Rapid advances in technology and 3D imaging have brought new challenges when interpreting medical images, with two academics from the University of Cumbria and their collaborators recently recognized for their crucial research contributions in the understanding of the perception of medical images.
Associate Professor Tim Donovan and Dr Peter PhillipsThe work to help establish better image interpretation has helped inform global guidance from the World Endoscopy Organization and the American Association of Physicists in Medicine.
The two researchers studied eye tracking for medical imaging on 2D and 3D datasets to improve our understanding of how practitioners perform visual searches and make decisions, as well as the occurrence and frequency errors. Rapid advances in imaging technology and its increased use compel us to learn more about its role in decision-making in clinical practice.
The results of this research have also directly benefited radiography students at the University of Cumbria, who are learning vital new skills in interpreting imaging data.
But what exactly is eye tracking and how has it made a difference in the frequency of x-ray errors that has remained unchanged for 60 years?
Dr. Donovan, who works in medical image perception and cognition at the university’s Institute of Health, said the procedure measured eye movements with an infrared illuminator and a high-resolution camera. , which allowed a better understanding of radiological performance.
He said: “We constantly move our eyes, up to five times per second, to scan the world around us and focus our attention on the most relevant part of a scene. Eye tracking tells us where people are looking, what information is being processed and for how long.
“It provides invaluable insight into visual and cognitive processing and, as a research tool, has helped understand why radiologists make mistakes. These are usually at the rate of three to five percent, but can be as high as 30 percent when examining cross-sectional scans such as MRI and CT scans.
Medical images such as X-rays have always had a degree of ambiguity and this key research examined typical search patterns with the aim of understanding and reducing inaccuracies and potentially improving professional training.
The assumption that a radiologist should be able to “look at” an image and make an informed decision based on instinct and experience has been challenged by this study. It showed where radiologists and radiographers had actually looked at the image and generated data that could improve diagnoses and reduce errors.
While there are fears that the use of artificial intelligence will displace radiologists, the counter-argument is that computer-aided detection will actually enhance their role.
“These are small incremental advances in our understanding of how visual expertise is applied to radiology and have become even more relevant now that computer algorithms are used in many tasks,” Dr Donovan explained.
“However, people will still need to be involved when decisions are made about medical diagnoses.”
Improved accuracy of radiological interpretation means a higher percentage of conditions can be spotted early. An important part of the research has been the imaging diagnosis of colorectal cancer and the detection of polyps.
Although revealed by CT scans, polyps are often overlooked because they may be unclear or difficult to spot. However, it was this type of decision-making that needed improvement, the researchers concluded.
Dr. Peter Phillips is a lecturer in medical imaging. He is currently working with Dr Yan Chen, Associate Professor of Cancer Screening at the University of Nottingham, while Dr Donovan continues his collaboration with Dr Damien Litchfield, Edge Hill University, on components of visual search .
Accustomed to renowned research, Dr. Donovan, in collaboration with Professor Vincent Reid, of the University of Waikato, New Zealand, and Dr. Kirsty Dunn, of Lancaster University, undertook research at the laboratory of ultrasound at Blackpool Victoria Hospital and University, examining the visual perception of babies in the womb – listed as one of the most important findings of 2017.
The research used a light source to project a pattern of three dots in the shape of eyes and mouths through the uterine wall. By measuring how the fetus responded using a 4D ultrasound, it was found that at 34 weeks they resembled newborns in preferring face-like stimuli.
Discovery Magazine included the project in the list of 100 discoveries made that year.
The image below is the CTC image of the scanning path of three trainee radiologists and an expert looking for bowel cancer from a virtual colon flyby created from the data of CT colonography. Everyone spots the cancer in the lower left, but only the expert looks at a cancer in the upper right of the image.
The image below shows the scan path of a radiologist looking for fractures. The red cross is where they think there is a fracture, but it’s not. They actually missed the fracture in the blue circle. In this case, the error is a research error because they did not observe the fracture.