Compression and Enhancement of Medical Ultrasound Images using Optimal Quantization

Background: Compression of medical images is vital for storage and transmission, while enhancement and de-noising is important to meet diagnostic requirements. We introduce a method for simultaneous enhancement and compression of medical ultrasound images, based on the notion that the inherent speckle pattern may be treated as noise. Since typically, the speckle noise is characterized by high spatial frequencies in the image, suppression of these frequencies yields a de-speckled image.

Method: To achieve this goal, we apply a transform representation of the input image and quantize the transform coefficients which correspond to the high frequencies. We use the wavelet transform to obtain a sub-band representation, in order to avoid undesired artifacts in the restored image. Furthermore, in order to obtain optimal quantization parameters, we construct a combined distortion measure of Peak-Signal-to-Noise Ratio (PSNR), Speckle Index (SI) and Edge Preservation (EP). This measure is computed with respect to a de-speckled image that is obtained using a pre-determined de-speckling algorithm, such as the Non-Local Means (NLM) method. For a given desired compression ratio, the quantization parameters are iteratively updated such that for each step, the distortion measure is minimized. Then, the obtained parameters are applied to quantize the wavelet coefficients and compress the image.

Results: Our algorithm was tested on liver and breast images and yielded de-speckled and compressed images, with compression ratios of the order of 40:1.

Conclusions: The results demonstrate that our method enables to compress medical ultrasound images while efficiently de-noising them as in commonly used de-speckling methods.