The research conducted by the lab spans diverse domains within healthcare, with a primary focus on integrating advanced technologies to enhance clinical decision-making. The work involves the development of novel frameworks and systems for real-time event stream processing, particularly in critical care settings. Notably, the research explores the integration of clinical and physiological data through streaming multi-modal infrastructure, enabling the creation of complex event processes for knowledge-driven clinical decision support systems.

In particular, we have focused on advancing healthcare research through innovative methods for the analysis of physiological signals, computer vision, adaptive machine learning, and unsupervised learning techniques. Our expertise lies in real-time event stream processing, integrating irregular time series e.g. clinical and physiological data to develop predictive models for early detection of conditions like sepsis and acute respiratory failure. We're at the forefront of algorithmic development, creating complex models to analyze continuous physiological waveform data, including novel methods for the synchronization of multiple signals to identify unique biological patterns. 

Our work also involves utilizing novel techniques, including transformer networks, generative AI, and deep neural networks for contributing novel foundational models to the domain of critical care medicine. We contributed to the open-source community with projects like improving learning under extreme class imbalances using CTGANs, and infrastructure like PhysOnline, a machine-learning pipeline for real-time analysis of streaming physiological waveform data. In addition to novel common data models for incorporating various clinical data elements into machine-readable and generalizable assets. Overall, our lab's commitment to physiological signal processing and machine learning is driving advancements in understanding and predicting highly time-sensitive clinical conditions.

The development of novel methods to address open problems related to healthcare machine learning is a significant aspect of the research, with a keen interest in the development of novel methods to aid in predicting and managing critical conditions. The Kamaleswaran Lab has developed and validated various models for predicting outcomes such as sepsis, acute respiratory failure, and acute respiratory distress syndrome (ARDS). The utilization of artificial intelligence extends to diverse areas, including identifying prodromal Parkinson's disease, predicting fever in critically ill children, and assessing complications after liver transplantation.

Furthermore, the Kamaleswaran Lab has contributed to advancements in data visualization techniques for physiological event streams and has actively engaged in collaborative initiatives, such as the National COVID Cohort Collaborative (N3C). The research aligns with a broader vision of leveraging technology to accelerate insights from complex data, with a particular emphasis on surgical sciences, anesthesiology, and critical care informatics. This comprehensive approach aims to revolutionize clinical support systems, fostering advancements in precision medicine and healthcare delivery.

Improving Sepsis Prediction Model Generalization With Optimal Transport

Sepsis is a deadly condition affecting many patients in the hospital. To circumvent the curse from noisy and unbalanced samples, we develop a novel two-step approach for sepsis prediction: given feature-label points from the source domain and feature points from the target domain, to obtain a sepsis predictor that has satisfactory performance at the target domain. The proposed algorithm first learns how to transform sample points from the source domain to the target domain, and then applies the distributionally robust optimization (DRO) technique with the Sinkhorn distance and asymmetric cost function to reliably obtain a classifier with satisfactory out-of-sample performance. 

Integrating Structured and Unstructured Data for Timely Prediction of Bloodstream Infection among Children

Hospitalized children with central venous lines (CVLs) are at higher risk of hospital-acquired infections. Information in electronic health records (EHRs) can be employed in training deep learning models to predict the onset of these infections. Integrating clinical notes with structured EHR data improved the prediction of serious bloodstream infections among pediatric patients with CVLs.

Machine Learning Approaches to Identify Discriminative Signatures of Volatile Organic Compounds (VOCs) from Bacteria and Fungi Using SPME-DART-MS

Point-of-care screening tools are essential to expedite patient care and decrease reliance on slow diagnostic tools (e.g., microbial cultures) to identify pathogens and their associated antibiotic resistance. This study presents a new approach for the identification of pathogens from VOC signatures collected using SPME and ambient ionization MS by training classifiers on just a few samples of data. This ambient plasma ionization and ML approach is robust, rapid, precise, and can potentially be used as a non-invasive clinical diagnostic tool for point-of-care applications.

OnAI-Comp: An Online AI Experts Competing Framework for Early Sepsis Detection

Sepsis has always been a main public concern due to its high mortality, morbidity, and financial cost. There are many existing works of early sepsis prediction using different machine learning models to mitigate the outcomes brought by sepsis. In the practical scenario, the dataset grows dynamically as new patients visit the hospital. Most existing models, being '`offline'' models and having used retrospective observational data, cannot be updated and improved using the new data. Incorporating the new data to improve the offline models requires retraining the model, which is very computationally expensive. To solve the challenge mentioned above, we propose an Online Artificial Intelligence Experts Competing Framework (OnAI-Comp) for early sepsis detection using an online learning algorithm called Multi-armed Bandit. 

Cluster analysis and profiling of airway fluid metabolites in pediatric acute hypoxemic respiratory failure

Hierarchal clustering of amino acid metabolites may identify a metabolic signature in children with pediatric acute hypoxemic respiratory failure. We used hierarchal clustering and partial least squares-discriminant analysis to profile the tracheal aspirate airway fluid using quantitative LC–MS/MS to explore clusters of metabolites that correlated with acute hypoxemia severity and ventilator-free days. Three clusters of children that differed by severity of hypoxemia and ventilator-free days were identified. Quantitative pathway enrichment analysis showed that cysteine and methionine metabolism, selenocompound metabolism, glycine, serine and threonine metabolism, arginine biosynthesis, and valine, leucine, and isoleucine biosynthesis were the top five enriched, impactful pathways. 

Uncertainty-Aware Convolutional Neural Network for Identifying Bilateral Opacities on Chest X-rays: A Tool to Aid Diagnosis of Acute Respiratory Distress Syndrome

Acute Respiratory Distress Syndrome (ARDS) is a severe lung injury with high mortality, primarily characterized by bilateral pulmonary opacities on chest radiographs and hypoxemia. In this work, we trained a convolutional neural network (CNN) model that can reliably identify bilateral opacities on routine chest X-ray images of critically ill patients. We propose this model as a tool to generate predictive alerts for possible ARDS cases, enabling early diagnosis. We used a novel training technique that enables the CNN to explicitly predict the ‘equivocal’ class using an uncertainty-aware label smoothing loss.