Noise in medical images sets fundamental limits for lesion detection accuracy. Patient structure also plays an important role for both humans and computer-aided diagnosis algorithms. The authors previous experimental strategy was to measure human performance for a variety of tasks in computer generated images. Performance with simple tasks and backgrounds is now reasonably well understood and the time is ripe to investigate performance with clinical images to test the models and ensure that there are no overlooked important issues. The long range goal is to obtain a useful model that describes physician performance for clinical diagnostic tasks. In the first period this project the authors did precise human performance measurements for (1) several tasks using synthetic, statistically defined backgrounds with added white noise and (2) a variety of tasks using uniform backgrounds and negatively correlated noise. The objective was to test three popular observer models, the modified non-prewhitening model, the Fisher-Hotelling (FH) model and the simplest channelized FH model. The new experiments showed that none of these models is satisfactory. The applicant showed that more complex channelized FH models did give satisfactory predictions. In the next period of the grant the applicants proposed to investigate observer performance using clinically realistic lesions added to two classes of digital clinical images, chest radiographs and mammograms. These classes of images are an important fraction of radiography, there are validated methods for simulating lesions and the amount of patient structure is significant. One goal of this research is to determine the relative contributions of patient structure and quantum noise to reducing the detectability of lesions under a variety of conditions. The second goal is to incorporate patient structure effects into a more sophisticated FH model with overlapping frequency channels and to evaluate this modified model by three methods. These research results will be useful to the designers of medical imaging equipment and CAD algorithms. The increased experimental and theoretical understanding of human visual capabilities and limitations will provide a stronger basis for the analysis of potential benefits of changes to data acquisition, image reconstruction, image compression and 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 |
