Current Events

Date: Tuesday, April 15, 2025 | 12:00PM-1:00PM, BMI Classroom 4004 or on Zoom
Speakers: Ehsan Hoque 
                 Professor of Computer Science
                 University of Rochester

Abstract: 
Parkinson’s disease (PD), a complex neurodegenerative disorder with a yearly economic burden projected to rise from $57 billion in 2017 to $80 billion by 2037 in the United States, demands innovative approaches to care. Digital health technologies—wearables, smartphones, and sensors—empower patients to monitor their health outside clinics, addressing the global shortage of clinicians.
Multimodal AI enhances this by integrating diverse data sources, such as speech and motor activity, to capture PD’s multifaceted symptoms. For instance, tremors may manifest in speech but not in physical movement. A smart multimodal fusion can enable early detection, personalized interventions, and equitable access to care. This talk will highlight cutting-edge multimodal AI research, showcasing its transformative potential for PD and beyond.

Date: Tuesday, March 18, 2025 | 12:00PM-1:00PM, BMI Classroom 4004 or on Zoom
Speakers: Shengpu Tang, PhD 
                  Assistant Professor of Computer Science
                  Emory University

Abstract: 
Decision making is at the core of healthcare: clinicians constantly make complex decisions that span diagnosis, treatment, care coordination, and resource allocation. Yet, human decisions are never perfect, leading to suboptimal patient care. My research aims to use AI to augment and improve decision-making in healthcare, following a synergistic approach that combines AI methods with practical, real-world implementation. In this talk, we will explore the two key themes of my research: (1) Application-Inspired AI Innovations, focused on novel AI methods grounded in practical healthcare problems; and (2) Path to Deployment and Impact, addressing AI integration into clinical workflows for real-world improvements. The talk will end with my future vision of human-AI teaming to enable better healthcare. 

Date: Thursday, March 13, 2025 | 2:00PM-3:00PM, BMI Classroom 4004 or on Zoom
Speakers: Ravi Parikh, M.D., MPP 
                 Associate Professor of Hematology and Medical Oncology
                 Emory University

Abstract: 
In this talk, I will describe 2 completed projects relating to using 1) digital twins – computational phenotypes of patients using real-world data – to understand real-world generalizability of clinical trials and improve future trial design, and 2) electronic health record-based language models to enhance trial matching and recruitment.

Date: Tuesday, March 4, 2025 | 12:00PM-1:00PM, BMI Classroom 4004 or on Zoom
Speakers: Zachary Danziger, Associate Professor, 
                 Department of Rehabilitation Medicine
                  W.H. Cluster Department of Biomedical Engineering, Emory University

Abstract: 
Simulating physiological systems is a powerful tool for generating hypotheses and for rapid prototyping of experimental treatments, but we often lack a full mathematical description of the system, which stimies our ability to simulate it. This talk explores a new approach to fill the gaps of missing ordinary differential equations (ODE) with small artificial neural networks (ANN). The goal is to train the entire hybrid ODE-ANN such that the embedded ANNs become approximations to the missing ODEs that can infer important but unmeasured physiological states of the system. The hybrid model (mostly) preserves interpretability and can be used to simulate the physiological system, thereby restoring our ability to study it computationally despite incomplete knowledge. We will explore the framework we are developing to build and train such hybrid ODE-ANN systems and deploy it for studying the lower urinary tract. 

Date: Tuesday, February 18, 2025 | 12:00PM-1:00PM, BMI Classroom 4004 or on Zoom
Speakers: Sang Won Bae, Ph.D., 
                 Assistant Professor, Department of Systems and Enterprises 
                 Schaefer School of Engineering and Science 
                 Stevens Institute of Technology

Abstract: 
This seminar examines the use of mobile sensors and machine learning to detect and intervene in acute substance intoxication in real-time, enabling just-in-time adaptive interventions. Dr. Bae will present her research on detecting binge drinking and marijuana intoxication through smartphones and wearable devices, emphasizing the role of explainable AI in providing transparency in decision-making. By leveraging data from smartphone sensors and wearables, her research explores how real-time predictions can empower individuals to make informed decisions, ultimately improving health outcomes and reducing substance-related harm. Dr. Bae will also discuss the technical and ethical challenges in implementing these technologies, including concerns around privacy, algorithmic transparency, and the need for personalized, adaptive systems that respect user autonomy. The talk will conclude with a forward-looking discussion on the future of digital health technologies, their potential to enhance public health, guide personalized interventions, and support clinical decision-making.

