈70%) in exact alignment and 22% (achieving ≈82%) in nested alignment. We demonstrate that our solution delivers reliable annotation results across the GSCs and it is an important contribution towards a homogeneous annotation of MEDLINE abstracts. AVAILABILITY AND IMPLEMENTATION: Totum is implemented in Java and its resources are available at http://bioinformatics.ua.pt/totum

BACKGROUND: Determining usefulness of biomedical text mining systems requires realistic task definition and data selection criteria without artificial constraints, measuring performance aspects that go beyond traditional metrics. The BioCreative III Protein-Protein Interaction (PPI) tasks were motivated by such considerations, trying to address aspects including how the end user would oversee the generated output, for instance by providing ranked results, textual evidence for human interpretation or measuring time savings by using automated systems. Detecting articles describing complex biological events like PPIs was addressed in the Article Classification Task (ACT), where participants were asked to implement tools for detecting PPI-describing abstracts. Therefore the BCIII-ACT corpus was provided, which includes a training, development and test set of over 12,000 PPI relevant and non-relevant PubMed abstracts labeled manually by domain experts and recording also the human classification times. The Interaction Method Task (IMT) went beyond abstracts and required mining for associations between more than 3,500 full text articles and interaction detection method ontology concepts that had been applied to detect the PPIs reported in them. RESULTS: A total of 11 teams participated in at least one of the two PPI tasks (10 in ACT and 8 in the IMT) and a total of 62 persons were involved either as participants or in preparing data sets/evaluating these tasks. Per task, each team was allowed to submit five runs offline and another five online via the BioCreative Meta-Server. From the 52 runs submitted for the ACT, the highest Matthew's Correlation Coefficient (MCC) score measured was 0.55 at an accuracy of 89% and the best AUC iP/R was 68%. Most ACT teams explored machine learning methods, some of them also used lexical resources like MeSH terms, PSI-MI concepts or particular lists of verbs and nouns, some integrated NER approaches. For the IMT, a total of 42 runs were evaluated by comparing systems against manually generated annotations done by curators from the BioGRID and MINT databases. The highest AUC iP/R achieved by any run was 53%, the best MCC score 0.55. In case of competitive systems with an acceptable recall (above 35%) the macro-averaged precision ranged between 50% and 80%, with a maximum F-Score of 55%. CONCLUSIONS: The results of the ACT task of BioCreative III indicate that classification of large unbalanced article collections reflecting the real class imbalance is still challenging. Nevertheless, text-mining tools that report ranked lists of relevant articles for manual selection can potentially reduce the time needed to identify half of the relevant articles to less than 1/4 of the time when compared to unranked results. Detecting associations between full text articles and interaction detection method PSI-MI terms (IMT) is more difficult than might be anticipated. This is due to the variability of method term mentions, errors resulting from pre-processing of articles provided as PDF files, and the heterogeneity and different granularity of method term concepts encountered in the ontology. However, combining the sophisticated techniques developed by the participants with supporting evidence strings derived from the articles for human interpretation could result in practical modules for biological annotation workflows.

BACKGROUND: We report the Gene Normalization (GN) challenge in BioCreative III where participating teams were asked to return a ranked list of identifiers of the genes detected in full-text articles. For training, 32 fully and 500 partially annotated articles were prepared. A total of 507 articles were selected as the test set. Due to the high annotation cost, it was not feasible to obtain gold-standard human annotations for all test articles. Instead, we developed an Expectation Maximization (EM) algorithm approach for choosing a small number of test articles for manual annotation that were most capable of differentiating team performance. Moreover, the same algorithm was subsequently used for inferring ground truth based solely on team submissions. We report team performance on both gold standard and inferred ground truth using a newly proposed metric called Threshold Average Precision (TAP-k). RESULTS: We received a total of 37 runs from 14 different teams for the task. When evaluated using the gold-standard annotations of the 50 articles, the highest TAP-k scores were 0.3297 (k=5), 0.3538 (k=10), and 0.3535 (k=20), respectively. Higher TAP-k scores of 0.4916 (k=5, 10, 20) were observed when evaluated using the inferred ground truth over the full test set. When combining team results using machine learning, the best composite system achieved TAP-k scores of 0.3707 (k=5), 0.4311 (k=10), and 0.4477 (k=20) on the gold standard, representing improvements of 12.4%, 21.8%, and 26.6% over the best team results, respectively. CONCLUSIONS: By using full text and being species non-specific, the GN task in BioCreative III has moved closer to a real literature curation task than similar tasks in the past and presents additional challenges for the text mining community, as revealed in the overall team results. By evaluating teams using the gold standard, we show that the EM algorithm allows team submissions to be differentiated while keeping the manual annotation effort feasible. Using the inferred ground truth we show measures of comparative performance between teams. Finally, by comparing team rankings on gold standard vs. inferred ground truth, we further demonstrate that the inferred ground truth is as effective as the gold standard for detecting good team performance.

BACKGROUND: Advances in biotechnology and in high-throughput methods for gene analysis have contributed to an exponential increase in the number of scientific publications in these fields of study. While much of the data and results described in these articles are entered and annotated in the various existing biomedical databases, the scientific literature is still the major source of information. There is, therefore, a growing need for text mining and information retrieval tools to help researchers find the relevant articles for their study. To tackle this, several tools have been proposed to provide alternative solutions for specific user requests. RESULTS: This paper presents QuExT, a new PubMed-based document retrieval and prioritization tool that, from a given list of genes, searches for the most relevant results from the literature. QuExT follows a concept-oriented query expansion methodology to find documents containing concepts related to the genes in the user input, such as protein and pathway names. The retrieved documents are ranked according to user-definable weights assigned to each concept class. By changing these weights, users can modify the ranking of the results in order to focus on documents dealing with a specific concept. The method's performance was evaluated using data from the 2004 TREC genomics track, producing a mean average precision of 0.425, with an average of 4.8 and 31.3 relevant documents within the top 10 and 100 retrieved abstracts, respectively. CONCLUSIONS: QuExT implements a concept-based query expansion scheme that leverages gene-related information available on a variety of biological resources. The main advantage of the system is to give the user control over the ranking of the results by means of a simple weighting scheme. Using this approach, researchers can effortlessly explore the literature regarding a group of genes and focus on the different aspects relating to these genes.