Founder’s Lessons: Dr. Enhao Gong, CEO of Subtle Medical
Enhao shares his journey from Tsinghua, his past startup experiences, his vision for medical imaging, and his best advice
About
Welcome to the sixteenth installment of Tsingyuan Ventures’ Lessons from Founders series. Every week, we publish an in-depth founder interview, ranging from early-stage entrepreneurs to successful businesses. Our conversations cover their personal journeys, the lessons that shaped them, their visions for the future, and their failures. We also learn more about their companies and about the challenges they try to solve. These insights and lessons are applicable to any entrepreneur — current or future.
Read past interviews here.
Subtle Medical
Update: Subtle Medical recently raised a significant $12.2 million Series A round of funding, with participation from Tsingyuan Ventures. You can read the 11/19 press release here.
Subtle Medical is a healthcare technology company founded out of Stanford University by practicing Neuroradiologist, Professor Greg Zaharchuk, MD, PhD, and Enhao Gong, PhD. The mission of the company is to make medical imaging faster and safer with their deep learning powered software solutions.Faster imaging allows centers to increase revenue, while offering a shorter and more comfortable scanning experience to patients.
Subtle Medical has successfully launched two FDA-cleared & CE Marked products, SubtlePET™ and SubtleMR™.SubtlePET allows hospitals and imaging centers to denoise images collected in 25% of the original scan time while SubtleMR uses denoising and resolution enhancement to achieve higher quality MRI scans for conventional and accelerated protocols. Subtle Medical also recently received a NIH SBIR Grant for $1.6 million in order to continue research on SubtleGAD™, their third software product that is aimed at 10x contrast dose reduction, enabling safer contrast-enhanced MRI
Subtle was named among the prestigious CB Insights AI 100 List of Most Innovative Artificial Intelligence Startups and CB Insights Digital Health 150 in 2020. Among other awards, it received the NVIDIA Inception Award for Top Healthcare+AI Startup Globally selected from over 3,000 contenders in 2018.
Subtle’s AI software solutions are being used in dozens of hospitals and imaging centers in the US and abroad. Current clinical partners include UCSF, Hoag Hospital, UCSD, MD Anderson, Affidea in EU, DASA in Brazil, Tiantan Hospital in China, and more. For more information, visit subtlemedical.com.
Dr. Enhao Gong is the Founder and CEO of Subtle Medical. He is a serial entrepreneur with a PhD in Electrical Engineering from Stanford University. His passion and research focuses on applying AI and deep learning to improve the reconstruction, analysis and quantification of medical imaging. His work has won several awards including RSNA research award, Forbes 30-under-30 (China and APAC) and has been featured in numerous academic journals and clinical conferences.
Why we invested in Subtle Medical: The founders, Enhao and Greg, have combined decades of research in radiology and deep learning together in a powerful image reconstruction system. This system dramatically reduces the time, cost, and radiation harm previously thought as necessary in medical imaging. The uniqueness of their approach combined with their head-start gives them a huge first-mover advantage. The multiple FDA approval and commercial clinical usage subsequently validated our initial thesis.
Meet Dr. Enhao Gong
Interview edited for clarity and length.
“You start with a specific technology endpoint and then you have to build it into a product.”
Enhao introduces himself
I’m Dr. Enhao Gong, CEO of Subtle Medical. I co-founded the company around the end of 2017, and Tsingyuan offered us one of the first checks helping us to start from zero. I graduated from Tsinghua University in 2012. My major was biomedical engineering, and I was working on medical imaging technologies ever since undergrad research. I came to Stanford for my master’s and PhD in electrical engineering, with a focus on medical imaging. The lab I worked in, MRSRL and RSL, focused on MRI acquisition, reconstruction, medical image processing, and related applications.
Medical imaging is fundamental to clinical diagnosis. For decades, people in this field have been working on improving quality, improving efficiency, and providing better care through medical imaging exams. That’s the technology we are developing to accelerate this process.
Medical imaging is fundamental to clinical diagnosis.
Enhao’s background
In 2013 and 2014 summer, I had a little deviation from the traditional PhD journey. I co-founded Polarr, an image processing photography start-up, with my friends at Stanford. Polarr was working on building an online tool for photo retouching with tens of millions of users. We developed AI to improve photograph editing and curation, which later also converted into well-performing B2B applications that power the latest Oppo and Samsung phones.
Photo processing is related technically to my research of image processing, but is not in the same area as medical imaging, which I was primarily working on at Stanford. Around 2015, I decided to continue and finish pursuing my PhD.
