Clinical time series are often irregularly sampled, with varying sensor frequencies, missing observations, and misaligned timestamps. Prior approaches typically address these irregularities by interpolating data into regular sequences, thereby introducing bias, or by generating inconsistent and uninterpretable relationships across sensor measurements, complicating the accurate learning of both intra-series and inter-series dependencies. We introduce WaveGNN, a model that operates directly on irregular multivariate time series without interpolation or conversion to a regular representation. WaveGNN combines a decay-aware Transformer to capture intra-series dynamics with a sample-specific graph neural network that models both short-term and long-term inter-sensor relationships. Therefore, it generates a single, sparse, and interpretable graph per sample. Across multiple benchmark datasets (P12, P19, MIMIC-III, and PAM), WaveGNN delivers consistently strong performance, whereas other state-of-the-art baselines tend to perform well on some datasets or tasks but poorly on others. While WaveGNN does not necessarily surpass every method in every case, its consistency and robustness across diverse settings set it apart. Moreover, the learned graphs align well with known physiological structures, enhancing interpretability and supporting clinical decision-making.
Federated learning (FL) for distributed data has gained significant attention by enabling model training on local data without transferring it to a central system. While this approach protects sensitive information, risks of data leakage still persist, necessitating the integration of privacy-preserving techniques such as differential privacy. In many FL applications, tasks like exploratory data analysis or tracking and monitoring data that change over time are essential. For these purposes, analysts rely on data visualizations to make decisions or draw conclusions. This vision paper emphasizes the importance of federated privacy-preserving visualization and outlines a general pipeline for its implementation. We discuss the challenges of integrating federated visualizations with differential privacy and demonstrate the feasibility of this approach through examples, such as federated privacy-preserving boxplots, scatterplots, and correlation visualizations in neuroimaging. This highlights the need for further research in this promising field.
Learning from massive amounts of domain-specific information requires new algorithms and models for parsing the ever-expanding field of big data. Such algorithms for exploring and identifying key features in vast databases require analysis of complex interactions to uncover critical features under a variety of circumstances. We study a comprehensive collection of health-related data, showing that our novel Choquet Integral activation function for deep neural networks transforms high-dimensional data into simpler sub-feature sets that better model complex interactions. While standard methods account for unitary feature tracking, they do not extend to multiple feature subsets, an impactful and necessary knowledge base. To this end, our novel activation function creates a sub-additive tool that better considers the weighted compilation of features within a robust set of standard benchmarks, advancing the synergistic and antagonistic relationships among features, capturing non-linear dependencies. We present the theoretical underpinnings, highlighting balanced fuzzy measures and sub-additivity for an optimized model based on real-world health data targeting weight loss. We further test different model settings, akin to hyper-parameter optimization. Despite computational time consumption, which could be improved via nowadays more powerful computing units, this novel method can be implemented as a pre-trained model using big data to identify heretofore unknown sub-additive feature interactions in a variety of fields such as biomedicine, fraud detection, cyber-security, and finance.
This paper presents CLINICSUM, a novel framework designed to automatically generate clinical summaries from patient-doctor conversations. It utilizes a two-module architecture: a retrieval-based filtering module that extracts Subjective, Objective, Assessment, and Plan (SOAP) information from conversation transcripts, and an inference module powered by fine-tuned Pre-trained Language Models (PLMs), which leverage the extracted SOAP data to generate abstracted clinical summaries. To fine-tune the PLM, we created a training dataset of consisting 1,473 conversations-summaries pair by consolidating two publicly available datasets, FigShare and MTS-Dialog, with ground truth summaries validated by Subject Matter Experts (SMEs). CLINICSUM’s effectiveness is evaluated through both automatic metrics (e.g., ROUGE, BERTScore) and expert human assessments. Results show that CLINICSUM outperforms state-of-the-art PLMs, demonstrating superior precision, recall, and F-1 scores in automatic evaluations and receiving high preference from SMEs in human assessment, making it a robust solution for automated clinical summarization.