Date: Tuesday, January 21, 2025 | 12:00PM-1:00PM, BMI Classroom 4004 or on Zoom
Speakers: Benny Budiman Sc.D., Director of Research Data Analytics, Office of Research Informatics, Data, and AI, Emory University
Abstract: 
Data, Analytics, and AI-based solutions have potential to deliver enormous value to decision makers. Because of this astronomical potential, HBR reported that many organizations aspired to become data-driven so much so that they embarked on digital transformations by investing in big data and AI projects for enabling them to "compete on analytics" or to be "AI-first" in their business. Despite the promise of immense value, the percentage of failed data science projects has been alarmingly very high. Gartner Research estimated 80% of analytics insights failed to deliver value. A Gartner analyst also estimated that 85% of big data projects failed. A recent whitepaper from the Centre for Business Analytics at Melbourne Business School, while confirming the high failure rate of 80%, also reported the failure rate to be as high as 90% for analytically immature organizations but only as low as 40% for analytically mature organizations. In addition to the high failure rate, a high percentage of analytics projects failed to move beyond their pilot phase, 87% according to an estimate by VentureBeat AI. Even those that went into production faced low adoption rates. In a recent survey, DataIQ estimated only 23.9% of AI-driven solutions have been widely deployed into production.

Since 2004, the 7Cs framework for delivering values has been proven to identify and capture values. It has been used in the design, development, and deployment of data, analytics, and AI-based solutions in industrial settings where it captured and delivered over $1 billion in measurable outcomes from virtually every project in which it was used. The 7Cs functions like a checklist comprising seven elements all of which begins with the letter 'C' — hence, the name. The first is Context: understanding of objectives and goals. Next is Concept: requirement to be able to describe the flows of materials, resources, and information in the system / entity where challenges or problems need to be solved by using Physics, Chemistry, Biology, Business Dynamics, etc. to understand inputs, process, and outputs. After Concept, interdependency or interconnectedness — referred to in the framework as Connection — among elements of the flows need to be clearly articulated. The fourth C is Constraints that usually limit the flows or put restrictions on feasible actions. Because nothing is constant, the approach will evaluate impacts of Change involving variability and uncertainty by identifying the known unknown and the unknown known as well as anticipating the unknown unknown through scenarios so that risk can be assessed and planned to be mitigated accordingly. Communication, the sixth C, plays a crucial role in gathering and putting together a "picture" from the previous five Cs. It also plays an important role in creating data, analytics, and AI models in the seventh element of the framework which is Calculus.

Examples of how the 7Cs have been used in industrial settings will be presented next. In the end, ideas of how the 7Cs may be useful for Emory will be discussed.

Date: Tuesday, January 7, 2025 | 12:00PM-1:00PM, BMI Classroom 4004 or on Zoom
Speakers: Farzaneh Najafi, Assistant Professor, School of Biological Sciences, College of Sciences, Georgia Institute of Technology
Abstract: 
Precise motor timing underlies essential human behaviors such as speech, and motor coordination, with deficits profoundly impacting quality of life in conditions like ataxia. Two brain regions, cerebellum and parietal cortex, are both involved in motor behavior; however, their interaction for generating precisely timed movements remains poorly understood. My talk shows preliminary results from our lab addressing this question. We trained mice on visually and self-timed movements, while measuring and manipulating neural activity in the cerebellar and parietal cortex. We found temporal information in the parietal cortex, which requires intact cerebellar activity, hence suggesting potential causal interaction between these brain regions leading to temporally precise movements.

