Diagnostically lossless compression strategies for x-ray angiography images
- Autores: Zhongwei Xu
- Directores de la Tesis: Joan Serra Sagristà (dir. tes.), Joan Bartrina Rapesta (codir. tes.)
- Lectura: En la Universitat Autònoma de Barcelona ( España ) en 2015
- Idioma: español
- Tribunal Calificador de la Tesis: Vicente González Ruiz (presid.), Ian Blanes Garcia (secret.), Antoni Bardera Reig (voc.)
- Materias:
- Enlaces
- Resumen
The past several decades have witnessed a major evolution in medical imaging techniques, making medical images become commonplace in healthcare systems and an integral part of a patient medical record. Among the existing medical imaging modalities, X-ray imaging is one of the most popular technologies due to its low cost, high resolution and excellent capability to penetrate deep within tissue. In particular, X-ray angiographies --which use minimally invasive catheterization-- and X-ray imaging are widely used to identify irregularities in the vascular system. X-ray angiography images can be classified into two types: general X-ray angiography (GXA) images, which present blood vessels in several body parts like arms, legs, foots, etc.; and coronary angiogram video sequences (CAVSs), which only focus on coronary vessel trees for diagnosing cardiovascular diseases. Because of the differences in functions, these two types of images have different features: GXA images normally have high spatial resolutions (the width and height sizes) but low temporal resolution (the number of frames), while CAVSs usually have lower spatial resolutions but higher temporal resolution.
Due to the increasing number of medical studies using X-ray angiography images and the need to store and share them, compression of these images is becoming critical. Lossy compression has the advantage of high data reduction capability, but it is rarely accepted by medical communities because of the modification of data that may affect the diagnosis process. Lossless compression guarantees perfect reconstruction of the medical signal, but results in low compression ratios. Diagnostically lossless compression is becoming the preferred choice, as it provides an optimal trade-off between compression performance and diagnostic accuracy. In diagnostically lossless compression, the clinically relevant data is encoded without any loss while the irrelevant data is encoded with loss. In this scenario, identifying and distinguishing the clinically relevant from the clinically irrelevant data in medical images is the first and usually most important stage in diagnostically lossless compression methods.
In this thesis, two diagnostically lossless compression strategies are developed. The first one is proposed for GXA images. The second one is proposed for CAVSs. For GXA images, the clinically relevant focal area in each frame is first identified; and then a background-suppression approach is employed to increase the data redundancy of the images and hence improve the compression performance. For CAVSs, a frame-identification procedure is implemented to recognise the diagnostically unimportant frames that do not contain visible vessel structures; then, lossy compression is applied to these frames, and lossless compression is applied to the other frames.
Several compression techniques have been investigated for both types of images, including the DICOM-compliant standards JPEG2000, JPEG-LS and H.264/AVC, and the latest advanced video compression standard HEVC. For JPEG2000, multicomponent-transform and progressive lossy-to-lossless coding are also tested. Experimental results suggest that both the focal-area-identification and frame-identification processes are automatic in computation and accurate in clinically relevant data identification. Regarding the compression performance, for GXA images, when compared to the case of coding with no background-suppression, the diagnostically lossless compression method achieves average bit-stream reductions of as much as 34% and improvements on the reconstruction quality of up to 20 dB-SNR for progressive decoding; for CAVSs, the frame-identification followed by selective lossy & lossless compression strategy achieves bit-stream reductions of more than 19% on average as compared to lossless compression.
