Federated orthogonal learning for detection of liver lesions from multi-phase contrast-enhanced CT images

2026-06-08 · Source: Machine learning : nature.com subject feeds · Field: Health & Wellbeing — Medical Devices & Health Technology, Artificial Intelligence & Machine Learning, Clinical Care & Medical Practice · Depth: Expert, short

Summary

FedOG, a novel federated orthogonal learning method, addresses challenges in detecting liver lesions from multi-phase contrast-enhanced CT (CECT) images across multiple institutions. Due to data protection and incomplete CECT phases, standard federated learning often degrades performance from data heterogeneity. FedOG mitigates this by adjusting gradients from local models, trained with incomplete CECT phases, using orthogonal gradient decomposition to reduce interference. Bayesian optimization then determines the optimal gradient for updating the global model. Tested on 3,668 multi-phase CECT scans from five institutions, FedOG improved the Dice score by 1.67% on a real-world clinical dataset, and by 1.13% and 3.03% on two public datasets. This approach offers a heterogeneity-robust, search-efficient, and privacy-preserving framework, particularly beneficial for regions with limited or low-quality CECT data.

Key takeaway

For AI Scientists and Research Scientists developing federated learning solutions for medical imaging, you should consider integrating orthogonal gradient decomposition and Bayesian optimization. This approach, exemplified by FedOG, significantly improves model performance and robustness when dealing with heterogeneous datasets and incomplete local data, such as multi-phase CECT scans. Implementing such a framework can enhance diagnostic accuracy in privacy-sensitive environments, especially in regions with varying data quality, ensuring more reliable lesion detection.

Key insights

FedOG enhances federated learning for medical image segmentation by using orthogonal gradient decomposition to counter data heterogeneity and incomplete phases.

Principles

• Data heterogeneity in federated learning can be mitigated by orthogonal gradient decomposition.
• Bayesian optimization can determine optimal global model updates in federated settings.
• Privacy-preserving models can be robust to incomplete local data.

Method

FedOG adjusts local model gradients via orthogonal gradient decomposition to minimize interference from incomplete CECT phases, then uses Bayesian optimization to determine the optimal global model update gradient.

In practice

• Implement orthogonal gradient decomposition in federated learning for heterogeneous medical imaging data.
• Apply Bayesian optimization to refine global model updates in privacy-preserving training.
• Consider FedOG for medical image analysis in regions with varied data quality.

Topics

• Federated Learning
• Liver Lesion Detection
• Medical Imaging
• Contrast-Enhanced CT
• Orthogonal Gradient Decomposition
• Bayesian Optimization

Best for: AI Scientist, Research Scientist

Related on AIssential

• Federated learning with swarm intelligence for efficient and secure medical image analysis — Swarm intelligence-enhanced federated learning improves medical image analysis accuracy and efficiency while preserving privacy.
• Cut instance mixing: A domain-specific data augmentation method applied to gastrointestinal lesion detection — A novel domain-specific data augmentation method, Cut Instance Mixing (CIM), has been introduced to address the scarcity of well-annotated medical image datasets and the subtle appearance of gastroint…
• Fairness-Aware Federated Learning with Trajectory Shapley Value — TSV and FedTSV provide dynamic, fairness-aware aggregation for federated learning by evaluating client influence on model trajectory.
• A federated graph learning method to realize multi-party collaboration for molecular discovery — FedLG enables privacy-preserving multi-party molecular discovery through federated graph learning and the Lanczos algorithm.
• M$^2$FedAQI: Multimodal Federated Learning for Air Quality Prediction on Heterogeneous Edge Devices — M^2FedAQI offers a multimodal federated learning solution for accurate, private, and efficient air quality prediction on edge devices.
• A Data-Centric Framework for Intraoperative Fluorescence Lifetime Imaging for Glioma Surgical Guidance — A data-centric AI framework improves FLIm accuracy for glioma surgery by refining labels and enhancing model robustness.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine learning : nature.com subject feeds.