Current and anticipated rates of acquisition of digital medical images threaten to overwhelm the available capacities for digital storage media and digital communication links, even with the arrival of optical disk storage and optical fiber communication. This necessitates efficient coding and compression of images for both archives and communication. The goal of the proposed research is to extend existing techniques and develop new techniques for image coding and compression well matched to medical image applications, The propose algorithms combine recent advances from vector quantization (vector coding, block quantization) and decision tree design and combine aspects of compression with low-level classification so as to permit the best (or fastest) reproduction in areas of an image of most interest to the user. The research will be conducted by specialist in coding and signal processing in cooperation with radiologists.
The specific aims of the research are: 1. To develop algorithms for compressing medical images to one bit per pixel (picture element) and less with little or no loss in perceived image quality or diagnostic accuracy using vector quantization, a form of lossy data compression that optimizes fidelity for a given bit rate in bits per pixel. This will provide compression ratios of at least 8:1 for typical 8 bit per pixel monochrome images. We believe good quality images will be feasible at rates of 1/2 bit per pixel (16:1) and less. The algorithms foe encoding and decoding should be implementable in real time using current technology. Specific algorithms to be considered include tree- structured vector quantizers and recursive vector quantizers (including predictive and finite-state codes). Special emphasis will be given to progressive transmission applications where quality improves as additional bits arrive from storage or communication media. 2. To evaluate clinically the quality of the compressed images at various compression ratios. Image evaluation will be done by diagnostic radiologists in experiments designed in conjunction with biostatisticians to judge quantitatively and qualitatively the coded images both for the detection of important features and for the preservation of selected measurements. The primary applications considered will be computes tomography (CT) and magnetic resonance (MR) images (including CINE mode MR images consisting of sequences of MR heart images).
| Cosman, P C; Davidson, H C; Bergin, C J et al. (1994) Thoracic CT images: effect of lossy image compression on diagnostic accuracy. Radiology 190:517-24 |