The widespread adoption of large language models (LLMs) has created an urgent need for robust tools to detect LLM-generated text, especially in light of paraphrasing techniques that often evade existing detection methods. To address this challenge, we present a novel semantic-enhanced framework for detecting LLM-generated text (SEFD) that leverages a retrieval-based mechanism to fully utilize text semantics. Our framework improves upon existing detection methods by systematically integrating retrieval-based techniques with traditional detectors, employing a carefully curated retrieval mechanism that strikes a balance between comprehensive coverage and computational efficiency. We showcase the effectiveness of our approach in sequential text scenarios common in real-world applications, such as online forums and Q&A platforms. Through comprehensive experiments across various LLM-generated texts and detection methods, we demonstrate that our framework substantially enhances detection accuracy in paraphrasing scenarios while maintaining robustness for standard LLM-generated content. This work contributes significantly to ongoing efforts to safeguard information integrity in an era where AI-generated content is increasingly prevalent.
Hematopoietic stem cell transplantation (HCT) is a potentially life-saving treatment that uses healthy blood-forming cells from donors to replace dysfunctional or damaged hematopoietic cells in patients with various blood disorders. This procedure is often employed to treat conditions such as hematological malignancies (e.g., leukemia, lymphoma, myeloma) and other severe blood or immune system diseases. Monitoring post-transplant complications is essential for tracking physiological effects and aiding in clinical decision-making. Biobehavioral aspects of care partners (i.e., unpaid caregivers) can also be influenced during the post-transplant stage of HCT. Wearable devices offer a non-invasive way to continuously track physiological parameters, making them a valuable resource for health monitoring. However, the physiological data collected from wearables is highly unstructured, often containing missing values, outliers, redundant features, and erroneous measurements leading to false conclusions/prediction. Therefore, enhancing data quality is essential for deriving meaningful insights. This paper introduces novel pre-processing methods to build a high quality, comprehensive, standardized, AI/ML ready, and clinically meaningful wearable dataset of HCT patients and caregivers. To test our data cleaning implementation, our cleaned, high-quality dataset is utilized to predict mood and compliance in HCT patients and their caregivers using machine learning algorithms. The paper illustrates our proposed approach and presents experimental results conducted on the data collected from Michigan Medicine for HCT patients and caregivers. Our preliminary experimental results are promising, demonstrating the effectiveness of the proposed methods and the high-quality dataset in predicting mood and compliance for the participants.
Predicting seizures ahead of time will have a significant positive clinical impact for people with epilepsy. Advances in machine learning/artificial intelligence (ML/AI) has provided us the tools needed to perform such predictive tasks. To date, advanced deep learning (DL) architectures such as the convolutional neural network (CNN) and long short-term memory (LSTM) have been used with mixed results. However, highly connected activity exhibited by epileptic seizures necessitates the design of more complex ML techniques which can better capture the complex interconnected neurological processes. Other challenges include the variability of EEG sensor data quality, different epilepsy and seizure profiles, lack of annotated datasets and absence of ML-ready benchmarks. In addition, successful models will need to perform inference in almost real-time using limited hardware compute-capacity. To address these challenges, we propose a lightweight architecture, called ESPFormer, whose novelty lies in the simple and smaller model-size and a lower computational load footprint needed to infer in real-time compared to other works in the literature. To quantify the performance of this lightweight model, we compared its performance with a custom-designed residual neural network (ResNet), a pre-trained vision transformer (ViT) and a pre-trained large-language model (LLM). We tested ESPFormer on MLSPred-Bench which is the largest patient-independent seizure prediction dataset comprising 12 benchmarks. Our results demonstrate that ESPFormer provides the best performance in terms of prediction accuracy for 4/12 benchmarks with an average improvement of 2.65% compared to the LLM, 3.35% compared to the ViT and 17.65% compared to the ResNet - and comparable results for other benchmarks. Our results indicate that lightweight transformer architecture may outperform resource-intensive LLM based models for real-time EEG-based seizure predictions.
Machine learning foundation models, particularly large language models (LLMs) such as GPT-4o, have revolutionized traditional applications in computer vision and natural language processing, marking a significant shift in recent years. Building on these advancements, recent efforts have explored the potential of foundation models in hypothesis generation, highlighting their possibility in aiding human researchers in scientific discovery. In this paper, we envision a future where academia increasingly integrates foundation models to accelerate and enhance the process of scientific discovery. Motivated by potential application scenarios of foundation models in scientific research, our vision is anchored in a central question: How can we accelerate scientific discovery with the aid of foundation models? To address this overarching question, we raise two key challenges that need to be addressed: (1) how to effectively harness the parametric knowledge embedded in foundation models to propel scientific discovery? and (2) how to develop rigorous yet scalable methods to evaluate the effectiveness of foundation models in supporting scientific research? To tackle these two challenges, we propose our approaches, termed knowledge-grounded Chain-of-Idea (KG-CoI) hypothesis generation and IdeaBench - Benchmarking LLM hypothesis generators in a customizable manner. Through addressing these challenges, we outline our vision in hope to inspire new ideas and innovations in harnessing foundation models for advancing scientific discovery, paving the way for a new era of research collaboration between humans and artificial intelligence.
