Pruned tree-structured vector quantization of medical images with segmentation and improved prediction

Area of Science:

• Medical Imaging
• Image Compression
• Computer Vision

Background:

• Medical image compression is crucial for storage and transmission.
• Maintaining diagnostic quality during compression is a significant challenge.
• Existing methods often struggle to balance compression ratio and image fidelity.

Purpose of the Study:

• To develop a novel image compression method for medical images.
• To achieve high compression ratios without compromising diagnostic image quality.
• To reserve bits for diagnostically relevant areas within medical scans.

Main Methods:

• Utilized predictive pruned tree-structured vector quantization (PP-TSVQ).
• Incorporated a priori knowledge for image segmentation to prioritize diagnostically relevant regions.
• Enhanced predictor memory and employed ridge regression for improved prediction accuracy.
• Tested the scheme on mediastinal CT scans.

Main Results:

• Achieved remarkable improvements in prediction accuracy and encoding quality compared to conventional methods.
• Successfully encoded test images at 0.5 bits per pixel and below.
• Demonstrated no visible degradation in diagnostically relevant regions at low bit rates.

Conclusions:

• The proposed PP-TSVQ method offers superior performance for medical image compression.
• Effective segmentation and enhanced prediction significantly improve compression efficiency and quality preservation.
• This technique holds promise for efficient storage and transmission of medical imaging data.