Embedded Files
The BDS Lab is at the forefront of innovative research and development (R&D) in healthcare modeling and technology. The lab's current projects span a wide range of critical areas, including mental health interventions, healthcare data interoperability, medical image analysis, and public health software development. These initiatives utilize cutting-edge technologies such as large language models (LLMs), machine/deep learning, blockchain, and FHIR interoperability to address pressing challenges in healthcare delivery and patient outcomes.
The lab's work in mental health focuses on developing advanced AI-driven systems for trauma response classification, enhancing medical language accessibility, and optimizing resource allocation. In healthcare data management, the team is working on improving interoperability and security through AI and blockchain technologies. The medical image analysis projects, conducted in collaboration with Henry Ford Health, aim to enhance early detection of conditions like low bone mass and lung cancer using deep learning algorithms.
On the public health front, Dr. Hembroff's team is developing innovative software solutions, including a FHIR-enabled health information exchange system to create truly patient longitudinal records and producing analysis and predictive analytics via AI/ML models, a behavioral health network patient registry providing mental health and substance use disorder support by remote psychiatric consultation, providing hopsitals and clinics with an architecture to provide secure mental health screenings for pediatric patients using iPads with assessment and education resources for primary care providers, and the crowd-sourced citizen science Tick-Talk geographical tick disease monitoring platform to assist with public disease surveillance. These projects demonstrate the lab's commitment to translating research into practical applications that can significantly impact public health and patient care. Please see below for more information regarding this work.
Research Projects
AI Aimed to Improve Human Mental Health and Wellness
Automated Mental Health Interventions with LLMs: A System for Trauma Response Classification and Customized Self-Regulation Interaction
This research advances our current work in large language models (LLMs) to define a customized intervention optimization model to include key self-regulation responses directly based on user’s input through trauma (fight, flight, freeze, and fawn) classification.
Optimization of the model incorporates teams of licensed adult/child psychiatrists and psychologists to validate the model’s output and policy based on RLHF.
Medical Language Accessibility: Extending Transformer Models for Grade-Level Comprehension
Research advances our work on our December 2024 paper titled “Advancing Health Literacy through Generative AI: The Utilization of Open-source LLMs for Text Simplification and Readability” to further improve enhanced simplifying of words from the Living Word Vocabulary + API lookup and simplification of new words and their respective meanings by grade reading level to enhance user comprehension.
Enhancing Mental Health Intervention Efficacy through Multi-Source Data Integration and Advanced Machine Learning Models
This research focuses on developing LLM models to evaluate the effectiveness of mental health interventions by analyzing integrated data from mobile apps, wearable devices, and self-reported assessments.
Mental Health Resource Allocation Optimization: A Machine Learning Approach to Cost-Benefit Analysis and Dynamic Intervention Adjustment
Development of deep learning models to assess resource utilization, evaluate cost-effectiveness, and dynamically adjust interventions based on real-time user feedback
Data-Driven Policy Modeling for Cost-Effective Mental Health Care: Predictive Analytics and Personalized Intervention Strategies to Ensure Valuable Policy Development
Research encompasses the objective of developing ML/DL models to generate cost-effective data, reports, and policies for optimizing mental health care across various stakeholders, including insurance companies, healthcare providers, and government agencies.
Healthcare Data Interoperability and Security
Interoperability and Security in Healthcare Data Systems: AI-Driven Auto-Mapping and Blockchain for IoMT Integration
Research reflects the key themes and challenges outlined in the problem statements, such as data interoperability, security, and the integration of Internet of Medical Things (IoMT) devices.
Highlights proposed solutions involving AI and blockchain technologies.
Development and Efficacy Assessment of a Blockchain-Enabled Data Exchange using Cross-Platform Personal Health Record Architecture
This research project aims to contribute to the field of healthcare informatics by exploring the intersection of blockchain technology, mobile application development, and electronic health records.
The findings will provide insights into the feasibility and effectiveness of blockchain-integrated patient health record systems, potentially influencing future developments in healthcare data management and patient empowerment.
