Contrastive Learning on Medical Intents for Sequential Prescription Recommendation

Published: Oct. 20, 2024

Recent advancements in sequential modeling applied to Electronic Health Records (EHR) have greatly influenced prescription recommender systems.While the recent literature on drug recommendation has shown promising performance, study of discovering a diversity coexisting temporal relationships at level medical codes over consecutive visits remains less explored.The goal this can be motivated from two perspectives.First, there is need develop sophisticated model capable disentangling complex across visits.Second, it crucial establish multiple and diverse health profiles for same patient ensure comprehensive consideration different intents recommendation.To achieve goal, we introduce Attentive Recommendation with Contrasted Intents (ARCI), multi-level transformer-based method designed capture but paths shared sequence visits.Specifically, propose novel intent-aware contrastive learning, that links specialized patients transformer heads extracting distinct associated profiles.We conducted experiments real-world datasets task using both ranking classification metrics.Our results demonstrate ARCI outperformed state-ofthe-art methods providing interpretable insights healthcare practitioners.

Language: Английский

---

Task-Specific Transformer-Based Language Models in Health Care: A Scoping Review (Preprint)

JMIR Medical Informatics, Journal Year: 2024, Volume and Issue: 12, P. e49724 - e49724

Published: Oct. 21, 2024

Background Transformer-based language models have shown great potential to revolutionize health care by advancing clinical decision support, patient interaction, and disease prediction. However, despite their rapid development, the implementation of transformer-based in settings remains limited. This is partly due lack a comprehensive review, which hinders systematic understanding applications limitations. Without clear guidelines consolidated information, both researchers physicians face difficulties using these effectively, resulting inefficient research efforts slow integration into workflows. Objective scoping review addresses this gap examining studies on medical categorizing them 6 tasks: dialogue generation, question answering, summarization, text classification, sentiment analysis, named entity recognition. Methods We conducted following Cochrane protocol. A literature search was performed across databases, including Google Scholar PubMed, covering publications from January 2017 September 2024. Studies involving transformer-derived tasks were included. Data categorized key tasks. Results Our findings revealed advancements critical challenges applying For example, like MedPIR generation show promise but privacy ethical concerns, while question-answering BioBERT improve accuracy struggle with complexity terminology. The BioBERTSum summarization model aids clinicians condensing texts needs better handling long sequences. Conclusions attempted provide role guide future directions. By addressing current exploring for real-world applications, we envision significant improvements informatics. Addressing identified implementing proposed solutions can enable significantly delivery outcomes. provides valuable insights practical setting stage transformative

Language: Английский

---

Transfer learning enabled transformer-based generative adversarial networks for modeling and generating terahertz channels

Communications Engineering, Journal Year: 2024, Volume and Issue: 3(1)

Published: Nov. 2, 2024

Terahertz communications are envisioned as a promising technology for the sixth generation and beyond wireless systems, which can support links with Terabits-per-second (Tbps) data rates. As foundation of designing terahertz communications, channel modeling characterization crucial to scrutinize potential this spectrum. However, current in band heavily relies on time-consuming costly measurements. Here, we propose transfer learning enabled transformer based generative adversarial network mitigate problem modeling. Specifically, fundamental building block, is exploited generate parameters. To improve accuracy, structure self-attention mechanism incorporated network. Still incurring errors compared ground-truth measurement, designed solve mismatch between formulated measurement. The proposed method achieve high accuracy modeling, while requiring only rather limited amount complement techniques.

Language: Английский

---

Big Epidemiology: The Birth, Life, Death, and Resurgence of Diseases on a Global Timescale

Epidemiologia, Journal Year: 2024, Volume and Issue: 5(4), P. 669 - 691

Published: Nov. 6, 2024

Big Epidemiology represents an innovative framework that extends the interdisciplinary approach of History to understand disease patterns, causes, and effects across human history on a global scale. This comprehensive methodology integrates epidemiology, genetics, environmental science, sociology, history, data science address contemporary future public health challenges through broad historical societal lens. The foundational research agenda involves mapping occurrence diseases their impact societies over time, utilizing archeological findings, biological data, records. By analyzing skeletal remains, ancient DNA, artifacts, researchers can trace origins spread diseases, such as

Language: Английский

---

Identification of Myocardial Infarction (MI) Probability from Imbalanced Medical Survey Data: An Artificial Neural Network (ANN) with Explainable AI (XAI) Insights

medRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Feb. 29, 2024

ABSTRACT In the healthcare industry, many artificial intelligence (AI) models have attempted to overcome bias from class imbalances while also maintaining high results. Firstly, when utilizing a large number of unbalanced samples, current AI and related research failed balance specificity sensitivity – problem that can undermine reliability medical research. Secondly, no reliable method for obtaining detailed interpretability has been put forth addressing numbers input features. The present addresses these two critical gaps with proposed lightweight Artificial Neural Network (ANN) model. Using 43 features 2021 Behavioral Risk Factor Surveillance System (BRFSS) dataset, model outperforms prior in producing balanced outcomes markedly survey data. efficacy this ANN is attributed its simplified design, which reduces processing demands, resilience identifying probability myocardial infarction (MI). This demonstrated by 80% 77% sensitivity, substantiated Receiver Operating Characteristic Area Under Curve (AUC) 0.87. across scopes each specified data domain were separately represented, thus demonstrating model’s robust sensitivity. model, as measured Shapley values, reveals substantial correlations between (MI) risk factors, including long-term conditions, socio-demographic personal health habits, economic social status, availability affordability, well impairment statuses, providing valuable insights improved cardiovascular assessment personalized strategies.

Language: Английский

---