With social media communities increasingly becoming places where suicidal individuals post and congregate, natural language processing presents an exciting avenue for the development of automated suicide risk assessment systems. However, past efforts suffer from a lack of labeled data and class imbalances within the available labeled data. To accommodate this task’s imperfect data landscape, we propose a semi-supervised framework that leverages labeled (n=500) and unlabeled (n=1,500) data and expands upon the self-training algorithm with a novel pseudo-label acquisition process designed to handle imbalanced datasets. To further ensure pseudo-label quality, we manually verify a subset of the pseudo-labeled data that was not predicted unanimously across multiple trials of pseudo-label generation. We test various models to serve as the backbone for this framework, ultimately deciding that RoBERTa performs the best. Ultimately, by leveraging partially validated pseudo-labeled data in addition to ground-truth labeled data, we substantially improve our model’s ability to assess suicide risk from social media posts.
Demographics, Social determinants of health, and family history documented in the unstructured text within the electronic health records are increasingly being studied to understand how this information can be utilized with the structured data to improve healthcare outcomes. After the GPT models were released, many studies have applied GPT models to extract this information from the narrative clinical notes. Different from the existing work, our research focuses on investigating the zero-shot learning on extracting this information together by providing minimum information to the GPT model. We utilize de-identified real-world clinical notes annotated for demographics, various social determinants, and family history information. Given that the GPT model might provide text different from the text in the original data, we explore two sets of evaluation metrics, including the traditional NER evaluation metrics and semantic similarity evaluation metrics, to completely understand the performance. Our results show that the GPT-3.5 method achieved an average of 0.975 F1 on demographics extraction, 0.615 F1 on social determinants extraction, and 0.722 F1 on family history extraction. We believe these results can be further improved through model fine-tuning or few-shots learning. Through the case studies, we also identified the limitations of the GPT models, which need to be addressed in future research.
Autism spectrum disorder (ASD) affects large number of children and adults in the US, and worldwide. Early and quick diagnosis of ASD can improve the quality of life significantly both for patients and their families. Prior research provides strong evidence that structural and functional magnetic resonance imaging (MRI) data collected from individuals with ASD exhibit distinguishing characteristics that differ in local and global, spatial and temporal neural patterns of the brain – and therefore can be used for diagnostic purposes for various mental disorders. However, the data from MRI are high-dimensional and advanced methods are needed to make sense out of these datasets. In this paper, we present a novel model based on graph convolutional network (GCN) that can utilize resting state fMRI (rs-fMRI) data to classify ASD subjects from health controls (HC). In addition to using the graph from traditional correlation matrices, our proposed GCN model incorporates graphlet topological counting as one of the training features. Our results show that graphlets can preserve the topological information of the graphs obtained from fMRI data. Combined with our GCN, the graphlets retain enough topological information to differentiate between the ASD and HC. Our proposed model gives an average accuracy of 64.27% on the whole ABIDE-I data sets (1035 subjects) and highest site-specific accuracy of 75.9%, which is comparable to other state-of-the-art methods – while potentially open to being more interpretable.
Effective disease surveillance systems require large-scale epidemiological data to improve health outcomes and quality of care for the general population. As data may be limited within a single site, multi-site data (e.g., from a number of local/regional health systems) need to be considered. Leveraging distributed data across multiple sites for epidemiological analysis poses significant challenges. Due to the sensitive nature of epidemiological data, it is imperative to design distributed solutions that provide strong privacy protections. Current privacy solutions often assume a central site, which is responsible for aggregating the distributed data and applying privacy protection before sharing the results (e.g., aggregation via secure primitives and differential privacy for sharing aggregate results). However, identifying such a central site may be difficult in practice and relying on a central site may introduce potential vulnerabilities (e.g., single point of failure). Furthermore, to support clinical interventions and inform policy decisions in a timely manner, epidemiological analysis need to reflect dynamic changes in the data. Yet, existing distributed privacy-protecting approaches were largely designed for static data (e.g., one-time data sharing) and cannot fulfill dynamic data requirements. In this work, we propose a privacy-protecting approach that supports the sharing of dynamic epidemiological analysis and provides strong privacy protection in a decentralized manner. We apply our solution in continuous survival analysis using the Kaplan-Meier estimation model while providing differential privacy protection. Our evaluations on a real dataset containing COVID-19 cases show that our method provides highly usable results.
