Gnarr, all you Al startup sales people. Stop trying to sell me your COVID-19 AI software!Felix Nensa, radiologist and researcher at University Hospital Essen
I’d rather buy a truckload of toilet paper.
The COVID-19 pandemic has put artificial intelligence (AI) imaging diagnostics at the center stage. Nearly every AI medical imaging startup has recently announced COVID-19 diagnostic product developments despite almost every major radiology body advising against imaging tests as a primary COVID-19 diagnostics method.
Can AI-based diagnostics support medical institutions during the pandemic? Or is this yet another AI industry hype? Oxipit co-founder and chief operating officer Jogundas Armaitis shares the perspective of the company.
Several companies in the chest X-ray AI ecosystem have announced special products for COVID-19 diagnostics. Meanwhile Oxipit has remained silent. Why is that?
In short, a chest X-ray image alone is not enough to manage a suspected COVID-19 patient.
From the onset of the pandemic several bodies representing radiologists and other clinicians have advised that imaging cannot substitute serological or PCR tests when diagnosing the novel coronavirus disease COVID-19. These tests are the only widely accepted and verified method for COVID-19 diagnostics at the moment. In particular, chest X-ray films provide a limited amount of information that is not sufficient to confidently distinguish COVID-19 from other viral or bacterial infections in itself.
Initial disease diagnostics aside, can AI imaging be beneficial over the course of patient treatment?
Absolutely. “Does this patient have COVID-19?” is the question that the media and the public care most about, but the reality of the COVID-19 workflow in a hospital is much richer than that. Medical imaging plays a crucial role in that workflow as advised by, for example, The British Society of Thoracic Imaging.
Furthermore, Oxipit ChestEye can identify 75 radiological findings in every chest X-ray film, including findings that may (or may not) be a result of COVID-19 infection. These findings are well established and very important, no matter if the patient has the novel coronavirus disease or not.
In general, AI imaging software can identify lung abnormalities and that way increase radiologist productivity. In addition, as overworked clinicians focus on COVID-19, secondary findings (such as tiny lung cancer nodules) may be overlooked. Software does not have this bias.
Would you regard ‘AI COVID-19 Diagnostics’ a play of words or a marketing ploy rather than genuine innovation?
I would rather shape this question in terms of current consensus in academia.
Can chest X-ray imaging by itself be used as a reliable tool for COVID-19 detection with high specificity? The scientific consensus is no.
Can medical imaging software based on AI be employed over the course of COVID-19 patient treatment? Yes, absolutely, as long as it is safe and reliable. This is the case not only with COVID-19 but with a plethora of other conditions as well.
What would you advise to medical institutions who are currently researching AI medical imaging tools?
I would advise to take the time to make the right decision, even if there is pressure to act quickly in the face of the pandemic or the post-lockdown labor shortage.
The medical imaging team should be front and center in the process. What will be the role of the AI product in the organization? What problems will it solve and what particular productivity gains are expected? Does the intended use of the product answer the problem in question (this point was recently highlighted in an excellent blogpost by Hugh Harvey)? The match between problems experienced by the radiology team and the solution in question should be crystal clear.
Ease of use is crucial, too. Will the new piece of software present the radiologist with additional windows and panels? Or will it integrate with your PACS and RIS? AI should be properly integrated into the radiology workflow as emphasized by, for example, EuSoMII president Erik Ranschaert in a recent webinar.
Quality and accuracy of AI diagnostics is what the organization will experience in the daily routine after the software is up and running. To avoid surprises on that front, we strongly advise organizations to evaluate software performance on their own data before making the final decision.
The COVID-19 pandemic put the AI medical imaging community in the limelight. What effect does this exposure have for the industry as a whole?
The route of AI medical imaging is no different from other emerging technologies. It follows the trajectory along the Gartner Hype Cycle. After the initial innovation the industry received a lot of buzz (including scaremongering of how it will replace human radiologists) and probably has already passed the peak of inflated expectations. The COVID-19 pandemic presents another expectation bump. Thus it is common for emerging technologies to be initially oversold and to then underdeliver.
As for Oxipit, we are in our third year of development and operations. This period of time has allowed us to gather a lot of valuable feedback from radiologists using our products, as well as to engage with the broader community of scientists and other organizations. Our product has significantly improved over this period of time, and the benefits that our software can provide are clear by now.
The industry as a whole is reaching maturity. The good thing about the hype cycle is that at some point the hype ends. Quite often new technologies result in products that contribute to the betterment of our reality.
Oxipit is a computer vision software startup specialized in medical imaging. With a team of award-winning data scientists and medical doctors, the company aims to introduce innovative Artificial Intelligence/Deep Learning breakthroughs to everyday clinical practice. Oxipit is the authors of CE certified multi-award winning ChestEye radiology imaging suite.