A Retrospective Analysis of Magnetic Resonance Imaging Data for Breast Cancer Screening in the Open Consortium for Decentralized Medical Artificial Intelligence
试验速览
- 阶段
- 不适用
- 状态
- 进行中（未招募）
- 入组人数
- 25,000
- 试验地点
- 2
- 主要终点
- Diagnostic performance for breast cancer detection (Sensitivity and specificity)
研究概览
简要总结
ODELIA is a project that aims to improve breast cancer detection in magnetic resonance imaging by utilizing artificial intelligence and swarm learning (MRI). The project will create an open-source swarm learning software framework that will be used to train AI models for breast cancer detection. These models' performance will be compared to that of conventional AI models, and the results will be used to assess the effectiveness of swarm learning in improving the accuracy and robustness of AI models. The project will use retrospective, anonymized breast MRI datasets with manual ground truth labels for cancer presence. The study is not associated with any patient treatment or intervention. The project's goal is to provide evidence of the clinical benefits of swarm learning in the context of breast cancer screening, such as accelerated development, improved performance, and robust generalizability.
详细描述
Artificial Intelligence (AI) is set to revolutionize healthcare as its diagnostic performance approaches that of clinical experts. In particular, in cancer screening, AI could help patients to make better-informed decisions and reduce medical error. However, this requires large datasets whose collection faces severe practical, ethical and legal obstacles. These obstacles could potentially be overcome with swarm learning (SL) where partners jointly train AI models without sharing any data. Yet, access to SL technology is currently limited because no studies have implemented SL in a true multinational setup, no freely usable implementation of SL is available, researchers & healthcare providers have no experience with setting up SL networks and policymakers are currently unaware of the broader implications of SL.
ODELIA will aim to solve these issues: ODELIA will build an open-source software framework for SL, providing an assembly line for the streamlined development of AI solutions in a preclinical setting. To serve as a blueprint for future SL-based AI systems, ODELIA partners collaborate as a consortium to develop AI models for the detection of breast cancer in magnetic resonance imaging (MRI). The size of ODELIA's distributed database will be substantial and ODELIA's AI models could reach expert-level performance for breast cancer screening.
Thereby, ODELIA will could not just deliver a useful medical application, but provide evidence to summarize the clinical benefit of SL in terms of accelerated development, increased performance and robust generalizability.
To achieve this, ODELIA partners will collect retrospective, anonymized breast MRI datasets with manual ground truth labels for the presence of cancer, and will train AI models conventionelly and via SL. The performance of these technical approaches will be compared. The aim of the study is to test the methodology of Swarm Learning and the performance of AI algorithms developed within ODELIA on retrospective data. There will be no effect on treatment of patients as all evaluations will be done retrospectively. No patient treatment or any intervention is associated with the study.
研究设计
- 研究类型
- Observational
- 观察模型
- Cohort
- 时间视角
- Retrospective
入排标准
- 年龄范围
- 18 Years 至 90 Years（Adult, Older Adult）
- 性别
- Female
- 接受健康志愿者
- 否
入选标准
- •age at the MRI examination from 18-90 years
排除标准
- •insufficient image quality as judged by a blinded radiologist before start of the analysis
- •non-identifiably ground truth (i.e., diagnosis has not yet been established)
结局指标
主要结局
Diagnostic performance for breast cancer detection (Sensitivity and specificity)
时间窗: 5 years
Diagnostic performance for breast cancer detection (Sensitivity and specificity) compared to the gold stnandard method of expert-based assessment of breast MRI, may be summarized in a receiver operating characteristic curve for multiple threshold values, comparing multiple technical approaches, including swarm-learning based AI models and local AI models.
次要结局
未报告次要终点