Sharing data and computation among concurrent queries has been an active research topic in database systems. While work in this area developed algorithms and systems that are shown to be effective, there is a lack of logical foundation for query processing and optimization. In this paper, we present PsiDB, a system model for processing a large number of database queries in a batch. The key idea is to generate a single query expression that returns a global relation containing all the data needed for individual queries. For that, we propose the use of a type of relational operators called ψ-operators in combining the individual queries into the global expression. We tackle the algebraic optimization problem in PsiDB by developing equivalence rules to transform concurrent queries with the purpose of revealing query optimization opportunities. Centering around the ψ-operator, our rules not only cover many optimization techniques adopted in existing batch processing systems, but also revealed new optimization opportunities. Experiments conducted on an early prototype of PsiDB show a performance improvement of up to 36X over a mainstream commercial DBMS.
Healthcare workers such as doctors and nurses are expected to be trustworthy and creditable sources of vaccine-related information. Their opinions toward the COVID-19 vaccines may influence the vaccine uptake among the general population. However, vaccine hesitancy is still an important issue even among the healthcare workers. Therefore, it is critical to understand their opinions to help reduce the level of vaccine hesitancy. There have been studies examining healthcare workers' viewpoints on COVID-19 vaccines using questionnaires. Reportedly, a considerably higher proportion of vaccine hesitancy is observed among nurses, compared to doctors. We intend to verify and study this phenomenon at a much larger scale and in fine grain using social media data, which has been effectively and efficiently leveraged by researchers to address real-world issues during the COVID-19 pandemic. More specifically, we use a keyword search to identify healthcare workers and further classify them into doctors and nurses from the profile descriptions of the corresponding Twitter users. Moreover, we apply a transformer-based language model to remove irrelevant tweets. Sentiment analysis and topic modeling are employed to analyze and compare the sentiment and thematic differences in the tweets posted by doctors and nurses. We find that doctors are overall more positive toward the COVID-19 vaccines. The focuses of doctors and nurses when they discuss vaccines in a negative way are in general different. Doctors are more concerned with the effectiveness of the vaccines over newer variants while nurses pay more attention to the potential side effects on children. Therefore, we suggest that more customized strategies should be deployed when communicating with different groups of healthcare workers.
This paper presents a novel use case of Graph Convolutional Network (GCN) learning representations for predictive data mining, specifically from user/task data in the domain of high-performance computing (HPC). It outlines an approach based on a coalesced data set: logs from the Slurm workload manager, joined with user experience survey data from computational cluster users. We introduce a new method of constructing a heterogeneous unweighted HPC graph consisting of multiple typed nodes after revealing the manifold relations between the nodes. The GCN structure used here supports two tasks: i) determining whether a job will complete or fail and ii) predicting memory and CPU requirements by training the GCN semi-supervised classification model and regression models on the generated graph. The graph is partitioned into partitions using graph clustering. We conducted classification and regression experiments using the proposed framework on our HPC log dataset and evaluated predictions by our trained models against baselines using test score, F1-score, precision, recall for classification, and R1 score for regression, showing that our framework achieves significant improvements.
Support vector machine (SVM) is a popular classification method for the analysis of a wide range of data including big biomedical data. Many SVM methods with feature selection have been developed under the frequentist regularization or Bayesian shrinkage frameworks. On the other hand, the value of incorporating a priori known biological knowledge, such as those from functional genomics and functional proteomics, into statistical analysis of -omic data has been recognized in recent years. Such biological information is often represented by graphs. We propose a novel method that assigns Laplace priors to the regression coefficients and incorporates the underlying graph information via a hyper-prior for the shrinkage parameters in the Laplace priors. This enables smoothing of shrinkage parameters for connected variables in the graph and conditional independence between shrinkage parameters for disconnected variables. Extensive simulations demonstrate that our proposed methods achieve the best performance compared to the other existing SVM methods in terms of prediction accuracy. The proposed method are also illustrated in analysis of genomic data from cancer studies, demonstrating its advantage in generating biologically meaningful results and identifying potentially important features.
