AIMER: An Open-Source Federated Learning Platform for Medical Research

AIMER is an open-source federated learning platform designed specifically for medical artificial intelligence research, aiming to resolve the conflict between data privacy and cross-institutional collaboration in the medical AI field. The platform integrates data privacy protection and distributed model training using a multi-package workspace architecture, consisting of three core modules: AIMER-ROOT (Web application and UI layer), MAGE (machine learning service gateway), and FARM (data and workflow support package). It features medical scenario-adapted capabilities such as differential privacy and secure aggregation, supports practical applications like oncology, drug development, and rare disease research, and promotes open collaboration and compliant innovation in medical AI.

In the field of medical artificial intelligence, data is the lifeline of model training, but patient privacy protection regulations (such as GDPR and HIPAA) pose significant barriers to data sharing: hospitals face difficulties in cross-institutional collaboration due to compliance concerns, and research institutions are limited by the scale and diversity of single datasets. Federated learning emerged as a solution, allowing participants to collaboratively train models without sharing raw data (data remains local, encrypted model parameters circulate). AIMER is the concrete practice of this concept in the medical AI field.

AIMER adopts a multi-package workspace design, broken down into three complementary sub-projects:

• AIMER-ROOT: Core Web application and UI layer, providing researchers with an intuitive operation entry (initiating tasks, monitoring performance, managing nodes);
• MAGE: Machine learning service and model testing coverage layer, with an API gateway supporting RESTful interfaces and MCP protocol, and the service layer following a microservice architecture;
• FARM: Data and workflow support package, responsible for data pipeline orchestration, preprocessing standardization, and cross-node task scheduling coordination.

AIMER has a complete CI/CD system and quality gate mechanism: Docker image building ensures environment consistency; code coverage checks enforce test adequacy; global code quality scanning integrates CodeQL to detect security vulnerabilities; license compliance checks avoid legal risks of open-source components. These engineering practices are crucial for medical software, shifting quality left to the development phase to ensure patient safety and institutional compliance.

Medical scenarios have strict requirements for privacy protection, and AIMER's architecture is designed accordingly:

• Differential privacy support: Calibrated noise is injected during parameter aggregation to prevent individual data from being inferred from gradients;
• Secure aggregation protocol: Cryptographic technology ensures only the aggregated global model is exposed, with individual updates encrypted;
• Fine-grained access control: Hospital administrators can precisely control data participation in training and model access permissions.

AIMER has a wide range of potential applications:

• Oncology: Multiple hospitals jointly train cancer detection models to improve generalization ability;
• Drug development: Pharmaceutical companies collaborate with clinical trial centers to accelerate the evaluation of new drug efficacy;
• Rare disease research: Aggregate scattered cases worldwide to solve the problem of data sparsity.

AIMER is not only a technical project but also a practice of the open collaboration concept in medical AI. It encapsulates the complex engineering of federated learning into a deployable and scalable open-source platform, lowering the threshold for medical institutions to participate in AI innovation and balancing data privacy with technological progress. For developers concerned with medical AI ethics and engineering practices, it is a reference implementation worth in-depth study.