Medical image compression by discrete cosine transform spectral similarity strategy

Area of Science:

• Medical Imaging
• Image Compression
• Signal Processing

Background:

• Bandwidth and storage limitations necessitate medical image compression.
• High compression ratios often lead to significant image degradation and artifacts, risking misdiagnosis.
• Existing compression methods like JPEG struggle to balance compression efficiency with image fidelity.

Purpose of the Study:

• To develop an efficient medical image compression strategy with a simple computational burden and superior decoded quality.
• To enhance compression ratios without compromising the diagnostic integrity of medical images.
• To improve upon existing transform coding schemes for medical image applications.

Main Methods:

• Decomposing image sub-blocks into bands using the Discrete Cosine Transform (DCT) as a bandpass filter.
• Applying a band gathering operation to identify and exploit high similarity among the decomposed bands.
• Utilizing the discovered similarity property to achieve significant bit rate reduction.

Main Results:

• The proposed method achieves substantial bit rate reduction by exploiting band similarity.
• Demonstrated superior performance over JPEG compression for various medical images, particularly angiograms.
• Angiogram images showed a 13.5 dB Peak Signal-to-Noise Ratio (PSNR) gain at 0.15 bits per pixel compared to JPEG.
• Other medical images achieved 4-8 dB PSNR gains at high compression ratios.
• Expert verification confirmed correct diagnoses for all tested images compressed below a ratio of 20.

Conclusions:

• The novel compression strategy effectively increases compression ratios while preserving crucial image characteristics.
• The method offers a significant improvement in image quality and diagnostic reliability compared to conventional JPEG compression.
• This approach provides a viable solution for efficient medical image transmission and storage, mitigating diagnostic risks associated with severe compression.