Bayesian theory applies to radiology interpretation

June 11, 2007

Since PACS, computed and digital radiography, computer-aided detection, and digital imaging in general have changed the way radiologists view images, understanding precisely how those images are viewed has gained importance.

Since PACS, computed and digital radiography, computer-aided detection, and digital imaging in general have changed the way radiologists view images, understanding precisely how those images are viewed has gained importance.

A new study uses Bayesian probability to more completely comprehend the radiology interpretation task (BJR 2007;10:1259).

"With the introduction of PACS, CR/DR, and CAD, the time has come to focus attention on radiologists to optimize the design of viewing stations with respect to the human visual system," said lead author Timothy Donovan, a medical imaging sciences lecturer at St. Martin's College in the U.K.

Donovan said a Bayesian approach is attractive because it allows researchers to model the ideal observer, then compare human and ideal search quantitatively. The approach divides the perceptual task into pieces and then combines them to understand the whole.

There is often conflict between the way CAD operates and the way radiologists reach decisions, according to Donovan.

"Despite all the advances in CAD and decision support software, it is the human factors that are a critical part of the interpretative process and that transform image appearances into a differential diagnosis," he said.

Understanding and modeling radiologists' perceptual processes and incorporating this information into imaging technology with the aid of Bayesian theory can help improve observer performance.

Earlier work found that while the detection of, say, chest lesions did not vary among the technologies of plain film, CR hard-copy, and PACS soft-copy images, statistically significant differences occurred among observers (Eur J Radiol 2003;47:206-214).

"It is pertinent that medical image perception research should concentrate on the observer as well as the technology, as it seems interobserver differences are often greater than differences between imaging techniques," Donovan said.

Bayesian theory is a handy tool when attempting to explain such phenomena as eye-tracking patterns and to provide insight into understanding image interpretation. It provides a mathematical framework for representing the properties of the image, describing the image interpretation task, and taking account of costs and benefits associated with different perceptual decisions, Donovan said.

"These findings are relevant in the design of viewing and reporting stations so that design can be matched to human abilities and limitations," he said.