Abstract
The Federated Learning (FL) approach has advanced traditional machine learning, enabling the handling of heterogeneous data, such as Non-Identically Distributed Data (Non-IID). This approach provides distributed data access across multiple research centers. The traditional approach only accesses data in a centralized way and misses data due to challenges in acquiring data from other healthcare research centers. Although FL can help handle real-world scenarios, it also poses security issues, such as adversarial attacks, backdoors, and others. In light of these security issues, it is challenging to detect and mitigate threats or attacks that compromise healthcare centers, resulting in reputational or financial losses. The victims even imagine that their data serves as a vector for an attacker to invalidate a system. Based on these concerns, Federated Learning environments pose distributed systems, increasing the risk of security breaches compared to traditional models in healthcare analysis. This project seeks to analyze security threats in Non-IID Federated Learning and mitigate them by leveraging knowledge from related areas, including defensive strategies grounded in the statistics and mathematical foundations of deep learning, to improve healthcare systems' security. We also proposed implementing, adapting, and integrating federated systems in the clinical environment to predict the risk of brain diseases, such as Alzheimer's, given their complex analysis and the need for additional data to develop more robust models.
Grant number: 26/03705-5
Start date: April 01, 2026
End date: March 31, 2028
Support Opportunities: Scholarships in Brazil - Post-Doctoral
Principal Investigator: Caetano Traina Junior
Grantee: Erikson Júlio de Aguiar
Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
Associated scholarship(s): 24/13328-9 - Intelligent management of multimodal health data for decision-making in big data scenarios: IHealth-MD, AP.TEM
Abstract
The proposal focuses on the challenges of managing and integrating multimodal health data for knowledge discovery, aiming at extracting meaningful information from large volumes of complex and diverse health data by leveraging Big Data and Artificial Intelligence (AI) for the Health Sciences. Almost every human activity now generate and require storing and processing vast quantities of diverse and complex data, from scientific, academic, and business to leisure activities. Health-related activities are no different, as they produce big data and can benefit from technological advancements to enhance decision-making processes through the information extracted from this data.In a clinical environment, electronic health records (EHRs) are the foundation for developing information extraction strategies. In this proposal, we intend to develop and integrate novel and scalable algorithms powered by data engineering (DE) and AI methods. These algorithms will leverage large amounts of EHR and clinical data repositories to gather valuable and significant information for decision-making. Data communication and security aspects will firmly be considered, ensuring the integrity and privacy of the data while enabling efficient knowledge discovery.The size and complexity of EHR databases, which include structured and unstructured text, signals, images, lab results, and genomic data, present significant challenges for processing. These challenges encompass the application of analysis techniques and the development of practical tools and subsequent applications. However, these databases also present numerous opportunities to develop new algorithms and methods capable of displaying smart and relevant information related to individual patients or groups of patients. This can transform EHRs into more effective platforms, enhancing support for healthcare professionals, optimizing medical applications, and informing strategic government decisions in line with the demands and benefits of big data. In this project, we aim to address the challenges of managing and integrating not only the information but also the knowledge from multiple modalities of health data, focusing initially on lung-related diseases, primarily lung cancer, as well as cardiovascular diseases. We will develop methods and algorithms that will ultimately be materialized in a modular platform, which will be made available to the healthcare community.
Grant number: 24/13328-9
Start date: August 01, 2025
End date: July 31, 2030
Support Opportunities: Research Projects - Thematic Grants
Principal Investigator: Agma Juci Machado Traina
Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
City of the host institution: São Carlos
Abstract
With the advent of big data, data is being produced and generated on a large scale, used by Machine Learning (ML) models to generate new knowledge. Several areas have benefited from big data and ML, one of them being healthcare, which can employ complex data such as images to assist medical experts in decision making. While these concepts are valuable for healthcare, they can lead to issues regarding patient privacy and security. Information leaks in healthcare systems frequently. For example, in 2020, data of 200,000 patients from public health systems in Brazil was exposed. The ML models employed in healthcare are susceptible to attacks that poison the input data, the model itself and cause problems in the test data. In addition, they can present both known and unknown backdoors. The area of study that proposes defense and attack strategies against ML models is adversarial machine learning, which aims to reduce the model's reliability and cause the model to misclassify the data. Therefore, this project aims to devise a framework consisting of defense, vulnerability exploitation, and attack models to understand and combat security and privacy violations in pattern recognition models in medical images. Medical images are used as input to ML models to recognize patterns and support medical decision-making. However, these images, such as the models that classify them, can suffer attacks to invalidate their robustness or compromise the patient's privacy. In this project, we hope to: (I) develop defensive algorithms against adversarial examples; (II) devise methods to preserve patient privacy; (III) exploit new vulnerabilities and backdoors that ML models may present; (IV) propose attack strategies and their respective defenses, to communicate to other researchers the possible paths an attacker may follow.
Grant number: 21/08982-3
Start date: March 01, 2022
End date: February 28, 2026
Support Opportunities: Scholarships in Brazil - Doctorate
Principal Investigator: Agma Juci Machado Traina
Grantee: Erikson Júlio de Aguiar
Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
Associated scholarship(s): 23/14759-0 - Privacy-preserving and backdoors defending: towards federated learning in medical settings, BE.EP.DR
Abstract
The Blockchain technology has been gaining visibility in the market and the academic environment, due to its characteristics that allow to boost the security and reliability and robustness of distributed systems. Several areas are being considered with advances in research using this technology, such as: (i) finance, (ii) environment sensing, (iii) data analysis, and (iv) health. It is possible to emphasize that the characteristics of data immutability, privacy, transparency, decentralization and distributed ledger all contribute significantly to Blockchain being applicable by solutions in several areas. In this sense, it is highlighted that one of the contexts that can benefit from this approach is the sharing of medical records for research purposes. This type of information contains confidential patient data, which makes this sharing process more complex because of the risks of privacy breach. From this perspective, this research project has as main aim to propose a Blockchain based protocol of multiple levels of access, that is safe and manageable for the sharing of medical images. In addition, such a proposal should provide the institutions (medical and research) of your network the ability to track and audit the shared data. In brief, this project proposal aims to contribute with advances in multidisciplinary areas by proposing and evaluating a protocol that is able to facilitate the intersection of information between the medical and computer areas.
Grant number: 18/18187-3
Start date: April 01, 2019
End date: October 31, 2020
Support Opportunities: Scholarships in Brazil - Master
Principal Investigator: Jó Ueyama
Grantee: Erikson Júlio de Aguiar
Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
Associated scholarship(s): 19/19913-2 - Preserving-privacy on blockchain-based model to healthcare: a case study to multimedia data, BE.EP.MS
Prof. Agma Traina, University of São Paulo (ICMC USP), Brazil
Prof. Caetano Traina Jr., University of São Paulo (ICMC USP), Brazil
Prof. Sumi Helal, University of Bologna (UNIBO), Italy
Dr. Alyson Dos Santos, Federal Institute of Manaus (IFAM), Brazil
Dr. José Rodrigues Neto, Federal University of Piau (UFPI), Brazil
Dr. Jean Ponciano, University of São Paulo (ICMC USP), Brazil