Clinical Report: Impact of Sequential Institution Integration on CAD Performance
Overview
This study investigates the effect of sequentially adding institutions in a federated learning framework on the performance of computer-aided detection (CAD) software for lesion detection.
Background
The integration of machine learning in medical imaging has led to significant advancements in CAD systems, which are increasingly utilized in clinical settings. However, the performance of these systems can be compromised if the training data do not reflect the clinical data. Federated learning offers a solution by allowing institutions to collaborate on model training while keeping their data secure and local.
Data Highlights
The study utilized data from 423 MRA images and 398 contrast-enhanced T1-weighted MR images. The evaluation included two types of CAD software: one for detecting cerebral aneurysms and another for identifying brain metastases. Various training approaches were implemented, including from-scratch training, fine-tuning, and full fine-tuning.
Key Findings
Sequential addition of institutions in a federated learning framework enhances CAD software performance.
Two types of CAD software were evaluated: one for cerebral aneurysms and another for brain metastases.
Different training approaches were tested, including from-scratch training, fine-tuning, and full fine-tuning.
Data from 423 MRA images and 398 contrast-enhanced T1-weighted MR images were used for the study.
Federated learning allows for local data retention while improving model generalizability.
Clinical Implications
The findings indicate that incorporating data from multiple institutions can improve the robustness of CAD systems.
Conclusion
The study demonstrates that sequentially integrating institutional data within a federated learning framework can significantly improve the performance of CAD software for lesion detection.
Related Resources & Content
Suraj Rajendran et al., ASCO Publications, 2021 -- Cloud-Based Federated Learning Implementation Across Medical Centers
Collaborative 3D segmentation of multiple organs using both labeled and unlabeled datasets, Springer, 2024
Is Federated Learning the Answer to Patient Privacy in AI?, ASCO AI in Oncology, 2026
Transcript for CDRH Webinar Final Guidance: Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions, FDA, 2024
Screening performance and characteristics of breast cancer detected in the Mammography Screening with Artificial Intelligence trial (MASAI), ScienceDirect, 2024
Federated Learning with Contrastive Prototypes for Detecting Fetal Standard Planes Amid Noisy Labels
Transcript for CDRH Webinar Final Guidance: Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions
Screening performance and characteristics of breast cancer detected in the Mammography Screening with Artificial Intelligence trial (MASAI): a randomised, controlled, parallel-group, non-inferiority, single-blinded, screening accuracy study - ScienceDirect
Advancements and challenges of federated learning in medical imaging: a systematic literature review | Artificial Intelligence Review | Springer Nature Link
