haca3-mri-harmonization

name: haca3-mri-harmonization version: 1.0.0 description: "HACA3+ MRI harmonization algorithm validated across 100+ scanners with traveling subjects. Incorporates improved artifact encoder, comprehensive multi-site validation, and real-world protocol robustness testing. Most comprehensive multi-site MRI harmonization validation to date. arXiv:2604.19474." date: 2026-04-23 arxiv_id: "2604.19474" authors: "Savannah P. Hays, Lianrui Zuo, Muhammad Faizyab Ali Chaudhary, Kathleen M. Bartz et al." categories: "eess.IV" activation: - MRI harmonization - multi-site neuroimaging - scanner harmonization - artifact removal - HACA3 - traveling subject validation - clinical trial imaging - ComBat MRI

HACA3+: Harmonizing MR Images Across 100+ Scanners

Overview

Presents HACA3+, an enhanced MRI harmonization algorithm validated with traveling subjects across 100+ scanners. Addresses the critical challenge of combining heterogeneous MR data from multi-center clinical trials by removing scanner-specific artifacts while preserving biological signal.

Key Methodology

HACA3+ Enhancements over HACA3

1. Improved Artifact Encoder: Better isolation and mitigation of scanner-specific image artifacts
2. Traveling Subject Validation: Ground-truth validation using same subjects scanned across 100+ sites
3. Real-World Protocol Robustness: Tested on pragmatic clinical trial acquisition protocols (not just research-grade data)

Algorithm Pipeline

1. Image preprocessing: Standardize resolution, orientation, intensity range
2. Artifact encoding: Extract scanner-specific artifact features using improved encoder
3. Harmonization transform: Apply site-adaptive normalization while preserving biological variation
4. Quality control: Automated checks for residual scanner effects

Validation Framework

• Traveling subjects: Same individuals scanned at multiple sites provide ground truth
• Quantitative metrics: Intra-class correlation (ICC), coefficient of variation (CV)
• Downstream tasks: Validate harmonization preserves diagnostic utility
• Scanner diversity: 100+ unique scanners across manufacturers (Siemens, GE, Philips)

Implementation Guidance

• Input: T1-weighted or T2-FLAIR MR volumes
• Preprocessing: N4 bias field correction, skull stripping, registration to template
• Model: Encoder-decoder architecture with artifact disentanglement
• Output: Harmonized volumes with reduced inter-scanner variance

Advantages

• Largest multi-site MRI harmonization validation (100+ scanners)
• Works with pragmatic clinical trial data (not just research acquisitions)
• Preserves biological variation while removing scanner effects
• Improved artifact handling over predecessor methods

Pitfalls

• Requires sufficient per-site samples for reliable harmonization
• May not fully handle extreme protocol deviations
• Computational cost scales with number of sites
• Validation limited to specific MRI contrasts (T1w, T2-FLAIR)

References

• arXiv: 2604.19474
• Key terms: MRI harmonization, multi-site imaging, traveling subjects, artifact removal, clinical trials, neuroimaging