31 Aug How Can Radiologists Detect Cancer In a Fraction of a Second?
MedicalResearch.com Interview with:
Dr. Karla Evans
Karla K. Evans, Ph.D.
Lecturer, Department of Psychology
The University of York
Heslington, York UK
MedicalResearch.com: What is the background for this study?
Response: This research started after initially talking to radiologists and pathologists about how they search a radiograph or micrograph for abnormalities. They talked about being able to tell at the first glance if the image had something bad about it. Jokingly, they talked about “having the force” to see the bad. We wanted to know whether this hunch after the brief initial viewing was real and to systematically test it. We collected radiographic and micrographic images, half of them that had signs of cancer in them and half of them that didn’t, and we briefly presented them (250 millisecond to 2000 milliseconds) to radiologists or pathologistsrespectively. They simply had to report whether they would recall the patient or not and try localize on the outline the location of the abnormality. We first reported these finding in the following paper.
Evans et al. (2013) The Gist of the Abnormal: Above chance medical decision making in the blink of an eye. Psychonomic Bulletin & Review (DOI) 10.3758/s13423-013-0459-3
In addition to finding that radiologists and pathologists can indeed detect subtle cancers in a quarter of a second we also found that they did not know where it was in the image leading us to conclude that the signal that they were picking up must be a global signal (i.e. the global image statistic or the texture of the breast as a whole) rather than the result of a local saliency. This led me to start further exploring this signal in order to characterize it when I moved to University or York, UK to establish my own lab.
MedicalResearch.com: What are the main findings?
Response: The current findings published in PNAS are the result of that work, looking into where might this signal lie and how to characterize it using the same brief presentation paradigm but this time looking only into radiographs, specifically mammograms. Radiologists in regular practice look at a pair of breasts and if the breasts appear asymmetrical or have certain density to radiologists that often conveys that there is something a miss. We find that though symmetry is important it is not the crucial part of the signal we observe, since when we break it or just show one of the breasts, the radiologists can still tell that there was an abnormality. Like symmetry we find that breast density is also not part of the signal.
The fact that radiologists can detect subtle signs of cancer in a single breast brought us to one of the most interesting findings and that is that not only did the radiologists do well when we briefly presented the breast with cancer but they also beat chance when they viewed the other breast (contralateral breast) that did not have reportable sign of cancer. They were able to detect an abnormality in the breast that was from a woman with cancer but it itself did not have any visible lesions.
We were also able to show in one of our experiments that the signal radiologist extract in a blink of an eye lies in the fine detail of the fibrous structures of the breast since radiologists were able to preform the detection on these images (i.e. mammograms with only high spatial frequencies) as well as when they were shown the unaltered whole mammogram. However if you blurred the image (showed them only the low spatial frequencies of the mammogram) they did much worse. This means that the signal resides in the high–spatial frequencies of the image.
MedicalResearch.com: What should readers take away from your report?
Response: The one thing that readers should take away from this report is that our results suggest that there may be something in the supposedly normal breast that can be detected and to experts looks abnormal. The radiologists when briefly viewing images without having to report on the location can pick up this early global signal of the presence of an abnormality that is still unknown thus far. Characterizing this signal is useful in that it could allow potential algorithms, based on perceptual analysis, to be developed to improve computer aided detection systems. Such knowledge can also be incorporated into training protocols for medical experts, improving cancer detection or even be used as a possible risk factor in regular screening.
MedicalResearch.com: What recommendations do you have for future research as a result of this study?
Response: My overall objective as that of my co-authors is to make cancer screening better. To that end we are planning a series of different studies, some looking into ways to use this signal and rapid presentation to improve training for radiologist, some looking at possibly using this signal to improve upon computer aided detection systems.
For example one very interesting next experiment would be to take mammograms from women who developed cancer, and look at their mammograms from four or five years ago (e.g. priors), present them to radiologists briefly and see if radiologists can detect the images above chance levels from women that will develop cancer even before the cancer ever appears. We are also interested in exploring whether other medical image experts who use images to make diagnosis such as dermatologists and pathologists can use analogous signals.
MedicalResearch.com: Is there anything else you would like to add?
Response: Radiologists often have ‘hunches’ about images on first glimpse. Our work shows that those hunches are based on something real in the image, possibly on global, widely distributed signal residing in fine detail of the fibrous structures of the tissue that is not the result of either breast density of symmetry. This signal is also visible in the nominally normal breast of a patient that has cancer. However, even though this gist signal has the potential to be useful, as I have mentioned before, it is nowhere near definitive and would never be used to substitute the normal exhaustive reading of mammograms. This work is the results of a transatlantic collaboration and I would like to thank all the radiologists who participated in these studies as observers.
MedicalResearch.com: Thank you for your contribution to the MedicalResearch.com community.
Citation:
Note: Content is Not intended as medical advice. Please consult your health care provider regarding your specific medical condition and questions.
More Medical Research Interviews on MedicalResearch.com
[wysija_form id=”5″]
Last Updated on August 31, 2016 by Marie Benz MD FAAD