Apr 2, 2021 — 4 min read
As a medical AI company, we value in building clinical evidence. We have published lots of studies featuring the performance of our AI solutions in major peer-reviewed journals.
We have also been closely working with physicians and researchers around the world to conduct trials on the clinical outcomes of AI. These trials aim to assess the benefits of using it in real-world clinical settings and to gain insight into how it could be best adopted into routine practice.
Dr. Fredrik Strand, MD PhD, from the Karolinska University Hospital in Sweden, is one of the leading researchers who has demonstrated the potential value of AI in breast cancer screening.
In his recent retrospective study published in JAMA Oncology, he demonstrated that AI algorithms differed in performance to detect breast cancers on mammograms and that Lunit algorithm outperformed the others. He also showed that Lunit AI when combined with the first-reader radiologist detected more breast cancers in the double-reading environment.
Interview with Dr. Fredrik Strand, MD PhD, from the Karolinska University Hospital in Sweden
To evaluate the sufficient diagnostic performance of Lunit AI solution as an independent reader, Dr. Strand has set out a prospective clinical trial in a true screening population.
Below is an email interview with him that will give you a glimpse into his prospective research.
We’d like to hear about your prospective clinical trial that starts in March of 2021.
Dr. Strand: The program is called “ScreenTrust” and consists of two studies: “ScreenTrust CAD” and “ScreenTrust MRI”.
The first one, ScreenTrust CAD, is a clinical trial of AI as an independent reader of screening mammograms, aiming to examine to what extent AI can replace or assist radiologists in screening.
The second one, ScreenTrust MRI, is a clinical trial of AI as selection mechanism to invite women for supplemental MRI after a normal screening mammogram, aiming to examine to what extent AI together with MRI can help identify cancer that did not show up on mammography.
Is it going to be the first prospective study that evaluates the performance of an AI algorithm combined with radiologists in a true screening population?
Dr. Strand: As far as I know, both trials are the first clinical trials within full population-based study populations. However, I would expect that similar ones are being planned since AI seems to hold so much potential in this field.
What is the main difference between your research and other AI research?
Dr. Strand: We have a multidisciplinary group consisting of breast radiologists, biostatisticians and computer scientists. We combine networks that were academically developed with networks that were commercially developed.
The Swedish screening program is well-established and it is easy to follow the participants over time, thanks to regional and national registries. Our approach to clinical trials have a clear practical focus, since I am working part-time as a breast radiologist and part-time as a researcher.
Did you have any difficulty planning and preparing for the trial?
Dr. Strand: It took slightly longer than expected to get the two hospitals onboard, but everyone was very enthusiastic from the start.
Do you expect to get the results you are looking for? Are you confident in the performance of AI?
Dr. Strand: We have our beliefs based on retrospective studies, but in a prospective clinical setting, there is also important interaction between AI and humans that does not show up in retrospective studies.
I am confident that AI can play an important role, as there are now a large number of retrospective studies supporting that. However, I also think that the studies will show that there are still opportunities to improve the role of AI and how to combine AI with human assessments.
When are we going to be able to see your interim results?
Dr. Strand: It will probably take 1,5 to 2 years from the start (the CAD project expected to be slightly quicker than the MRI project), so that would mean sometime during 2023.
How would the results of your study be applied to the European breast screening environment?
Dr. Strand: We have to wait and analyse the results of course. If AI proves to work as expected, we could see a change from double-reading by human readers to the first combination of AI and one human reader, and later perhaps some cases can be read by AI alone.