Date: Tuesday, November 19, 2024 | 12:00PM-1:00PM, BMI Classroom 4004 or on Zoom
Speakers: Chuan Huang, Associate Professor, Department of Radiology and Imaging Science, Director of PET-MIR Research, Emory University
Abstract: 
AI in PET/MR is an active field of research. While the current PubMed search identified limited PET/MR AI publications, it is clear that AI has the potential to be applied to the entire life cycle of PET/MR imaging. PET/MR researchers can learn from advancements both in the MR field and PET field, as well as the general computer vision field. In this talk, the speaker will discuss his lab’s experience in AI in PET/MR Imaging and the state of the art.

Date: Tuesday, November 5, 2024 | 12:00PM-1:00PM, BMI Classroom 4004 or on Zoom
Speakers: Sundaresh Ram, Assistant Professor, Department of Radiology and Imaging Science, Emory University
Abstract: 
Computational imaging has a rich history of using tools in the areas of signal processing, imaging physics, and machine learning to extract clinically relevant information from data acquired using medical imaging systems in order to support and improve the diagnosis, and therapy planning and follow-up of various diseases. The emergence of artificial intelligence has further expanded the footprint of this field. In this talk, I will present examples of research in our lab that will demonstrate how, by using tools from these areas, we are able to develop innovative solutions for addressing clinically impactful problems.

Date: Tuesday, October 1, 2024 | 12:00PM-1:00PM, BMI Classroom 4004 or on Zoom
Speakers: Puneet Sharma, Associate Professor, Pediatrics - Division of Neonatology, Emory University School of Medicine, Children's healthcare of Atlanta
Abstract: 
Artificial intelligence is transforming the landscape of neonatal care, offering powerful tools to analyze vast amounts of healthcare data, identify patterns, and support clinical decision-making. In this presentation, we explore the current and potential future applications of artificial intelligence in neonatal intensive care. We also discuss the potential hurdles to adoption of this technology and what we can do to ensure that it is deployed effective and safely in the care of vulnerable infants.

Date: Tuesday, September 17, 2024 | 12:00PM-1:00PM, BMI Classroom 4004 or on Zoom
Speakers: Azra Ismail, Associate Professor, Department of Biomedical Informatics, Emory University
Abstract: 
The prevailing conversation around the future of work often emphasizes digital advancements shaped by a select few, especially in hubs like Silicon Valley. The increasing global inequalities we see today remind us of the need to decentralize technology development and access. So how do we move towards crafting more democratic technological futures? In this talk, I will reflect on our experiences at MakerGhat in creating spaces for underserved youth to exercise their imagination through hands-on making. I invite us to think about how we can build more equitable futures, by addressing who gets to meaningfully participate in creating technology, and the role of education (and educators) in getting us there.

Date: May 21, 2024 | 12:00PM-1:00PM, BMI Classroom 4004 or on Zoom
Speakers: Giri Krishnan, Associate Director, Center for Artificial Intelligence in Science and Engineering, Georgia Institute of Technology
Abstract: 
Sleep is observed across species and seems essential for life. Yet, we are still trying to identify its function and how the complementary role of awake and sleep phase impacts biological and cognitive functions. In a series of work, we identify some of the neural mechanism that result in spontaneous activity during different sleep stages and how replay emerges and its impact on learning and memory. Further, we take inspiration from the neural mechanism of sleep to develop an algorithm for deep learning. The sleep-like state implemented for deep neural networks improved continual learning, and generalization under low data and noisy conditions.

