Evaluation of Gabor elementary function-based medical image compression

Due to the high volume of medical imaging studies performed nationwide, and the desire to get away from the maintenance of long-term film archives, storage of digitized medical images is being promoted as an alternative. Computer-based storage would be more practicable if compressed rather than original images could be stored. In this study, compression methods based on Gabor functions were implemented for simulated nuclear medicine liver images, and their performance was assessed objectively. A large number of nuclear medicine liver images with and without space-occupying lesions were simulated. Then various compression schemes based on transformation of the images into the "information space" proposed by Gabor were implemented. Two tasks were examined: (1) determination of the presence or absence of the lesion in a given location, and (2) determination of the presence or absence of the lesion in one of several locations. Gabor-based compression has not previously been implemented on medical images, nor has any rigorous task-based measure of quality been used to assess the compression.; Task-based performance using the compressed/reconstructed images was compared to that using the original images according to the following measures: (1) mean square error, (2) Hotelling trace criterion, an index shown by Fiete and others to correlate with performance for nuclear medicine images, (3) Bayesian maximum likelihood ideal observer signal to noise ratio, and (4) area under the receiver operating characteristic curve. For compression based on thresholding of the complex Gabor coefficients, a better than 2:1 compression of the simulated nuclear medicine liver images was obtained without appreciable reduction in image quality, which when combined with gains expected from bit-reduction schemes, corresponds to an overall approximate 8:1 compression.