New study questions patient understanding on AI in radiology

A new study says patients need to be educated more on the precise use of AI in the scanning process

April 11, 2019
by John R. Fischer , Staff Reporter
A new study says patients vary in their understanding of the different roles that make up a radiology department, and are skeptical of and lack knowledge on the capabilities offered by artificial intelligence.

Conducted by Dutch researchers, the qualitative assessment argues that more education is necessary to help patients understand how AI is used in the scanning process, and where the roles of different staff members in radiology stand in relation to its use. Once attained, this knowledge will enable them to be better able to contribute input for determining best practices and use of AI and machine learning in these types of settings.

“Since patients' level of knowledge on both artificial intelligence and radiology is generally low, the combination generates mixed feelings among patients,” Dr. Marieke Haan, assistant professor in the department of Sociology at University of Groningen, told HCB News. “In all of the six domains we formed based on our findings, patients expressed concerns but also showed their beliefs in a promising future.”

The study refers to patients as “important but neglected” stakeholders who are “crucial” to determining the development and uses of AI systems for various clinical tasks in routine radiology practices.

Surveying 20 individual members from a group of 11 men and nine women at the department of radiology of the University Medical Center Groningen, the research team broke their responses down into six sections of information that sum up patient needs, concerns and viewpoints:

• Proof of technology: Patients desire scientific evidence to validate the use of an AI system in radiology, before it is actually used. When presented with findings that showed AI as equivalent in skill to humans, patients preferred to have a human perform their exam. Machines, however, were preferred when research showed computers to be superior.

• Procedural knowledge: Patients want to know how AI is exactly used, and want to receive incidental and unrequested findings in addition to those that are based on questions of the referring physicians. They are unclear about how AI affects scanning procedures and the delivery of findings, as well as unaware of who is involved in exams and how roles such as radiologist, technician and referring physician differ from one another in relation to the use of AI.

• Competence: Patients are skeptical of AI and believe using it only could lead to restricted views with wrong diagnoses. They are unsure about the skills of computers, and believe they should be used as secondary sources to validate the conclusions of radiologists.

• Efficiency: Efficiency was assessed based on the duration of scans, with patients expecting AI to shorten the amount of time necessary for full assessments, enabling them to be assisted sooner and at lower costs.

• Personal Interaction: Patients unambiguously expressed the desire for personal interactions with doctors when receiving exam results, as it enables them to safely ask questions and allows them to understand their results and their reliability. They believe human dialogue is important in such matters.

• Accountability: Patients question who is responsible for errors made by computers, with some saying humans will always be responsible because computers are just “giant calculators” or “dead things”.

Despite their skepticism and lack of knowledge on the use of AI, most feel that it is a development that will eventually take root in healthcare and other industries, though not in the near future.

The findings illustrate the need for clear communication among patients, referring physicians and radiologists on how AI is implemented in diagnostic radiology procedures, including scan acquisition, scan evaluation and the sharing of results. Haan, however, cautions that more quantitative analyses are required to validate the findings of this study.

“We are in the process of writing another article on this topic in which we aim at validating a patient survey on AI in radiology. Before developing and implementing an AI system for a particular radiological task, it would be very useful to perform a patient survey, since patient preferences determine the boundaries within which an AI system should function. The aim of our new study is therefore to develop and validate a standardized patient questionnaire on the use of AI in radiology using the six domains of our qualitative study. The data are already collected among patients and we are now analyzing them.”

The findings were published in the Journal of the American College of Radiology.

More Artificial Intelligence

Aidoc announces $27 million in VC funding to advance AI in imaging

Fredrik Palm

ACR engages in collaborations for AI development with launch of AI-LAB platform

New AI software identifies make and model of cardiac implants in seconds

Dicom Systems scores enterprise imaging contract with Radiology Partners

Apple study suggests wearable technology may be useful in detecting atrial fibrillation

Nvidia unveils Clara AI platform at GPU Technology Conference

BSWH to install Glassbeam's CLEAN blueprint to leverage machine uptime

Beyond the hype: How practical AI is enhancing radiology

Machine learning reduces false positives for lung cancer in low-dose CT