Date: April 16, 2024 | 12:00PM-1:00PM, BMI Classroom 4004 or on Zoom
Speakers: Yaomin Xu, Assistant Professor, Biostatistics and Biomedical Informatics, Vanderbilt University Medical Center
Abstract: 
Multimorbidity, where multiple health conditions co-exist non-randomly within an individual, is a growing challenge for healthcare and society. Understanding multimorbidity patterns can lead to better prevention, treatments, and personalized care. The advent of electronic health record (EHR) systems provides a vast trove of data for studying real-world patient health dynamics. However, concerns about the primary design of EHRs for billing and administration raise questions about the consistency and reproducibility of EHR-based research. In this study, we used the International Classification of Diseases (ICD) codes to analyze disease comorbidity patterns and employed network modeling to examine multimorbidity across two major EHR systems. Our findings revealed highly correlated multimorbidity patterns across HER systems, with graph-theoretic analysis confirming the consistency of the multimorbidity networks at local (nodes and edges), global (network statistics) and meso (neighboring connection structures) scales. This result offered new insights for developing an efficient framework to analyze and compare complex structures within the multimorbidity network. Our case study demonstrated that identifying subgraphs within multimorbidity networks is an effective method for detecting disease condition clusters, and, supported by graph spectral characteristics of the multimorbidity networks, we developed a complete online network clustering algorithm as an efficient approach to identify those clusters. To facilitate access to these complex datasets and promote further discovery research and hypothesis generation, we have developed a suite of interactive visualization tools for complex online data analysis leveraging data from multiple EHR/Biobank data sources. These tools are open source, available to the public, and are designed to enable researchers to intuitively explore the complex disease relationships within the multimorbidity networks, thereby enhancing our collective understanding and fostering the development of novel precision medicine solutions in the context of multimorbidities.

Date: April 2, 2024 | 12:00PM-1:00PM, BMI Classroom 4004 or on Zoom
Speakers: Varun Mishra, Assistant Professor, Khoury College of Computer Sciences, Northeastern University
Abstract: 
The pervasiveness of sensor-rich mobile, wearable, and IoT devices has enabled researchers to passively sense various user traits and characteristics, which in turn have the potential to detect and predict different mental- and behavioral-health outcomes. Upon detecting or anticipating a negative outcome, the same devices can be used to deliver in-the-moment interventions and support to help users. One important factor that determines the effectiveness of digital health interventions is delivering them at the right time: (1) when a person needs support, i.e., at or before the onset of a negative outcome, or a psychological or contextual state that might lead to that outcome (state-of-vulnerability); and (2) when a person is able and willing to receive, process, and use the support provided (state-of-receptivity). In this talk, I will present my research about when to deliver interventions by exploring and detecting both vulnerability and receptivity. I will start by discussing my work that advances the current state-of-the-art by developing reproducible methods to accurately sense and detect various mental and behavioral-health outcomes like stress and opioid use disorder. Next, I will discuss my work regarding methods to explore and detect receptivity to interventions aimed at improving physical activity and how it can guide the design, implementation, and delivery of future mHealth interventions. Finally, I will discuss some of the current projects my lab is working on to build complete solutions that span the entire life-cycle of a digital health intervention (from sensing to intervention delivery) for various mental and behavioral health outcomes by answering "what,'' "when,'' and "how'' to deliver interventions.

Date: March 5, 2024 | 12:00PM-1:00PM, BMI Classroom 4004 or on Zoom
Speakers: Arian Ranjbar, Postdoctoral Fellow
Abstract: 
The last decade witnessed a rapid development of data driven methodologies. As the technology matures, there is a growing desire for its implementation into the healthcare sector, to meet the future demands of an aging population. Norway benefits from a public healthcare system where vast quantities of relatively standardized data are already available. However, the healthcare sector in general suffers from a large technical debt, while adhering to strict modes of operation. In this presentation I will talk about the current status and challenges of AI research and implementation at the largest hospital in the country – Akershus University Hospital. This includes data quality improvement, infrastructure and navigating in a system of IT silos, AI and healthcare ethics, and regulation such as the EU AI act; as well as our first research projects developing models on production data.