While working on medical imaging, my focus was on improving imaging technologies with algorithms. The usual approach is with conventional algorithms, which are very computationally intense and use a lot of iterations to search for solutions using mathematical models. After I developed several algorithms to improve the iterative optimization methods, I tried re-approaching it as a prediction problem, using deep learning (which had just started to boom around 2014 and 2015.) I trained deep learning models to enhance image quality and found that it worked very well — way better than conventional algorithms.
When people talk about using AI to improve radiology, the first thought they have is to use AI to replace radiologists. But that’s not what I was primarily working on at Stanford and Subtle Medical. We’re working on augmenting technology by replacing the conventional algorithms currently used inside a scanner. For a new MR or PET scanner, the R&D-to-product time takes years or decades. One of the earlier generations of fast imaging technology, Compressed Sensing in MR, was invented in MRSRL at Stanford and the first paper can date back to 2007. But it took until 2017 for all the bigger vendors, like GE and Siemens, to get an FDA-cleared version of that algorithm and start to promote it as a clinical routine technology. Medical imaging and the healthcare industry is very conservative and slow-paced, whereas AI is rapidly changing every week or day. It needs a new framework to convert into clinical products and commercialize.
We’re working on augmenting technology by replacing the conventional algorithms currently used inside a scanner.
In the past, Stanford researchers usually licensed technology to big players in medical imaging. But since we believe we should try a new framework, we decided to do it ourselves. From 2015 to 2017 I did research, in close collaboration with my cofounder and then clinical advisor, Dr. Greg Zaharchuk, who is a neuroradiologist and professor at Stanford. In 2017, we decided to co-found this company and recruited great talents and industry veterans. Right now, we have a team of 20 people and have gotten two FDA-cleared, CE-marked, and industry-first AI solutions in this domain.
His experience growing up in China
Technology, especially in the earliest days, has been more rapidly changing in China versus the U.S. With a company, sometimes we find the markets or the dynamic trends in China are way faster, and we need to really adapt to that. As Chinese, we know how fast things are changing.
A lot of successful Chinese entrepreneurs are technical-driven. We try to aim at starting with engineering or science and having a great technology first, and then work towards products with product-market fit.
We really appreciate technology and science education. I think it’s rooted in the culture that if we work hard, we will get returns.
Why he pursued medical imaging
On one hand, I learned medical imaging technologies from my family. In my extended family, I have a lot of close relatives working as doctors in hospitals. I learned a lot from them about how technologies can be really helpful for diagnosis.
On the other hand, I also learned imaging can be transformed into a novel approach to decode biomedical signals. I went to 北京四中, Beijing No. 4 High School. In the 晚自习 (evening study session) we were “encouraged” to watch CCTV news. One day, there was a news segment about brain-computer interfaces, which is where people can do “mind control.” If we time shift to now, it’s neural engineering and related to Elon Mask’s new Neuralink.
At Tsinghua, there was a lab developing a brain-computer interface using signals from EEG and the cortex of the brain. The news was about an easily accessible signal to control a toy dog to play soccer. It was really fascinating. It reminded me of cartoons or movies like The Matrix. I was excited to learn both directions are within the scope of biomedical engineering, and that technologies can not only save lives, but also lead us to futuristic applications.
The news was about an easily accessible signal to control a toy dog to play soccer. It was really fascinating.
I was pre-admitted to Tsinghua because I won first award in a computer programming competition called the National Olympiad in Informatics Province competition. I could choose the department, so I chose this as my first interest.
At Tsinghua, there were a lot of opportunities to do research. So in addition to working on a regular curriculum, I started visiting from lab to lab. I stayed in each lab for a semester to learn new things since the second semester of my freshman year. Around junior year I was working more in the brain-computer interface lab and developed several technologies. So I started working to decode signals like EEG and fMRI, and got into the world of medical imaging and artificial intelligence.
The summer before my senior year, I went to Stanford as a summer research student, part of the Stanford-Tsinghua program called UGVR. We were invited to do research, and I was working on optical imaging to create super-resolution imaging. Instead of images that are purely optically driven (reliant on the shots you take), optical super-resolution algorithms can compute the images and break the information limit: you can see the resolution you’re not supposed to see. From the research program, I got more interested in medical imaging, especially in using an algorithm to power the imaging.
Optical super-resolution algorithms can compute the images and break the information limit: you can see the resolution you’re not supposed to see.
Further, I was also working on a machine learning algorithm to predict from a signal what sentence a patient is saying, so basically decoding the language signal using artificial intelligence.
All this training made me technically equipped to work on my engineering research. In my PhD at Stanford, there was extensive collaboration between electrical engineering and radiology. It was a great opportunity and place to work in this area.
