2024-2027 (FAPESP 2014/50889-7) – Natural Language Processing in Cardiovascular Medicine
Collecting vast amounts of health data can transform healthcare by enabling personalized care, reducing inefficiencies, and promoting prevention-focused systems. However, challenges arise due to the complexity of medical conditions, diverse data sources, and evolving methodologies, creating gaps between current models and real-world clinical needs. To address this, we propose a comprehensive framework for collecting and processing cardiovascular health data using artificial intelligence. Our objectives include: (1) Developing a robust environment for processing diverse cardiovascular data with advanced machine learning, bridging the gap between raw data and meaningful clinical insights. (2) Creating an NLP-based application to extract clinical entities and their temporal relationships from medical narratives, simplifying patient history analysis and improving diagnostic accuracy. (3) Standardizing data using medical terminologies like SNOMED-CT and ICD-10, ensuring interoperability with broader healthcare systems.
These innovations will overcome key barriers to integrating vast health data, paving the way for a more efficient, accessible, and predictive healthcare system.
2022- 2026 (FAPESP 2014/50889-7) – Automatic Quantification of Prognostic Biomarkers in Cardiovascular Disease Using Deep Learning in Chest Computed Tomography Images
The assessment of opportunistic findings in computed tomography (CT) scans has gained relevance in refining cardiovascular disease (CVD) risk, especially with the use of algorithms that automatically identify and quantify prognostic biomarkers. In the United States, it is estimated that before the COVID-19 pandemic, around 60,000 dedicated coronary artery calcium (CAC) score scans and 12.7 million chest CT scans were performed annually, a number that likely increased post-pandemic due to the widespread use of this exam. Recent advancements in CT technology now allow for imaging with radiation doses comparable to chest X-rays, making CT a viable alternative for screening while reducing patient risks. This project aims: (a) the development of deep learning (DL) algorithms for the automatic quantification of CAC in chest CT scans, including those performed during the pandemic and images acquired using a new X-ray filtration system for radiation dose reduction, in a cohort of patients referred for lung cancer screening; and (b) the development of new algorithms for the identification and automatic quantification of other prognostic biomarkers in chest CT, such as aortic calcification, valvular calcification, and markers of sarcopenia. These advancements can enhance cardiovascular risk stratification and contribute to more precise and early diagnosis.
2022 – 2025 (FAPESP 2014/50889-7) – System for Continuous Monitoring of Cardiovascular Biomarkers through Wearable Devices
Continuous monitoring of vital signs is crucial for preventing and managing cardiovascular diseases. However, global adoption relies on wearable medical devices that are accurate, affordable, and easy to use. This project proposes an innovative ecosystem integrating advanced algorithms with a monitoring platform for low-cost wearable devices, enabling continuous, non-invasive tracking of key cardiovascular biomarkers from photoplethysmography (PPG) signals. The monitored parameters include blood pressure, heart rate, cardiac output, oxygen saturation, blood glucose, sleep patterns, and physical activity. While some biomarkers are well-established, others face challenges due to physiological complexity and current invasive methods. To address these, the project will develop machine learning algorithms for precise estimation, with secure cloud storage and real-time analysis. A multi-platform solution will facilitate integration with other systems, alongside an Edge device for hospital use. Validation will occur with real-world Brazilian data, with results targeted for high-impact publications, patents, and industry transfer—maximizing social, clinical, and economic benefits.
2020 – 2025 (FAPESP 2020/11258-2) – Queries for similarity and interoperability in medical databases
Health institutions store patient data (Electronic Patient Records – REP) in their own repositories, using the structural particularities defined in each institution to carry out their activities. However, a given individual’s data is often spread across many repositories in different institutions and formats leading to the frequent need to share information between institutions. Given the differences in how and what data are recorded, the need for standards for their interoperability arises, and one of the most recent and promising technologies for sharing medical data is based on the Common Model of Observational Data in Medicine (OMOP-CDM) proposed by OHDSI initiative. The data stored in REP is fundamentally multimodal (images, videos, time series, etc.), acquired from different sources and with varied structures or formats that must be properly managed. This project aims to make fonts and formats compatible, developing techniques to integrate previous GBdI research results obtained with data from medical exams based on images in OMOP. Although there is still no standardized support for similarity comparisons or for complex data types such as images in OMOP, this research intends to develop it in the MIVisDB Project and make it available to the community in the area.
2017 – 2025 (FAPESP 2016/17078-0) – Mining, indexing and visualization of Big Data in the context of clinical decision support systems (MIVisBD)
Today, almost all human activities generate and/or demand to store and process ever-increasing sets of data, which are often diverse and complex. This scenario of vertiginous growth, which occurs both in the scientific, academic, business and even leisure activities, demands new efficient methods for organization and access. Such a scenario is being called as the “age of big data”. Health-related activities and systems are at the center of this scenario, as they produce large amounts of diverse and complex data. It is important that we advance technologically, in order to benefit from this volume of data to expand knowledge of the areas, in order, for example, to support the decision-making process. This decision support in complex systems is increasingly guided by the information extracted and what is learned from these large volumes of data. In a clinical environment, the Electronic Patient Records (ERP) constitute a propitious platform for the development of strategies for extracting information from patients, their profiles and even from groups that have the same casuistry. In this project, we intend to integrate innovative database supports, image processing and visual data analysis methods based on REPs and clinical data repositories to gather valuable and meaningful information for decision making that support diagnosis and treatment. of patients. The size and complexity of REP databases pose major processing challenges, both in terms of the development and application of analysis and knowledge extraction techniques, and in terms of supporting the development of practical tools for clinical use. However, they also incorporate a myriad of opportunities to create algorithms and methods capable of displaying relevant information related to a particular patient or groups of patients, which would usually be hidden by the large volume of data. Furthermore, an efficient manipulation of this data has high potential to make REPs a more effective platform to support healthcare professionals, dealing with fast-demand medical applications as well as strategic government decisions in healthcare. In this project we will develop methods and algorithms that will be materialized in a modular platform to be made available to the community in the area, supporting the daily decision-making in health systems.
2014 – 2025 (FAPESP 2014/50889-7) – National Institute of Science & Technology in Medicine Assisted by Scientific Computing (INCT-MACC)
The main objective of this proposal is to transfer innovative medical applications to the Health System, oriented towards diagnosis, treatment, surgical planning, training and computer-aided decision support. The National Institute of Science and Technology in Medicine Assisted by Scientific Computing (INCT-MACC) has been allowing, since 2008, to successfully integrate a network of knowledge in modern communication techniques and multimedia transmission, developing and managing computing environments in cloud and high performance for computer modeling and simulation of physiological systems that make up the human body, involving different scales at the molecular, cellular and systemic levels, promoting the development of medical image processing, scientific visualization and virtual reality in the development of innovative medical applications.