Date: Feb. 20, 2024 | 12:00PM-1:00PM, BMI Classroom 4004 or on Zoom
Speakers: Mously Dior Diaw, PhD Student
Abstract: 
Machine learning (ML) for healthcare holds many promises such as expediting repetitive tasks thus far sustained by human expertise and facilitating the discovery of novel biomarkers. In this talk, I will explore these two paradigms in the context of electrocardiography. First, I will present our human-in-the-loop framework that allows us to measure QT intervals, during drug safety trials, at low labeling cost. The framework consists of 3 key components: (1) deep learning (DL) based QT measurement with uncertainty quantification (2) expert review of a few DL-based measurements, mostly those with high model uncertainty and (3) recalibration of the unreviewed measurements based on the expert-validated data. Second, I will present our AFICIONADO project, which aims to leverage the accessibility of the ECG test to pre-screen high-risk patients for left ventricular dysfunction (LVD), which is traditionally detected with echocardiography. Such a pre- screening strategy would allow to only refer, for echo, patients displaying an abnormal profile of LVD-related echo parameters as estimated with a ML model built on paired ECG- echo data.

Date: Feb. 6, 2024 | 12:00PM-1:00PM, BMI Classroom 4004 or on Zoom
Speakers: Xuhai "Orson" Xu, Postdoctoral Associate
Abstract: 
As the intelligence of everyday smart devices continues to evolve, they can already capture basic health behaviors such as physical activities and heart rates. The vision of a complete human-AI pipeline for health -- from behavior modeling to intelligent intervention -- seems to be within easy reach. Why are we not there yet? Existing computational techniques are still far from being deployable, especially for longitudinal health behavior. In this talk, I will introduce our efforts on datasets, algorithms, and a benchmark platform, towards more robust and generalizable behavior modeling using everyday data. Based on these models, I will introduce interventions driven by both behavior science theory and AI that influence users' behavior and promote their health and well-being.

Date: Jan. 30, 2024 | 12:00PM-1:00PM, BMI Classroom 4004 or on Zoom
Speakers: Manasa Kalanadhabhatta, PhD candidate
Abstract: 
Nearly one in six U.S children aged 2-8 years has a diagnosed mental, emotional, or behavioral disorder, with actual rates of prevalence likely being even higher. Obtaining an accurate diagnosis is essential for facilitating treatment and interventions, but is often challenging due to a range of structural and phenomenological issues. Mobile and wearable devices offer an opportunity to fill this gap by enabling convenient, at-home behavioral screening. However, most screening technologies for young children rely on parent reports or behavioral observations, thus ignoring the neural and physiological underpinnings of behavioral disorders. In this talk, I will present my work on building neurophysiologically-grounded mental health screening tools for preschool aged children. First, I will discuss my research toward developing scalable, at-home assessment tools that use mobile and wearable devices to derive new insights of clinical value. Next, I will describe how neurophysiological measures including brain activity, cardiorespiratory signals, and electrodermal activity can be integrated with behavioral data to improve the reliability of assessment tools. I will conclude by outlining opportunities for future research in behavioral screening tools, including personalization and repeatability of assessments as well as the integration of neurophysiologically informed tools into clinical practice.

Date: Jan. 16, 2024 | 12:00PM-1:00PM, BMI Classroom 4004 or on Zoom
Speakers: Wei Jin, PhD
Abstract: 
Many learning tasks in Artificial Intelligence require dealing with graph data, ranging from biology and chemistry to finance and education. Graph neural networks (GNNs), as Deep Learning models have shown exceptional capabilities in learning from graph data. Despite their successes, GNNs often grapple with challenges stemming from data quality and size. This talk emphasizes a data-centric approach to enhance GNN performance. First, I will introduce a model-agnostic framework that enhances the quality of imperfect input graphs, thereby boosting prediction performance. Next, I will demonstrate methods to significantly reduce graph dataset sizes while retaining essential information fro model training. These data-centric strategies not only enhance data quality and efficiency but also complement existing models. Finally, I will introduce recent advances in graph generation and data-efficient learning. Join us to explore innovative approaches for overcoming data-related challenges in graph data mining.