With technological evolution moving at a rapid pace, the ubiquity of artificial intelligence (AI) in healthcare and radiology is an inevitability. But what does that mean for hospitals and clinicians in relation to their workflow? We spoke with Dr. Philippe Grenier, a Consultant Chest Radiologist at Foch Hospital in Suresnes, France, about the role of AI in clinical workflow and how the development of different types of tools and software are helping radiologists to improve productivity and efficiency.
How are AI tools changing and affecting clinical workflow in hospitals? What are the classical areas of healthcare in which AI tools can be helpful?
Actually, some AI Tools already shorten the interpretation time by radiologists for several radiological examinations, particularly plain radiographs and CT scans. Some examples may be cited, such as automatic detection of pulmonary emboli on CT scans, automatic detection and volumetric measurement of pulmonary nodules that may be suspected to be malignant. Another advantage of AI tools is the ability to detect abnormalities that are not a-priori expected to be present according to the clinician’s demand, and in the same time to perform quantitative measurements of automatically segmented anatomic structures or detected lesions. This aspect is particularly well illustrated by the situation of a chest CT scan performed in a 50-year-old smoker to screen for lung cancer. A specific software, AI Rad companion, can not only detect, segment, and measure suspected pulmonary nodules, but can also in the same time (1) detect and quantify emphysema, a marker of increased probability of nodule malignancy, (2) automatically measure the thoracic diameters of aorta at different levels to detect any early aneurysmal dilatation, and (3) segment and quantify calcifications in coronary arteries, a reliable marker of cardiovascular event prediction. It can also measure the density and height of vertebral bodies.
Another example of a significant change in clinical workflow is the interpretation of plain chest radiographs by an AI-based solution that provides efficient triage between normal and abnormal films and make a certain number of diagnoses. Such tools can simplify and shorten the radiologists’ interpretation. However, on the opposite end, more complex AI tools are developed to predict the outcome, prognosis, probability of malignancy of a detected lesion, or to predict treatment response to this lesion. This provides an improvement of the quality of the medical decision, particularly for optimized treatment, but does not affect clinical workflow in hospitals.
Where do you see AI helping with the exponential growth of radiological exams that we are seeing now (e.g. assisting the doctor with medical imaging; offering another level of gatekeeping between the clinicians requesting the exams and the radiologists; double-checking referral guidelines to see if the recommended exam is appropriate)?
The development of AI-based algorithms to control and validate the clinicians’ requests for imaging examinations according to the referral guidelines should be widely appreciated. Such tools already exist in hospitals to analyze clinicians’ prescriptions in order to facilitate the pharmacists’ validation.
What concerns do you think patients may have about their doctors using, and being assisted by, AI?
Patients are not always aware that their doctors are being assisted by AI. However, when they come to the hospital, they have to sign an agreement form in which it is clearly specified that their medical data will be digitized and stored with potential opportunities to be used for information processing that may include AI.
Is there a resistance by hospitals and their employees to implement AI tools or systems?
Resistance to AI implementation in hospitals exists, but the best way to delete this resistance is to validate the proof of concept. In other words, the high value of confidence index of AI algorithms performances have to be demonstrated before integration into the workflow. For instance, the demonstration of a very high predictive negative value of AI-based chest radiographs interpretations is necessary to make emergency physicians and radiologists confident enough to use such AI tools.
Are there financial barriers to implementing AI systems?
Of course there are financial barriers. The business models of the different AI solutions are various, including usage cost, licence acquisition, or annual licence renting. For the moment, the cost is quite limited as most of the AI tools are provided by small startups – this could be different when global solutions are supplied by big companies. At this point in time, implementation of AI global solutions should necessitate improvement of hospital productivity and of reduction of expenses, including human resources, to ensure financial balance.
Dr. Philippe Grenier received his medical degree from the school of medicine at the University of Paris in 1972. He then completed a residency in diagnostic radiology at the Assistance Publique – Hôpitaux de Paris and a fellowship in the Department of Radiology of the Hôpital Beaujon, Faculté de Médecine Xavier Bichat. Upon completion of his fellowship in 1982, Grenier remained at Faculté de Médecine Xavier Bichat as an associate professor of radiology. From there, he went on to the Faculté de Médecine de Bobigny as professor of radiology in 1988. The following year he accepted his present position as professor of radiology with the Faculté de Médecine Pitié-Salpêtrière, Université Pierre et Marie Curie, where he served as vice-president of the university from 1998 to 2001. Since 1989, he has been chairman of the Department of Diagnostic Radiology at the Hôpital Pitié-Salpêtrière in Paris. Grenier currently works as a consultant at Hôpital Foch in Suresnes, France.
Siemens Healthineers: Gain insights about the digital transformation of radiology and how data-driven solutions like AI-powered technologies are helping provide better outcomes and improve patient care.