MB4: Machine Learning Advances in Healthcare Aystems

Room: Old Main Academic Center 3090

Webex Link

OrganizerHaifeng Wang, Mississippi State University

Lir-Wan Fan, University of Mississippi Medical Center

Time: 10:25 am - 10:50 am (CST)

Title: Neonatal Inflammation Induces Attention-deficit/hyperactivity Disorder (ADHD)-like Behaviors and Effects on REM Sleep and Homeostatic Responses to Sleep Disturbances in Adolescent Rats

Abstract: Inflammation may play an important role in the association between sleep disturbances and neurodevelopmental disorders such as attention-deficit/hyperactivity disorder (ADHD) development. The objective of the current study was to examine whether perinatal systemic LPS exposure results in chronic inflammation, ADHD-like behaviors, and related sleep disturbances in adolescents. Intraperitoneal injections of LPS (2mg/kg) or saline was performed on postnatal day 5 (P5) Sprague-Dawley male rat pups, and surgery/sleep recording electrode implantation was conducted on P39. A baseline sleep recording, sleep disturbances, and recovery sleep was recorded on P46, P47 and P48 for 24 hours, respectively. The brain inflammation and neuronal damage were examined at P49. Our results showed that neonatal LPS treatment induced ADHD-like behaviors, including hyperactivity and inattention on P35. Neonatal LPS treatment interfered with REM sleep and sleep homeostatic responses (recovery sleep) to sleep disturbances in adolescent rats (P49). Neonatal LPS treatment also induced chronic microglia activation (Iba1+) and brain damage including the loss of TH+ neurons in the locus coeruleus of the P49 rat brain. These data suggest that neuroinflammation initiated at early development persists in adolescent age, which may contribute to neurodevelopmental disorders and sleep disturbances by disrupting sleep homeostatic responses, namely, recovery sleep.

Joint work with Silu Lu, Joseph C Crosby, Jonathan W Lee, James P Shaffery, Lu-Tai Tien, Michelle A Tucci, Haifeng Wang, Zhiqian Chen, Minal Patel, and Norma B Ojeda.

Chun-An Chou, Northeastern University

Time: 10:00 am - 10:25 am (CST)

Title: Data-Driven Modeling and Analysis for Understanding Complex Dynamics of Physiological Networks

Abstract: The human brain as an integrated complex system is consisted of neural units that continuously interact with each other to coordinate brain functions. The underlying interactions of these units forming a complex network are correlated with distinct brain states or functions. However, it is challenging to characterize the interactions directly as these units behave in a non-linear and non-stationary manner.In this talk, we will introduce novel data-driven computational methods to model and analyze the brain network and its dynamics using multi-variate EEG signals in computational neuroscience applications. More specifically, we present an embedding optimization method to reconstruct and project the brain system of multi-variate EEG signals in a low dimensional space. With the proposed quantification measurements, we are able to present the complex brain network as a whole. We apply this idea to epileptic seizure detection and schizophrenia pattern classification. The computational results demonstrate very competitive detection and classification performances comparing to state-of-art signal processing and machine learning methods. We will also present the extended work to discover the joint dynamics of human body network of multiple physiological systems.

Norma B Ojeda, University of Mississippi Medical Center

Time: 10:50 am - 11:15 am (CST)

Title: Machine Learning to Improve Management of Developmental Delay Associated with Prematurity in a Hospital-based Pediatric Population in Mississippi

Abstract: Developmental Delay (DD) is a long-term complication associated with prematurity (PT) with wide variability in prevalence rates worldwide. This variability is associated with regional differences affecting diagnosis and management. Optimization of healthcare services and utilization of resources are critical to improve management and outcome of these patients. Machine learning approaches can help identify patterns to improve management and effectiveness of health services for these patients. We investigated the prevalence and relative risk of developing developmental delay among premature patients, and the utilization of resources in these patients at the University of Mississippi Medical Center. We utilized the Patient Cohort Explorer (PCE), a public de-identified database from the University of Mississippi Medical (UMMC) Center compiled by the UMMC Research Data Warehouse (RDW). We included information from patients receiving healthcare service at the Medical Center from 01-01-2013 to 012-31-2021. We used SPSS23 software and MedCalc.net website for statistical analysis. We identified Adult and Pediatric patients total: 1,170,491 with total number of encounters: 40,297,667. Pediatric patients total: 404,019 (34%) encounters total: 11, 050, 479 (27%). Premature diagnosis total: 11,751 (1%) with total encounters: 900,471 (22%). Premature + Developmental Delay diagnosis total: 2,262 (0,23%) with total encounters; 517,391 (2%). Prevalence rate for Developmental Delay among Premature patients: 19%. Relative Risk (RR) 4.8 fold increased for Developmental Delay in Premature infants compared to full-term patients (RR; 4.8, 95%CI 4.6874 to 5.0412, P<0.0001). Utilization of resources measured by the ratio encounters/ patients showed an increase of 39% for patients with diagnostic of Premature + Developmental Delay. (Premature: 77 vs. Premature + Developmental Delay: 194) Average Max Length of Stay (LOS) Premature patients: 50days, and for Premature + Developmental Delay patients: 70 days. Financial Guarantor status for Premature: Medicaid 47%, Self Paid: 41%, Insurance: 12%, and for Premature + Developmental Delay: Medicaid 50%, Self Paid: 43%, Medicare: 4%, Insurance: 3%. In conclusion, the prevalence and risk of Developmental Delay are higher among premature patients, and the utilization of resources is higher in patients with diagnosis of premature and developmental delay. Machine learning could help to identify patterns and associations to improve management and healthcare effectiveness for these patients.

Joint work with Irene Arguello and Lir-Wan Fan.

W Neil Duggar, University of Mississippi Medical Center

Time: 11:15 am - 11:40 am (CST)

Title: Machine Learning in Radiation Oncology

Abstract: Radiation oncology is an area of healthcare that is highly suitable for advancement with Machine Learning Techniques. We will review the conventional workflow for most patients receiving radiation therapy and how Machine Learning may impact each workflow step. Finally, we will review real examples of how machine learning is on cusp of advancement in this field.

Haifeng Wang, Mississippi State University

Time: 11:40 am - 12:05 pm (CST)

Title: Brain Network Reorganization for Attention Deficit Hyperactivity Disorder Diagnosis

Abstract: This research examines the dynamics of brain resting-state functional connectivity (rf-FC) using functional magnetic resonance imaging (fMRI) data for attention-deficit/hyperactivity disorder (ADHD). Machine learning is a high-potential approach for brain disorder diagnosis based on the constructed rf-FC brain network. The dynamics of brain connectivity directly impact the choice of algorithm design and model performance evaluation. In this study, we applied a sliding window to fMRI time-series data from ADHD-200 dataset for constructing a time-varying network, and we experimented with three window sizes (30, 40, and 60 seconds). Then, 10 different network metrics are calculated for each network and compared between the ADHD vs. Control groups. Our research investigates the changes insignificant brain network features over time. The statistical results show that the brain rf-FC shows strong dynamics, which makes the network structure change significantly over time. In addition, results show that the behavior of the significant brain network metric is inconsistent for ADHD vs. Control groups, which challenges the development of an effective machine learning model for ADHD diagnosis. However, experimental results also indicate that some network metrics are highly potential informative features, which show a strong relationship to diagnose ADHD from the Control group, such as betweenness centrality. We expect this study could provide insights into the future machine learning-based ADHD diagnosis.

Joint work with Harun Pirim and Miaolin Fan.