Al medical images are inherently noisy. This image noise sets fundamental limits of how accurately one can detect, identify, or recognize lesions and how accurately one can estimate lesion parameters in medical images. The limit is determined using an information theory analog of the second law of thermodynamics; the concept of an ideal observer that """"""""does everything correctly"""""""" and operates at this limit. My experimental strategy has been to measure human performance for detection, identification and localization of visual signals in computer generated, noise limited images. Human performance is then compared with this ideal to identify sources of human inefficiency and to test hypotheses about possible strategies used by humans to do various visual tasks. I have slowly increased the complexity of the observer tasks and propose to continue in this direction. The long range goal is to obtain a useful model and describes physician performance for visual diagnostic tasks of the type encountered in noise-limited medical images. The objective of this project is to test some fundamental and important hypotheses suggested by my previous work and that of others. There are aims which involve investigation of 1) performance with statistically defined backgrounds, 2) performance with anti-correlated noise, 3) effect of varying prior probabilities, 4) variation of efficiency with display luminance, 5) spatial statistics properties of certain classes of clinical images, and 6) measurement of performance reduction by these clinical backgrounds. All the experiments except the last set will be done using forced choice psychophysics methods and human observer results will be compared to specific models of sub-optimal Bayesian observers. The images have noise similar to that found in x-ray, CT, MRI, and nuclear medicine images. The results of this research are useful to the designer of medical imaging equipment. The increased experimental and theoretical understanding of human visual capabilities and limitations provides a stronger basis for the theoretical analysis of potential benefits of changes to medical imaging data acquisition, image reconstruction and image display methods.
| Burgess, A E; Jacobson, F L; Judy, P F (2001) Human observer detection experiments with mammograms and power-law noise. Med Phys 28:419-37 |
| Burgess, A E (1999) The Rose model, revisited. J Opt Soc Am A Opt Image Sci Vis 16:633-46 |
| Burgess, A E (1999) Visual signal detection with two-component noise: low-pass spectrum effects. J Opt Soc Am A Opt Image Sci Vis 16:694-704 |
| Burgess, A E; Li, X; Abbey, C K (1997) Visual signal detectability with two noise components: anomalous masking effects. J Opt Soc Am A Opt Image Sci Vis 14:2420-42 |
| Burgess, A (1995) Image quality, the ideal observer, and human performance of radiologic decision tasks. Acad Radiol 2:522-6 |
