#APSA50: Artificial Intelligence
Topic overview
This episode is the third in our #APSA50 series, where we teamed up with the Behind the Knife Podcast to cover the 50th Anniversary Meeting of the American Pediatric Surgical Association. In this episode, we interviewed the Dr. Michael Muelly, who gave the first of the meeting's PED (Pediatric Surgery, Education, and Disruption) talks on the topic of artificial intelligence in healthcare.
Dr. Muelly is a product manager at Google Cloud’s Healthcare & Lifesciences team and is a radiology faculty member at Stanford University. He has spent time researching machine learning, and has co-founded companies dedicated to bringing radiology to the cloud.
Stay tuned for more coverage of the 50th annual APSA meeting.
- How has AI already changed healthcare?
- Surgical specialties are currently less affected than others
- For specialties that are already digitized, such as Radiology, there are more data available, and AI is already making an impact in this field
- Examples of AI in Radiology
- Several centers have implemented an AI system that prioritizes CT Heads that are concerning for bleeds and moves them to the top of the reviewing Radiologist’s queue
- Most AI systems are in the operational realm, with augmentations to work flow instead of actually making diagnoses
- Will AI replace physicians?
- "I’m not working at Google because I think my Radiology job is going to go away.”
- It is important to educate about the actual capabilities and limitations of AI
- AI and evidence-based medicine
- Instead, AI should allow physicians to truly move towards an evidence-based system of treatment
- This is achieved by collecting large datasets and gaining new insights from them that were not previously possible
- Potential AI applications for surgery
- Laparoscopic surgery presents a great opportunity for AI, as you are digitally capturing the video
- Surgical robots could eventually use data captured from laparoscopic surgery
- More frequently, surgeons will likely have their work impacted indirectly, such as triaging trauma patients or ICU patients in a more data-driven manner
- Actual impact in the OR will take much longer, if at all
- AI in the ICU
- Sepsis alerts, which have historically not been especially helpful due to their simple algorithms, will be able to become much more sophisticated with AI support
- Limitations of AI
- The main challenge is that a huge amount of data is needed, and this can frequently be difficult to obtain in the healthcare setting
- The smaller the patient population, the more difficult it is to obtain enough data for AI to be helpful
- The quality of the output from AI is only as good as the quality of the data entered into it
- This challenge is less of an issue for search engines or self-driving cars, as millions or billions of data points can be relatively easily obtained
- For healthcare, data collection has to fit into the work flow
- AI models require guidance; otherwise, they may learn that arbitrary inputs are important, such as the presence of a ruler in a picture of a skin lesion making melanoma more likely, or a type of CT scanner being associated with a greater likelihood of pneumonia
- Using the Cloud to gather datasets
- Once data is centralized, it is much easier to create the framework that allows data sharing
- Having data in the cloud does not mean that anyone can access it, so other regulatory challenges will still have to be addressed; however, first solving the technical challenges will make solving these regulatory challenges much easier
- Ethical concerns of AI
- Bias in models is a huge potential issue; a model that is trained in one country may not be applicable to others
- Since AI models are completely dependent on the data entered into them, any bias that is in the dataset will also be seen in the model itself
- There are also questions for how AI should be regulated in healthcare, such as how the FDA will determine which devices are safe to use
- Future of AI in medicine
- Even if AI models are not better than humans, they are much more consistent
- Reducing variance between physicians and providing consistent care to our patients are the great opportunities of AI
Comments