Dr. Jeff Chang, radiologist and co-founder of Rad AI, tackles the AI and radiology questions he’s asked most often…
Dr. Jeff Chang is a radiologist and co-founder of Rad AI. After starting medical school at NYU at age 16, Jeff became the second youngest US physician on record. He did graduate work in machine learning at the University of Edinburgh, his fellowship in musculoskeletal MRI, and has an MBA from UCLA Anderson. He has also been a practicing radiologist with Greensboro Radiology for the past ten years, launching their Emergency Radiology section in 2010.
Rad AI is building an AI-driven platform to streamline radiology workflow, automating redundant manual tasks to save radiologists time and effort. Rad AI is funded by Gradient, Google’s AI venture fund, as well as several other well-known funds.
Prior to Rad AI, Jeff co-founded Doblet, a Y Combinator hardware startup, coordinating its engineering, prototyping and manufacturing efforts. Jeff has prior experience in both venture capital and private equity, and is an angel investor in several dozen AI and other deep tech startups.
- How did you first get involved with AI and in what capacity?
In 2012, I became very interested in a predictive analytics startup called Prior Knowledge, which Peter Thiel discussed in his CS138 Startups course at Stanford. I reached out, met with the team and was offered a role, but then they went quiet and were shortly thereafter acquired by Salesforce.
Halfway through the Master’s program, I was recruited to co-found a machine learning-centric hardware startup (using SVMs to track emotions based on biosensor data) back in San Francisco, and decided to leave the University of Edinburgh early to work on the startup.
- What motivated you to go from clinical radiology as an ER radiologist to moving into the business world?
While in radiology residency, I had a good friend who was applying to business schools. In the process of helping him prep on the GMAT, reviewing essays, etc. I became really interested in learning more about business.
Since I was already heading to Los Angeles for musculoskeletal MRI fellowship, I applied to UCLA’s FEMBA (fully-employed MBA) program, and began B-school at the same time as fellowship. Since the goal was to learn as much as possible about many different industries, I became a member of all 17 professional associations at UCLA Anderson, organized the FEMBA weekly events calendar and ran the FEMBA side of UCLA Anderson’s Entrepreneur Association.
After completing my MSK fellowship, I found a radiology role that provided extra time to explore other industries, including an internship in venture capital, followed by a role in private equity, and eventually, after completing business school, moving to San Francisco for Dev Bootcamp.
- What is Rad AI and why did you start it?
Rad AI is a radiology AI startup focused on streamlining radiology workflow to save radiologists time. Its first product automatically generates a customized impression from the findings and clinical indication dictated by the radiologist, using the most advanced neural networks. It learns each radiologist’s language preferences from all of their prior reports, to create an impression that the radiologist can simply review and sign off. The impression consists of the summarization of findings, associated conclusions drawn, and corresponding follow-up recommendations, and is typically the part of the radiology report read by most ordering clinicians.
In addition, Rad AI improves report accuracy and consistency by making sure to include significant incidental findings, answering the main clinical question, and providing the most appropriate consensus guideline recommendations for follow-up.
When using Rad AI, radiologists experience total time savings of just over 1 hour per typical 9-hour shift, while noting less burnout and increased job satisfaction. Rad AI integrates seamlessly with the major voice recognition solutions in the radiology market, and works across all x-rays, CTs, and MRIs. Later this quarter, we’ll be adding functionality for all ultrasound reports.
In addition, Rad AI is working with a large health system on the automatic tracking and follow-up of incidental findings, to ensure that the appropriate follow-up imaging studies are performed at the correct times. This significantly improves the quality of patient care, while decreasing overall healthcare costs by reducing long-term morbidity and mortality.
We started Rad AI to really focus on developing AI products that provide tangible benefits to radiologists, reducing repetitive tasks and helping radiologists better enjoy their daily practice of radiology. Rad AI’s products are developed by radiologists, for radiologists, and that shows in the product implementation – it involves zero changes in existing radiology workflow, and is extremely intuitive to use.
Rad AI is based in Berkeley, CA, and is backed by Gradient Ventures, Google’s AI fund, which invests in the top companies using AI to positively transform their fields. We’re now working with 5 of the 10 largest radiology practices in the US, and expanding quickly.
- What is a typical day for you?
Typically, I spend quite a bit of time each day prioritizing product development and business development strategy with our highly passionate and amazing team, talking with our radiologist users and gathering their feedback, and figuring out how best to improve our products. In conjunction with product demos and discussions with potential customers, these meetings generally take up about half of each day. In addition, I help with model results validation, debug and test code, and handle a variety of paperwork and emails each day.
Since the start of the COVID-19 pandemic, our team has become fully remote, so team communication happens mainly via Zoom and Slack, while customer and partner communication happens mainly via Zoom and email.
- What can I do now as a resident/fellow to prepare for the future of radiology?
If you’re interested in better understanding both the promise and limitations of AI more directly, I would suggest learning Python first, using a course such as this. Then you can take several of the courses on AI at deeplearning.ai or fast.ai, which provide a good grounding. If you’d like to run your own machine learning experiments, you can find a variety of public radiology image and report datasets readily available online for analysis.
From there, you can keep abreast of the latest developments in AI in radiology, by reading Radiology: Artificial Intelligence and SIIM’s Journal of Digital Imaging, and the latest developments in the field of AI overall, by reading some of the most cited papers in arXiv, NeurIPS proceedings, and similar resources. Form an AI journal club at your residency program, and see if there are any interesting AI research projects that you can begin at your clinical center.
If you’d prefer to focus on how to adapt for the future once you join a radiology practice, you can keep thinking and reading on the subject. Join the AI steering committee in your practice, or if there’s no AI committee yet, look at starting one. Work closely with AI startups that you find promising, and assess whether they provide actual benefits and return on investment for radiologists and radiology practices. Keep identifying ways for your radiology practice to adapt, and adopt new products that actually improve radiologists’ well-being, while also improving the quality and efficiency of patient care and diagnosis.
Think about what the typical radiologist’s work day might look like in 2030 or 2040. Which parts of the radiology workflow can be streamlined, and which parts of clinical practice will radiologists be free to focus most of their time on? Figure out what steps we need to take now, to be prepared for that future.