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

Area of Science:

• Medical Imaging
• Radiotherapy Technology
• Data Compression

Background:

• Magnetic resonance imaging-guided radiotherapy (MRIgRT) generates large 2D+t cine-MRI datasets.
• Efficient storage and transfer are essential due to increasing usage and acquisition rates.

Purpose of the Study:

• To systematically evaluate lossless and lossy compression algorithms for cine-MRI data.
• To assess the impact of compression on image quality and downstream machine learning utility, specifically for target tracking.

Main Methods:

• Tested entropy-, image-, and video-based codecs on the TrackRAD2025 dataset.
• Evaluated lossless compression at 16-bit and 8-bit, and lossy compression at 8-bit.
• Assessed image quality using PSNR and MAE, and machine learning performance using Dice Similarity Coefficient (DSC).

Main Results:

• JPEG XL achieved superior lossless compression ratios (up to 9.9 at 8-bit) with real-time speeds.
• Quantization enabled effective compression with minimal quality loss (PSNR > 59.8 dB).
• Compression at PSNR > 45 dB maintained AI target tracking accuracy (DSC <= 0.005), allowing high compression ratios (up to 66 with H.265/HEVC).

Conclusions:

• Identified compression methods offer efficient lossless or lossy compression for 2D+t cine-MRI.
• Lossless compression is suitable for archival and quality assurance.
• Lossy compression, within defined quality thresholds, preserves target tracking performance at significantly higher rates.