the birth of a SPINOFF: Subtle Medical
part 5: spotlight on a spinoff from Stanford
DECEMBER 2020
by Giammarco Pacifico, Marta Gaia Zanchi
This is Part 5 of a short series of posts in which we explain how turning the next game-changing technology from a lab into a high-impact product available to men and women everywhere has always been the result of a mix of domain knowledge, intellectual curiosity, gut feeling, and being in the right place at the right time, equipped with the right mentors and role models.
Subtle Medical was born at Stanford University in 2017, thanks to the combined efforts of Professor Greg Zaharchuk and Enhao Gong, who at the time was completing his PhD Program in electrical engineering with a research focus on deep learning application in medical imaging.
Today, Enhao is at the helm of the startup, having raised over $20 million in funding to pursue its mission to enhance the quality of radiological images worldwide. In just three years, Subtle Medical has risen as a healthcare technology company with a suite of deep learning solutions that improve workflow efficiency and patient experience. The company has the first AI software solutions FDA-cleared for medical imaging enhancement, SubtlePET™ and SubtleMR™ for Positron Emission Tomography (PET) and Magnetic Resonance (MR) respectively. It counts clinical partners the likes of University of California San Francisco, University of California San Diego, Radnet, Tiantan Hospital, and Middlesex Hospital, among others.
The company has solved a challenge as old as the medical imaging field itself, that to increase the speed of image acquisition and reduce the dose of contrast in MR and PET without trade off in image quality, in order to increase throughput and comfort of patients during image acquisition. Notably, Subtle Medical’s impact on patients is direct and is immediate. Its acquisition software has been demonstrated to reduce the required time for an exam by up to 75% with no detriment to the clinical utility of the images. Importantly, images acquired with Subtle Medical’s solutions can still be fed to and processed by all downstream traditional analytics and diagnostic software available in the market
Enhao’s interest in imaging compression and reconstruction began very early in his academic career. He had been working on methods to optimize MR imaging since at least 2014, using more conventional techniques. His first peer-reviewed articles, published in 2015, led to a close collaboration with Professor Greg Zaharchuk. Their alignment in mission was evident from the start, as Greg recalls vividly in his interview: “ Enhao and I started working on some of the projects that would lead to Subtle Medical in early 2016. […] Once we did our first experiments with deep learning for image quality improvement, we realized that there were many potential applications to a technology that could use AI to predict images from other images. Also, as a clinician, I felt that this was a rare opportunity to develop a technology that could benefit patients, hospitals, imaging centers, and the healthcare system as a whole.”
Before Subtle Medical was both, the founders-to-be focused on tackling the high technological challenges of their innovative solution, as Enhao highlights: “When we are working on a building deep learning model for medical image reconstruction, one of the major issues is the generalizability and extension from computer vision algorithms on single photo data to medical image datasets that can be a 3D volume with various resolution and multiple contrasts and even complex values that are not the same as in natural RGB format. Many times the backbone model directly available in computer vision is not the best.” When they started, there were limited reference works and resources to work on this challenge, which forced them to creatively approach the challenge from entirely new perspectives. “We did different exploration and eventually invented some ways to not only better handle the medical image dataset, but also incorporate imaging physics models into the deep learning framework so that the algorithm can solve the problem with better accuracy and generalizability.”
Their first results were extremely encouraging, showing comparable quality of PET image acquisition with state-of-the-art solutions but reducing the standard dose by 50%, which opened the way to high-potential real-world applications. Further research led to the creation of a convolutional neural network to predict the occurrence of infarct from acute image scans, which is particularly relevant in triage and severity prioritization. Moreover, the development of compressed sensing analytics for MRI meant that finer texture details could be shown in a much shorter time delay, drastically reducing the time needed for a single exam.
But as we have learned during our previous interviews with spinoff founders, having a game-changing technology is only the first step towards solving a complex healthcare need. “While I personally did not have experience with starting a company, the great resources and supportive environment in and around Stanford was certainly a factor in the decision to take this leap,” Greg told us. Access to supportive faculty and in turn, their relevant networks helped them not only to access capital but also enter the market and fuel their international ambitions.
Despite such a supportive ecosystem, well known for a prolific generation of startup companies, the choice to spinoff was not obvious. It was inspired by a vision for a breakthrough in the status quo, and their strong partnership. “In the past, Stanford researchers usually licensed technology to big players in medical imaging. However, since we believe we should try a new framework, we decided to do it ourselves.” says Enhao, adding: “there are several moments that inspired us to start. One of the most important moments was when we found out that we shared the same idea of building the startup by ourselves to commercialize the technology. It is critical and valuable to find a partner to work together on this who shares the same passion.”
“It was clear that Enhao and I shared the vision that research needs to get beyond the walls of academia if it is really going to help people” points out Greg. “He was also very good at supporting other researchers, both at Stanford and in other institutions, to leverage his expertise to demonstrate how widely the technology could be used. This suggested to me that he could manage a team and would be a good leader.”.
With a compelling problem, an innovative technology, and strong partnership as its foundations, Subtle Medical was born — but the learning continued well after leaving the academic environment. Asked about what he wishes he knew before embarking on the process of spinning a company off Stanford University, Enhao points to the critical difference between technology and product. “For medical imaging products, the AI technology is at the core, but still just a part of the entire solution. At Subtle Medical, we have been investing a lot more on everything around it to deliver the best product.” From a smooth user experience to security audits, delivering value requires taking all perspectives into the product design experience. “For example, we know how to build the user experience so there is zero disruption to workflow except making it much more efficient. Another example is that we have also invested a lot of time and resources to get several third party audits like ISO, HIPAA etc. since in healthcare IT security is very important and getting it right and audited helps us to facilitate later conversations with hospitals and imaging centers. Last, but not least, we build a great customer success team with people heavily experienced in clinical imaging workflow and with experience working with radiologists, technologists, and administrators in the radiology practice.” It is critical not only to deliver the right technology but also “the solution in the best format, understand what customers need, and communicate effectively with them,” concludes Enhao.
Only the first chapters of Subtle Medical’s story have been written and yet the company is already having a direct impact on healthcare systems in three continents with its first two products. Thanks to a continuous effort on research and development, most certainly inspired by their academic roots, many more exciting improvements are coming.
In summary, here are the lessons in spinoff creation that Subtle Medical shares with us:
- Starting a project inside the academic environment offers a chance to test the partnerships of individuals in the context of a research collaboration, before they become funding teams in a startup. (Disharmony between founders is one of the top reasons startups fail.)
- Alignment of all stakeholders’ benefits is key. Everyone’s incentives and constraints should be the inputs of the process to design a successful health technology and business.
- When selling to hospitals, understand deeply the work of physicians, nurses, and all others involved in the provision of patient care: the way in which a product changes workflow can make or break its adoption.
- Great technologies do not always make for great products. Bridging the gap between the two lies in the understanding of all stakeholders and clinical workflows.
This post concludes our spinoff series for 2020. Look out for more posts on this topic, next year!
