Novel JPEG 2000 compression for faster medical image streaming and diagnostically lossless quality

摘要

Electronic health records can significantly improve productivity for clinicians as well as quality of care for patients. However, implementing highly available and universally accessible electric health records can be very challenging. This is in part due to the tremendous amount of data produced every day by modern diagnostic imaging devices. This data must be instantly available for remote consultation and must be archived for very long periods, at least until the patient’s death. Image compression can be used to mitigate this issue by reducing both network and storage requirements. Lossless compression can reduce file sizes by up to two thirds. Further improvements require the use of lossy compression where the original signal cannot be perfectly reconstructed. In that case, great care must be taken as to not alter the diagnostic properties of the acquired image. The current standard practice is to rely on compression ratio guidelines published by professional associations. However, image compressibility is known to vary significantly based on image content. Therefore, in order to be consistently safe, recommendations based on compression ratios have to be very conservative. At the same time, medical images are usually displayed after a value of interest (VOI) transform that can mask some of the image content leading to needless data transfers. Our objective is to improve medical image compression and streaming to achieve better efficiency while ensuring adequate diagnostic quality. To achieve this, 1- we have highlighted the limitations of compression ratio based guidelines by analyzing the effects of acquisition parameters and image content on the compressibility of more than 23 thousand computed tomography slices of a thoracic phantom, 2- we have proposed a streaming scheme that leverages the masking effect of the VOI transform and can scale from lossy to near-lossless and lossless levels and 3- we have proposed an alternative to compression scheme tailored especially for diagnostic imaging by leveraging the beneficial denoising effect of compression while preserving important structures. Our results showed significant compression variability, up to 66%, between series. Furthermore, 15% of the images compressed at 15:1, the maximum recommended ratio, had lower fidelity than the median of those compressed at 30:1. With our VOI-based streaming, we have shown a reduction in network transfers of up to 54% for near-lossless levels depending on the targeted VOI. Our solution is also capable of streaming between 20 and 36 slices per second with the first slice displayed in less than a second. Finally, our new compression constraint showed drastic reduction in structure degradations and the performances of the derived metric were on par with other leading metrics for compression distortions.