Research is key for innovation and development of state-of-the-art solutions to improve healthcare and patient care. Most ground-breaking research is done in academic institutions, where new and often great AI-algorithms are developed. However, a strikingly small percentage of these algorithms actually make it to the clinic, meaning that a lot of potential to save and improve the lives of patients is not used.
I believe this gap between research and practical applications can partly be solved by a stronger collaboration between academia and industry, sometimes called public-private partnerships. With industry, we here focus on companies or vendors that develop AI-software solutions within the healthcare space.
In this article, I would like to show that, when done correctly, academic-industry collaborations can bring a lot of positive aspects to both academia (or physician researchers) and industry.
A lot of intense research is needed to develop a new AI algorithm in healthcare. The process starts with a great idea, and a design of how AI algorithms could be implemented in this use-case. Many research institutions are, or have connections with, clinical institutions can that provide the data to train and test these algorithms on. In the best case scenario these algorithms are furthermore tested on data from other institutions or public datasets. But too often the process ends here.
Sadly, it is hard to get these algorithms into the hands of physicians that work in clinical practices. One of the reasons why this happens is that, in order to be used in a clinical setting, these tools need to be easily implemented in the clinical routine. This means that it needs an interaction with the hospital PACS system, it needs to be user friendly, and, in the case of medical imaging, it needs things like viewers or even interaction-steps. Not to mention the very important follow-up step of clinical certification. Creating such an environment that is unrelated to the algorithm needs time and resources and are often either not included in the initial project or are unfamiliar to the algorithm researchers.
To solve these hurdles, industry collaborations might be a valuable addition. Many of the companies that develop AI solutions, in their commitment to improve healthcare, are open to collaborate and assist research to further improve their products or to develop other solutions that may extend beyond their specific use-case.
AI healthcare companies are in certain situations willing to provide for example, a clinically cleared platform for implementation of the AI-algorithms, or their expertise in how to create the necessary framework for a clinical prototype of the AI-algorithm. This can accelerate the time between algorithm development and actual clinical implementation tests. A good public-private partnership, for example within the context of a grant consortium, can create projects that draw upon the combined efforts of expertise and experience leading to rapid results and effects on healthcare. Within a proper collaboration, clear agreements will allow the protection of intellectual property.
Research is not, and should never, be finished once a product has passed the validation steps and certification that enables it to be sold and used in a clinical setting. This is the moment another very important part of research comes into play; the clinical real-world validation. Strong scientific evaluations are necessary to discover real patient impact. These include questions about how well the software performs in all situations in order to avoid bias, for example.
Furthermore, it is also important to evaluate the impact on the physicians, for the precise AI-use case within the disease process, but also in the larger care-pathway. The former will allow an iterative process of product development, and the latter is important since the value of the products can increase depending on their impact on the larger care-pathway. For example, if a faster or more detailed diagnosis has no effect on later treatment, the clinical utility might be limited.
Researchers, especially those working with patients, have an incredible understanding of what is useful and necessary for real clinical impact on patient care. They are well suited to perform such studies independently and objectively. It is thus without question that academic-industry collaborations can be incredibly important for companies.
As previously mentioned, many companies will also be interested in research that is not necessarily directly related to their commercial products. Some companies might have research-only software available, this can include software with a broad use-case that makes FDA clearance/CE marking difficult, or software that is in the process of being cleared by regulatory bodies. Their use, their research interest, and the feedback from these early adopters can tell a company a lot about the utility of these software products and their possible clinical impact.
Of course, research doesn’t have to be done solely by researchers in academia. Frequently, it is the physician that regularily see patients the one that will have an interesting hunch or idea. For example, that the often observed assessment X within a specific disease has a link to patient outcome. AI-healthcare products might be a useful tool to answer these questions, and a company might very likely be interested in exploring that relationship together.
It can also be the physicians who have access to a large database of patient data. This data is often acquired on machines that might not be as high quality as those in academic institutions, or include missing data as patients get transferred between centers. In short, the real-world situations. As healthcare data is protected and messy, it is often in these non-academic settings that real-world clinical utility of AI products are best tested since AI software in healthcare should still work optimally in these situations.
Ideally, industry would like to have such real-world validation studies published in peer-reviewed journals. But this certainly does not have to be the case in all situations.
Many companies will have a team of clinical scientists that often are more than willing to help a researcher or a physician set up a strong research project. These teams will know how to create a good protocol and can aid in defining the exact research questions and statistical methods of analysis. Furthermore, depending on your needs, the collaboration and the pre-defined task distribution, they might be willing to collaborate on other parts of the research project. If you have a good idea, or feel like your upcoming grant could benefit from an academic-industry partnership, the best thing to do is to contact a company and explore a potential collaboration. But how?
First thing to do is to find a company that shares your vision, and with whom a collaboration might be beneficial for your research questions. For example, depending on the research question(s) you have, an AI company working mainly on MRI and medical imaging, might or might not be the best partner.
First contact will often lead to discussions and explorations of mutual interest. If this is present, further discussions will be very similar to those between different research institutions that aim to collaborate, or to grant preparations.
My personal tips to make this AI research project successful are:
In my experience, making sure this is well thought through and written down will facilitate the collaboration going forward.
During these moments of contact, raise any issues you might have. For example, one might ask themselves how such a collaboration could impact academic freedom, or if they would still be able to publish if results are not necessarily beneficial for the company. Do steer away from collaborations that might indicate that publication is dependent on results! From my own experience, benefits will outweigh any challenges that may arise in these discussions. Both the researchers and any good AI company in healthcare will be driven by the positive impact research can have on patient care, and would like to improve patients’ lives as quickly and safely as possible.
Still not sure if you, as a single researcher or small group should contact and set up collaboration with a larger company? Check with your university or institution, many can offer advice and guidance within their knowledge transfer teams.
Within Quantib, we are always interested in collaborating on research projects that have a large potential for patient-impact. Our clinical research endeavors take several forms.
Real-world clinical validation of our products is one of the most important axes of clinical research. We aim to gather strong scientific evaluations of our products, to assess if our products work as expected in all situations, and how they improve patients’ lives. Other kind of studies may focus more on the impact on the users (e.g., radiologists), or the effect in the clinical workflow and larger care-pathway. Collaborative scientific studies might also relate to wider research questions, to evaluate if, and how, our software can establish new knowledge about this or other diseases and conditions.
From time to time, we also take part in larger research consortia. We have done so in the past and are actively working on grant proposals for current and future research. Researchers and clinicians can also use other Quantib software that it is dedicated solely to research, like our Body Composition software. This software segments abdominal fat on CT images at the level of L3 and can lead to various interesting insights in many clinical situations.
Personally, I have experience on both sides of such academic-industry partnerships. While in the beginning of my career such collaborations might have seemed scary, over the years I’ve learned that the benefits vastly outweigh the possible hurdles.
A successful implementation of healthcare products in clinical use requires well founded research to prove it’s value. Strong partnerships between companies and researchers can facilitate the clinical implementation of such products as well as evaluate it’s validity and performance. Together we can reach improvements in healthcare faster.