Errors in interpreting medical images have significant health care consequences. Errors that result from failures of perception occur with all medical images, even those based on sophisticated tomographic technology such as CT, MRI, PET/CT and tomosynthesis. Basic scientific research on image perception strives to uncover the causes of these errors and provide remedies. In the past, perception research has relied on a very limited range of simulated abnormalities from phantoms or on diverse “found” abnormalities in clinical studies. Compiling samples of proven normal and abnormal imaging examinations by that approach is extremely time consuming and limits perception and image processing research.
The broad, long-term objective of this research is to advance perception research in tomographic imaging modalities. The overall goal is to provide editing and manipulation tools to greatly expand the range of images available for perception studies including those in which computers serve as observers.
These goals will be accomplished through four specific aims: (1) develop software to remove localized three-dimensional abnormalities from medical tomographic images without leaving any trace of image manipulation; (2) develop software to capture abnormal areas from specific organs and specific modalities and organize the volumetric collections of these abnormal areas into library files; (3) develop interactive software to insert abnormalities from the libraries into medical tomographic images without introducing artifacts that would identify them as artificially placed; and (4) provide for dissemination of the software and libraries to the medical image perception, computer-aided diagnosis, and observer modeling communities.
By making available a large domain of imaging abnormalities for use in experiments, this project will enable more and better research to reduce or eliminate diagnostic error, thereby improving public health.