Back to all papers

Compression of 2D+t cine-MRI data and its impact on AI-based target tracking in MRIgRT.

May 22, 2026pubmed logopapers

Authors

Blöcker TJ

Affiliations (1)

  • Department of Radiation Oncology, University Hospital of Munich, Am Coulombwall 1, Munich, 81377, Germany.

Abstract

Magnetic resonance imaging-guided radiotherapy (MRIgRT) uses 2D+t cine-MRI to track intra-fractional motion with high spatial and temporal resolution. Growing usage and acquisition rates require efficient storage and transfer, especially for large-scale cine-MRI datasets.&#xD;Approach. Lossless and lossy compression algorithms for cine-MRI are systematically evaluated, focusing on efficiency and preservation of downstream utility. Using the TrackRAD2025 dataset, entropy-, image-, and video-based codecs were tested in three scenarios: lossless at 16-bit, lossless after quantization to 8-bit or 10-bit, and lossy at 8-bit. The peak signal-to-noise ratio (PSNR) and mean absolute error (MAE) per normalized pixel were used as standard image quality metrics to assess compression artifacts. Additionally, the impact on machine learning performance metrics such as the Dice similarity coefficient (DSC) of artificial intelligence-based target tracking models was used to determine thresholds for negligible degradation.&#xD;Main results. JPEG XL achieved the highest lossless compression ratios (median 4.4 at 16-bit; 9.9 at 8-bit) with real-time speeds, outperforming the zlib/DEFLATE baseline (2.1 at 16-bit; 4.0 at 8-bit). Quantization reduced file sizes and enabled alternative compression algorithms with small quality loss (median PSNR 59.8 dB for 8-bit and 72.5 dB for 10-bit). For target tracking, compression at PSNR > 45 dB maintained segmentation quality (|DSC| <= 0.005), allowing for compression ratios up to 66 (H.265/HEVC) or 22 (JPEG XL), with or without inter-frame coding. Only at PSNR < 40 dB did model output quality deteriorate slightly for some models (|DSC| > 0.005).&#xD;Significance. The identified compression methods allow for efficient compression of 2D+t cine-MRI data, either losslessly, suitable for archival and quality assurance, or lossily, under well-characterized quality thresholds that preserve target tracking performance, at much higher rates.

Topics

Journal Article

Ready to Sharpen Your Edge?

Subscribe to join 11k+ peers who rely on RadAI Slice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.