Near-lossless compression of PET/CT images using singular value decomposition

Area of Science:

• Medical Imaging
• Data Compression
• Signal Processing

Background:

• Positron Emission Tomography/Computed Tomography (PET/CT) imaging generates large datasets.
• Efficient data compression is crucial for storage and transmission of medical images.
• Singular Value Decomposition (SVD) is a potential method for image compression.

Purpose of the Study:

• To compare the single matrix and slice-by-slice approaches for SVD-based PET/CT image compression.
• To evaluate compression efficiency and image fidelity between the two SVD methods.
• To determine the optimal SVD approach for near-lossless PET/CT image compression.

Main Methods:

• SVD was applied to 49 F-18-FDG PET/CT studies using both single matrix and slice-by-slice methods.
• Image compression was achieved by truncating insignificant singular values.
• Compression ratios, ROI count differences, and image quality metrics were assessed.
• Clinical evaluation by nuclear medicine physicians was performed.

Main Results:

• Both SVD approaches achieved near-lossless compression with less than 5% error.
• The slice-by-slice approach yielded significantly higher compression ratios for PET and CT images.
• Physicians found compressed images visually identical to originals, with no loss of clinical detail or artifacts.
• Objective image quality assessments supported the clinical findings.

Conclusions:

• The slice-by-slice SVD approach provides superior near-lossless compression for PET/CT images compared to the single matrix method.
• This technique enables significant data reduction while preserving essential diagnostic information.
• The slice-by-slice method is recommended for efficient PET/CT image compression.