Recently, we turned 10 years old. With that, it is fair to say that the team has grown, learned, and evolved. The field of radiology, or medical imaging in general, has seen a big change in the past years with the introduction of artificial intelligence (AI) and, more specifically, deep learning (DL) based methods. Their generalized use in research opened the door for commercial software development to be clinically implemented.
For that reason, we have collected insights from the team to give you a deeper look into what developing AI software in healthcare for 10 years has taught us, its ongoing challenges and what we expect the field to look like in the future.
Primarily, we are glad that the myth about AI replacing radiologists, has been proven to be exactly that, a myth. Radiologists have been able to see through the threat and the unrealistic expectations of AI and are perceiving AI as a tool to support and improve their work even further.
The conversation about AI also surrounds practically every single field, so every stakeholder involved in healthcare is much more in touch with what AI is. So, while AI in the field of radiology was only implemented by early adopters years ago, it is now something with a clear perceived value as imaging workload continues to increase.
AI has also introduced itself into the educational system. This has had an impact on the newer generation of professionals that is now more prepared to work with AI.
Despite the “hype” and attention the latest research innovations get from radiologists, the biggest challenge vendors face when it comes to developing software in healthcare is rapidly innovating in a strongly regulated market.
On the one hand, developing fast, robust, and explainable algorithms for all clinical settings remains a challenge, as it is generally difficult to find large and varied enough datasets to train and test the algorithms.
On the other hand, the regulatory landscape is still developing and, therefore, constantly changing. This fractured landscape can also come with conflicting regulatory requirements and delayed notified bodies which elongate the clearance process and make it harder to establish long-lasting processes to bring these algorithms to physicians.
But the challenges don’t stop there. Implementing a new piece of technology in a clinical setting requires time and financial resources from several stakeholders, which make installation processes longer than desired.
Other main challenge of clinical implementation is the lack of interoperability with the other systems radiologists use in their workflow. However, AI software developers in healthcare now understand that a seamless integration into the physicians’ workflow has an important role in their clinical implementation. So, there is a shift towards focusing more on User Experience (UX) and the ease of access of the software.
In recent years, we have also experienced less reticence towards cloud installations, primarily in Europe. This is positive because cloud installations could be considered easier to implement and they allow for easier customer support versus the traditional onsite installations. We have also noticed that this new outlook regarding cloud installations has increased the awareness of cybersecurity and patient information privacy regulations, such as GDPR, which is something vastly beneficial for the patients.
Quantib has changed a lot throughout the 10 years that it has been operating. Our product portfolio and customer base has expanded, our team has grown and specialized and, lastly, we became part of RadNet, one of the largest radiology organizations in the world.
Although our product portfolio consisted initially of a neuro-focused solution, it has now expanded to different use cases with the development of commercial software focused on supporting prostate cancer diagnosis, and other research tools. When designing and developing our products, we operate with a strong basis in medical imaging research and with a strong focus on regulatory compliance to make sure we can take those products to radiologists as soon as possible.
For that reason, we have created processes that allow us to listen to our customer’s feedback and to be flexible to adapt to new needs and new regulations, for example, we became early adopters of the European MDR 2017/745 regulation. Therefore, we actively focus in creating solutions that fit the existing clinical workflows to aid clinical implementation and truly impact radiologists’ way of working. Beyond the solutions themselves, we work hard to increase the interoperability of our software’s results and outputs to make an even bigger impact into diagnostic pathways as a whole.
With the expansion of our product portfolio and our customer base, we have focused on growing our team in parallel to ensure that we can effectively keep up with the growing demand. That has meant that the team has doubled in size in the last few years, allowing for a wide range of specialized teams and the specialization of each of the roles within them.
However, we have kept the 5 core values Quantib was created upon at the heart of all our operations and they have shaped our company culture.
And with a team that has a clear vision of the impact they want to make in the lives of radiologists and patients, we are confident we can keep succeeding in delivering solutions that can support radiologists’ current and future needs.
Becoming part of RadNet has also given us a very big advantage to enable the accelerated improvement of our products: data. Finding big and varied datasets for the development of AI in medical imaging has always been and continues to be a challenge for every company out there. That is a hurdle that we can now overcome while still having flexibility to bring those improvements to our customers and ultimately the patients.
Currently, there are several clear routes that will shape the future of radiology with one clear outcome: increased workload.
On the one side, early detection in cancer diagnostics is proving its value in increasing the survival rate of patients1, which is triggering multiple discussions about the need for population-screening programs to enable it, for example, when it comes to prostate cancer. The increase of the already high workload for radiologists calls for ways to find higher efficiencies in the workflows and find ways to support inexperienced radiologists to give higher quality reads without having to go through the traditionally extensive learning curve.
On the other side, there are multiple new drugs with ongoing clinical trials and gaining clearance from the FDA, for example regarding Dementia and Alzheimer’s Disease treatments2. The clinical use of these drugs will increase the imaging workload for neuroradiologists as they will require follow-ups to track changes and progression of the disease.
To solve this challenge, we believe there will be a trend in software development to head towards the development of solutions that will cover whole workflows instead of individual tasks within them. And, further along the line, if the regulatory landscape allows it, the trend will most probably materialize in the development and implementation of autonomous AI in radiology.
We also anticipate that the innovations seen in language models, such as ChatGPT, will spark interest in the medical imaging field, thus sparking research in its application to radiology report generation.
So, although radiologists, stuck with the promise of the all-powerful AI, might be disappointed with the impact that it has had in radiology today, we have great expectations of what the contribution of AI to radiology will be in the near future.
Lastly, we believe collaborations are going to play a key role in taking these advances closer to physicians and patients.
On the one side, collaborations between vendors and academia are extremely beneficial as they can prove and accelerate the research of the products’ clinical utility. This can accelerate the approval of regulatory bodies and increase physicians’ eagerness to implement them in the clinical setting.
On the other side, the need of integrated workflows and great interoperability between systems and software will make collaborations between vendors a common practice in the field of AI software in radiology.
This article was possible thanks to Arthur Post-Uiterweer , CEO, Jorrit Glastra, CTO, Floor van Leeuwen, Quality and Regulatory Director, Renske de Boer, Principal R&D Engineer and Product owner of Quantib Brain and Quantib ND, Pim Moeskops, Senior R&D Engineer that has been involved in multiple peer reviewed articles about AI algorithms applications in healthcare, and Almar van Loon, Customer Success Director.
BibliographyAmerican Cancer Society. Cancer figures 2022. (2022).
Biogen. FDA Approves LEQEMBITM (lecanemab-irmb) Under the Accelerated Approval Pathway for the Treatment of Alzheimer’s Disease. https://investors.biogen.com/news-releases/news-release-details/fda-approves-leqembitm-lecanemab-irmb-under-accelerated-approval (2023).