The continuous development of imaging technology offers great promise in areas where visual detection is needed, such as with cancer screening. Three-dimensional images in particular have become popular because it gives a fuller picture of the object and its context.
“More doctors and radiologists are looking at these 3D books, which are new technologies that allow you to look at not just one image, but a set of images,” said UC Santa Barbara professor of psychology Miguel Eckstein (external link), whose experience lies in the field of visual inspection. “In some imaging modalities this informs doctors about size and allows them to separate the objects they are interested in.”
It is common wisdom that the detection success rate should be greatly increased with this additional information provided. But that’s not always the case, Eckstein said. In a study (link is external) published in the journal Conventional biology, it is, lead author Miguel Lago and his colleagues express a strange view of human perspective: We find it even worse to find small targets in 3D image stacks than it would be if they were in one 2D image.
“For those kind of small targets, what happens is that they become increasingly difficult to find in those 3D books,” Eckstein explained. Unlike humans, observers of devices (e.g., deep neural networks) did not show this deficiency with small targets in 3D analysis, suggesting that the effect is related to some human visual bottlenecks.
It is a phenomenon that could have a significant impact in the medical field, especially in the world of breast cancer screening with more and more breast tomosynthesis (3D mammography) to detect not only large masses, or -normal but also microcalcifications that may indicate the onset of cancer. development. According to the study, analysis through 3D proportions led to high target loss rates and significantly reduced decision confidence from the viewer.
“Another thing we found was that when you asked people who were studying these 3D books what they studied, they tended to overestimate what they thought. based on results from eye-tracking software, subjects doing the 3D analysis only looked at about half of the survey area while reporting up to more than 80 % image analysis.
Much of the reason for this reduced performance, according to the paper, is how we use our perspective when conducting research. We use both focused and peripheral insights to examine what lies ahead and decide where next to focus our attention. People searching through a 2D image tended to rely more on their fovea (the part of the retina that brings objects into sharp, vertical focus) and shifted the focus around the more complete image. Those who searched through 3D fractions – collections of multiple images – were found to shift their gaze less and rely on peripheral visual processing.
“What happens is when doctors look through these 3D images, they are basically disregarding the entire data set,” said Eckstein, who had the colleagues in the University of Pennsylvania Department of Radiation replicate the effect by some radiologists. “They don’t look everywhere on every image, because it takes a long time.” The lack of eye movement in 3D searches could also be a strategic issue, he said, in whether clinicians fix in the same place in all images as they go back and forth through the stack.
Small targets, Eckstein explained, were very visible at or near the settlement point but were less obvious when moving toward the edge. This fundamental visual limitation, under-examination of eye movement and reliability in peripheral vision resulted in a large number of errors in the 3D detections.
The same could not be said for large targets, which followed the common wisdom about the benefits of 3D images; their discovery was enhanced in the 3D findings.
The conclusions of this paper highlight the gaps that sometimes arise between the technology we create and our ability to make the most of it, according to Eckstein.
“We’re good at making technology, but sometimes we don’t make a good connection with that,” he said. “And we don’t know that we’re not making a good connection with it. “
For radiologists combing through 3D images for small targets, this bottle of human vision and understanding, once identified, could be enhanced by extended use and study times. In some cases, clinicians already pursue synthesized 2D images for the small targets while using 3D proportions for the large products. Performance can also be improved by the use of computer vision, artificial intelligence and / or by a number of viewers studying images.
###
UCSB lead author and former postdoctoral scholar Miguel Lago is now a researcher in the Department of Food and Drug Administration Imaging, Diagnostics, and Software Reliability, which contributes to the evaluation of new medical imaging technologies .
Disclaimer: AAAS and EurekAlert! they are not responsible for the accuracy of press releases posted to EurekAlert! by sending institutions or for using any information through the EurekAlert system.