Pursuing entrepreneurship
I had serious thoughts about becoming a professor in this area when I was about to graduate. We were one of the first groups applying AI to medical imaging research back in 2016. We got one of the first papers, had several conference presentations, and I was invited to give different talks at universities. So I got a lot of academic exposure and I thought about whether I should pursue a professorship. It was a great time window to find faculty positions.
With entrepreneurship, you have to find funding yourself, you have to build the team, you had to develop something new, and to see whether you can sell it. But I still chose entrepreneurship. I started part of the journey even in Tsinghua, when I participated in an undergrad university entrepreneur competition. I did a lot of pitches and learned to do pitches through the mentors.
With entrepreneurship, you have to find funding yourself, you have to build the team, you had to develop something new, and to see whether you can sell it. But I still chose entrepreneurship.
When I got to Stanford, I participated in several hackathon and got a taste of quick prototyping. I co-founded Polarr, which was kind of a unique journey. Especially for Chinese students and engineers/scientists in the U.S., there’s a lot of “wantrepreneurship” going on, but very few really start doing a startup.
It was a fortunate event that I was involved in the early founding journey of Polarr and building a startup. My Polarr cofounder reached out to an entrepreneur group, and we got connected and he shared with me his initial idea of building the intelligent photo editing engine. Significant credit should also go to my girlfriend, who motivated me and shared the pain points of editing photos at scales, as she is also very into photography. All these thoughts came together for me to start Polarr and apply AI in image processing.
Then, I transformed these lessons into my new research, and I started Subtle Medical with my co-founder Dr. Zaharchuk, with my primary advisor Dr John Pauly as an advisor to the company on the technical side.
What to look for in choosing co-founders
Sometimes it’s 缘分. (Yuanfen, roughly translated as destiny or serendipity.) There’s definitely a luck and fortune component.
At Polarr, my co-founder was very technically driven, but was also a photographer, so he knew the market, whereas it’s not my major research industry. With Subtle Medical, I thought this was a journey I should not miss to take. It was exactly what I was doing for years and was exactly my research area. I understood the technical side and the market.
It was exactly what I was doing for years and was exactly my research area. I understood the technical side and the market.
My co-founder Dr Zaharchuk was my clinical advisor, who is a radiologist. We were really complementary, because he understood a lot of the clinical side that I didn’t understand. We have a lot of clinical-driven marketing. We have connections to the radiologists in top hospitals that we can start with as early collaboration sites. We know them and can talk to them to see what the really important piece of puzzle is, how the product performs, what defines the problem. That’s really helped. A shared pattern through my Polarr and Subtle experience is that co-founders need to be complementary.
You need to make sure you have the technical side and advantage, but also to understand the market need and understand the product. If you only have a hammer of technology, and try to find the nail, it often won’t work. You really need to think about what the viable product is and how to transform your existing technology or potential technology into a product, so you can deliver something that people really need.
If you only have a hammer of technology, and try to find the nail, it often won’t work.
On finding people to work with
At the beginning, you definitely hire through your existing network. So a lot of folks at Subtle are from Stanford and Tsinghua. That’s where we start. And then with more exposure, we connect with semi-networks, through a colleague or someone we know. We connect with people from both the industry and academia sides.
It’s important to hire people at the start where you think he or she is much better than you in a specific areafff. Start with your own connections. Even with a team of 20 we still rely a lot on our existing network. Maybe when we reach 50, we’ll have a more dedicated HR-type of recruiter.
There is a lot of talent within the field of medical imaging and image acquisition. It’s quite a niche area, we know almost everyone through research. A lot of them are either clinical partners, technical partners, or advisors of the company already. From this group, we are able to hire their graduate students and postdocs to build a team. It’s a big advantage when you have almost all the talent you know in this field. Finding talent is the most important aspect for us everyday. One of my major challenges as CEO is looking for talent.
Finding talent is the most important aspect for us everyday.
Working with investors
At the very beginning, I found a lot of investors through my experience at Polarr and through my Tsinghua/Stanford connections. I got to know Tsingyuan through Zhen Fund and a Tsinghua alumni I knew previously. Then I got to know more U.S. investors via connections through the Stanford, StartX program and from the Stanford Biodesign program. We got connections to investors like DCVC, Bessemer, Breyer Capital, and so on.
It was all through the network or secondary connections. Never hesitate to reach out for help. It’s like if you apply for a job. You have to get a referral. Don’t try to directly apply on LinkedIn. It shows your capacity if you can find connections to get you a referral. It’s important as an entrepreneur to find good connections, not only for investors, but also for positive customers and partners. We want to talk to more people and get introduced to additional resources we need.
It shows your capacity if you can find connections to get you a referral.