Retrospective data harmonization across multiple research cohorts and studies is frequently done to increase statistical power, provide comparison analysis, and create a richer data source for data mining. However, when combining disparate data sources, harmonization projects face data management and analysis challenges. These include differences in the data dictionaries and variable definitions, privacy concerns surrounding health data representing sensitive populations, and lack of properly defined data models. With the availability of mature open-source web-based database technologies, developing a complete software architecture to overcome the challenges associated with the harmonization process can alleviate many roadblocks. By leveraging state-of-the-art software engineering and database principles, we can ensure data quality and enable cross-center online access and collaboration. This paper outlines a complete software architecture developed and customized using the Django web framework, leveraged to harmonize sensitive data collected from three NIH-support birth cohorts. We describe our framework and show how we successfully overcame challenges faced when harmonizing data from these cohorts. We discuss our efforts in data cleaning, data sharing, data transformation, data visualization, and analytics, while reflecting on what we have learned to date from these harmonized datasets.
Our aging population increasingly suffers from multiple chronic diseases simultaneously, necessitating the comprehensive treatment of these conditions. Finding the optimal set of drugs for a combinatorial set of diseases is a combinatorial pattern exploration problem. Association rule mining is a popular tool for such problems, but the requirement of health care for finding causal, rather than associative, patterns renders association rule mining unsuitable. To address this issue, we propose a novel framework based on the Rubin-Neyman causal model for extracting causal rules from observational data, correcting for a number of common biases. Specifically, given a set of interventions and a set of items that define subpopulations (e.g., diseases), we wish to find all subpopulations in which effective intervention combinations exist and in each such subpopulation, we wish to find all intervention combinations such that dropping any intervention from this combination will reduce the efficacy of the treatment. A key aspect of our framework is the concept of closed intervention sets which extend the concept of quantifying the effect of a single intervention to a set of concurrent interventions. Closed intervention sets also allow for a pruning strategy that is strictly more efficient than the traditional pruning strategy used by the Apriori algorithm. To implement our ideas, we introduce and compare five methods of estimating causal effect from observational data and rigorously evaluate them on synthetic data to mathematically prove (when possible) why they work. We also evaluated our causal rule mining framework on the Electronic Health Records (EHR) data of a large cohort of 152000 patients from Mayo Clinic and showed that the patterns we extracted are sufficiently rich to explain the controversial findings in the medical literature regarding the effect of a class of cholesterol drugs on Type-II Diabetes Mellitus (T2DM).
We present a collection of geodatabase functions which expedite utilizing differential privacy for privacy-aware geospatial analysis of healthcare data. The healthcare domain has a long history of standardization and research communities have developed open-source common data models to support the larger goals of interoperability, reproducibility, and data sharing; these models also standardize geospatial patient data. However, patient privacy laws and institutional regulations complicate geospatial analyses and dissemination of research findings due to protective restrictions in how data and results are shared. This results in infrastructures with great abilities to organize and store healthcare data, yet which lack the innate ability to produce shareable results that preserve privacy and conform to regulatory requirements. Differential privacy is a model for performing privacy-preserving analytics. We detail our process and findings in inserting an open-source library for differential privacy into a workflow for leveraging a geodatabase for geocoding and analyzing geospatial data stored as part of the Observational Medical Outcomes Partnership (OMOP) common data model. We pilot this process using an open big data repository of addresses.
Big data often has emergent structure that exists at multiple levels of abstraction, which are useful for characterizing complex interactions and dynamics of the observations. Here, we consider multiple levels of abstraction via a multiresolution geometry of data points at different granularities. To construct this geometry we define a time-inhomogemeous diffusion process that effectively condenses data points together to uncover nested groupings at larger and larger granularities. This inhomogeneous process creates a deep cascade of intrinsic low pass filters on the data affinity graph that are applied in sequence to gradually eliminate local variability while adjusting the learned data geometry to increasingly coarser resolutions. We provide visualizations to exhibit our method as a "continuously-hierarchical" clustering with directions of eliminated variation highlighted at each step. The utility of our algorithm is demonstrated via neuronal data condensation, where the constructed multiresolution data geometry uncovers the organization, grouping, and connectivity between neurons.