Medical Image Analysis
Improved Fracture Risk Predictions through the Detection of Low Bone Mass based on Opportunistic Screenings of Existing Knee Radiographs for Density Modeling
In collaboration with Henry Ford Hospital, the goal of this research project is to develop a deep learning model capable of automated early detection of low bone mass on existing radiographic images of the knee. Our hope is through opportunistic radiographic screenings, such as of the knee, we can develop a deep learning algorithm which provides accurate information to identify if a patient has low bone mass and should be recommended for treatment options. Therefore, our hypothesis is that the use of this methodology will result in reduced imaging, leading to improved patient safety through less exposure to radiation, the reduction of healthcare imaging costs, and a reduction in the time to diagnosis and appropriate treatment for low bone mass.
This research project with Henry Ford Hospital aims to fill a critical gap in lung cancer screening by harnessing radiomic features to enhance the predictive power of the Lung-RADS classification. By identifying low-risk nodules with higher malignancy potential, we can enable earlier and more tailored interventions, potentially improving patient survival rates and reducing the burden of lung cancer. Additionally earlier detection of malignant transformation may allow for intervention with targeted radiation therapies which might avoid more invasive surgical options. This project represents a novel approach to lung cancer screening, integrating radiomic analysis with existing classifications. By leveraging cutting-edge machine learning techniques to analyze imaging data, this research could lead to a paradigm shift in how we assess and manage lung nodule risk, making lung cancer screening more precise and personalized.
Classification of Benign Appearing Lung Nodules using Artificial Intelligence and Radiomics
This collaborative research project with Henry Ford Hospital aims to fill a critical gap in lung cancer screening by harnessing radiomic features to enhance the predictive power of the Lung-RADS classification. By identifying low-risk nodules with higher malignancy potential, we can enable earlier and more tailored interventions, potentially improving patient survival rates and reducing the burden of lung cancer. Additionally earlier detection of malignant transformation may allow for intervention with targeted radiation therapies which might avoid more invasive surgical options. This project represents a novel approach to lung cancer screening, integrating radiomic analysis with existing classifications. By leveraging cutting-edge machine learning techniques to analyze imaging data, this research could lead to a paradigm shift in how we assess and manage lung nodule risk, making lung cancer screening more precise and personalized.
This Work Continues our Existing Research:
Lung Nodule Classification from Radiology Report Using Bidirectional Encoder Representations from Transformers
Public Health Software Development Projects
Pediatric Mental Health Decision Support for Primary Care Physicians
In collaboration with Michigan State University, Superior Health Foundation, and rural hopsitals and clinics in the Upper Peninsula of Michigan, Dr. Hembroff’s team is collaborating with psychiatrists, primary care physicians (PCPs), nurses, psychologists, and social workers to provide vaulable pediatric mental health education and information to assist PCPs and other hospital/clinic staff with mental health clinical decisions and timely resources for patients. Specifically, the BDS Lab will: 1) develop a full stack secure web-based repository to permit secure and timely information to clinic personnel for their pediatric patient population; and 2) develop train, and test large language model (LLM) agents for clinicians to be used in mining data to provide customized clinical decision answers to questions in real-time.
Tick-Talk: Geographical Tick Disease Monitoring and Geolocation Mapping
In this public health monitoring collaborative project, Dr. Hembroff has been leading the development of public health software at Michigan Technological University by utilizing crowdsourced data to provide real-time mapping of tick disease monitoring. This initiative, supported by the MI-SAPPHIRE grant, involves engaging the public in collecting ticks from Michigan's Upper Peninsula and nearby regions. The collected ticks are analyzed to generate up-to-date geolocation data, which is then visualized on an interactive dashboard. This approach not only focuses on data collection but also aims to educate the public on tick identification, associated risks, preventive measures, and treatment options. The integration of citizen science, technology, and education in this project creates a robust public health safety net, ensuring communities are well-informed and proactive in addressing tick-borne illnesses.
FHIR-Enabled Health Information Exchange to Improve Patient and Public Health Disease Surveillance Outcomes
In collaboration with a nonprofit hospital network, Dr. Hembroff’s team is working on developing a full stack secure web-based patient registry for disease surveillance. Through FHIR interoperability and data-driven insights, the project enables real-time decision-making and policy development with rich visualizations aimed to improve public health outcomes and safety.
Development of a Customized Behavior Health Network (BHN) through an Integrated Psychiatric Collaborative Care Model (CoCM) Patient Registry
This collaborative project focuses on creating a secure patient registry for use in integrated healthcare settings. The registry records valuable information about patients’ behavioral health, so care management personnel can make informed decisions. By translating data into critical visualizations and analytics, the project enhances healthcare and outcomes for patients.
Report abuse