There will always be lots of challenges, so it’s a great experience working with investors who are always there to listen to your concerns and address them. It’s very helpful to get another opinion and a fresh eye to look at a problem and help me. As a Chinese person, we tend to be a little shy to communicate, so that’s something that I personally need to overcome.
His best advice
I got a lot of advice that you have to be adaptive and move fast. In Silicon Valley a lot of people think ”move fast and break things,” and then fix it later. But I also got the same amount of advice saying: you have to be patient and move slow and steadily.
I think both are valid. So that’s the challenge: you have to know when to take which. As an early stage startup, I think we have to be patient. But once we see certain submarket driving forces, we are all-in. We’ve been developing a lot in different submarkets to try to understand different modalities and patient populations. That’s a challenge. So my advice is: you have to know both, and you have to try to find a way to navigate between them.
You have to know both, and you have to try to find a way to navigate between them.
Vision for Subtle Medical
We’re starting with two products which create clinical, operational, and financial value to hospital imaging centers. The centers can see images clearly and are willing to pay for the technology to help them to deliver better care to more patients. We’ll initially focus on commercializing our existing FDA-cleared product. But there are a lot of different dimensions we plan to scale to.
First is scaling to new modalities. MR and PET are the most inefficient and expensive modalities, which is why there’s a market and why we did research on the technology. We’re looking at other modalities, developing new algorithms and doing more extensive market research to prioritize.
We got a $1.6 million NIH SBIR grant to work on a new product, SubtleGAD, with the goal of making medical imaging exams safer by reducing the contrast dose injected into the body. The recent findings suggest the potential safety risk of the compounds to deposite inside your brain or body. You definitely don’t want that to happen. But ⅓ of MRI exams need a contrast dose so the radiologists can distinguish pathologies. We developed new AI technology and IPs to intelligently highlight the low contrast signals that humans may find difficult to tell. We can reduce the dose by 90%, a 10 times lower dose, to achieve equal or even better image quality.
In healthcare, a provider usually has to trade off costs: time, safety, and the quality of diagnosis. Our products help them renegotiate this trade-off so they can get better quality and a safer, faster exam all together.
We’ve started with the U.S. market, and we’ve already deployed in imaging centers in Europe, Latin America, and China. Healthcare providers everywhere share the same pain point, so we want to scale the solution. Right now, we can’t travel because of COVID. But we’ve developed a solution to deploy 100% remotely and seamlessly, which is an asset we’ll leverage to scale to new markets.
Lastly, we have some contracts signed with big medical imaging OEMs (Original Equipment Manufacturers) and pharmaceutical companies. With OEMs, we’re building together features to make a scanner run faster and with better quality. For some modalities we need to partner with OEMs to integrate technology into the scanner itself.
We have the infrastructure to develop technology, so it’s quite easy to scale with different designs or products. We can have a quick turnaround time from concept to product. That’s something we try to always improve, so we can quickly iterate and find new opportunities.
We can have a quick turnaround time from concept to product. That’s something we try to always improve, so we can quickly iterate and find new opportunities.
On social media
As Chinese, we use WeChat the most often. I use LinkedIn for industry news and industry connections. And not as often, but sometimes I use Twitter, where there are a lot radiologist and healthcare professionals. I also get some connections and keep track of the latest news there. There are a lot of radiologists and AI scientists who are very active on Twitter.
Media recommendations
There’s a YouTube channel called Two Minute Papers that I’ve found is a very efficient way to learn new research papers. On WeChat there are a lot of good blogs like 机器之心 and 动脉网 for healthcare and AI.
In addition to official news outlets, podcasts, LinkedIn and 朋友圈 (WeChat’s “Moments” and friend networks) of KOLs are of great value to get information. They share the latest news and different media which helps you digest as well. That’s a way to both get exposed to new information and to interact with people.
On reflection
I like sharing my experience. I find that all the dots connect eventually. I’m not sure that’s representative. But I think that anything you get exposed and anything you experience will somewhat contribute to your career path and your love. That’s something I really value.
To get updates on this series, please follow our publication on Medium.
Follow us on LinkedIn: https://www.linkedin.com/company/tsingyuan-ventures/
Follow us on Twitter: @TsingyuanVC
Tsingyuan Ventures is a $100M seed-stage technology firm. We back technical founders across software, life sciences, and frontier technologies. Learn more about our origin story and our approach here.
Questions, thoughts, reflections? Let us know in the comments below. We’re always looking for great entrepreneurs and early stage ideas, and we’re always interested in having a discussion about venture, technology, and anything related. To see more about Tsingyuan Ventures, please visit our website: tsingyuan.ventures.
