Tag Archives

3 Articles
<span class="live-editor-title live-editor-title-145" data-post-id="145" data-post-date="2019-04-07 06:40:50">Roche study solidifies AI’s immense value in diagnostics</span>

Roche study solidifies AI’s immense value in diagnostics

by Srihari Yamanoor

 

I have said this before, whenever speaking about AI in the context of healthcare, and here we are, with Roche, once again proving that AI has a huge potential to transform diagnostics. Now, I also strongly believe that AI (and in the current context we are talking about Narrow, Weak AI) has the potential to transform therapy and the fused emerging field, theragnostics. However, diagnostics is a low-hanging fruit, and will be first conquered by AI.

The current Roche study

Roche scientists deserve kudos for fighting genericide, having picked Diabetes Macular Edema (DME), a specific condition affecting specific cohorts within those living with Diabetes. The biggest effect of diabetes is in the secondary effects that cause more harm to patients many a time. Without appropriate diagnosis, Macular Degeneration can lead to permanent damage and consequent degeneration of sight.

The choice of a specific eye condition is not a weakness, but a very good way to conduct research, and make a comparison between AI, which uses the two-dmensional Color Fundus Photos (CFP) versus the three-dimensional Optical Coherence Tomography (OCT), which is the gold standard, and requires human expertise for successful diagnosis. Plus, as Susan Shepard lays out in her review (link below), the former technique is available through telemedicine centers, while the latter is not, as it is more expensive as well. This also has implications for underserved populations in the developing world, as diabetes is a global health condition.

AI and Imaging based Diagnostics

Another aspect that makes this particular study powerful, and also demonstrates the power of narrow AI, specifically Deep Learning is the choice of diagnostics that relies on imaging. Over the past few decades, quite a bit of research and development has focused on making Deep Learning powerful, especially when it comes to imaging viz-a-viz applications in facial recognition and more broadly, image recognition. This has resulted in knowledge in the form of algorithms and techniques that can now be successfully extended to AI.

 

 

Explainability

The authors of the study also allude to the fact that they have been able to get the algorithm to spit out its secrets as to how it is arriving at the diagnoses. This is a fundamental requirement of AI. The more an algorithm is understood in terms of how it recognizes patterns and arrives at conclusions, the more value it has in aiding diagnoses.

Explainability is critical, not just for healthcare applications, but in many fields where AI will be put to use. It has also been an ongoing challenge, and will be a significant hurdle when it comes to application. It is of such high importance that DARPA is continuing to make investments to have explainability or XAI be a functional component of AI going forward.

 

Where we go from here

AI can make diagnosis, like  many other activities and tasks faster, cheaper and more accurate, supplementing and complementing humans. As I mentioned before, this can have an effect and applications of immense value with very high Returns on Investment (ROI) for governments, private organizations, patients and stakeholders sundry. There are also several social applications where it will greatly benefit humanity.

There will be progressive churn, and the competition to make AI a market reality, already very intense is only going to grow stronger over time.

Yet, I do not expect the transition to be smooth, and there will definitely bumps in the road ahead, in terms of technological, regulatory and other challenges.

 

References:

  1. The Qmed Article by Susan Shepard: https://www.mddionline.com/study-shows-using-ai-eye-screenings-could-improve-vision-outcomes?ADTRK=UBM&elq_mid=8095&elq_cid=74447
  2. Close-up Image of the Eye: https://www.pexels.com/photo/eye-iris-anatomy-biology-8588/
  3. Image of person checking blood sugar: https://pixabay.com/photos/diabetes-blood-finger-glucose-777002/
  4. Image of X-ray: https://unsplash.com/photos/ouyjDk-KdfY
  5. Image of Doctors Collaborating: https://stocksnap.io/photo/GSBJXWRSDV
  6. Image of Subway Stairs: https://burst.shopify.com/photos/up-subway-stairs?q=stairs

The FDA Designates an Eye-Diagnostic System as a “Breakthrough Device”

by Srihari Yamanoor

 

Last year, in my talk at the San Jose BIOMEDevice Conference, I had postulated that one of the key factors in the rate and level of adoption of AI based devices and systems would depend on how the FDA would act on such submissions. It is good to see that one company seems have to have proceeded with taking the FDA to task. The result is positive, and good news for companies pushing AI based systems, but also carries a tale of caution, for the road ahead for everyone involved. I will lay out the case briefly in this blog, but this is not the last time you will hear me talk about the effect of regulatory agency response to such device submissions and their long-term implications.

I just reviewed news that IDx, a company that designed what is being described as “an AI based autonomous diabetic retinopathy” detection system has been given “breakthrough” status by the FDA for their system. This provides the system with an expedited review, and potentially quick approval. This is very encouraging, not only for the organization with the retinopathy system, but also for all companies vending AI systems.

Here are a couple of key thoughts I have on this:

  1. A very promising outlook: I am not aware of how much Scott Gottlieb had, in terms of a personal hand in this decision, but I am sure his leadership had a measurable impact on the decision to expedite review. He has taken a stance of modernizing some of the approval processes, especially in the area of Digital Health. The FDA’s backing for AI based systems, is a very positive thing for companies with an AI focus on healthcare.
  2. Good for competition in the healthcare industry: Once the FDA sets AI based apps and systems on the path to commercialization, there will be no excuse for organizations that do not adopt Machine Learning and Neural Networks. Of course watch for a good spike in M&A activity in this regard.
  3. The Hype Cycle is around the corner: While companies with real ML/DL products and services will be out there, helping patients, there will also be the fakes and the wannabes that we will have to contend with. And it is quite possible that the rotten apples will ruin it for everyone. This is something to watch for.
  4. Unrealistic expectations and unintended consequences: I know this is closely related to the point about the hype cycle, but I want to make a refined point here. Even the companies with good AI based tools might push the envelope on hyping up the utility and efficiency of their products. Over-promise and under-delivery lead to angry and disappointed customers and eventually will create problems for everyone in the industry.
  5. How will CMS and private insurance companies handle reimbursement? This is a big question and can be the topic (and probably will be) of several blog posts. However, I just wanted to highlight that while AI might improve efficiency and automate diagnoses and treatments, companies have to resolve how and where the reimbursement will come from.
  6. A word of caution on regulation: While it is commendable that the FDA is moving forward with initiatives on things such as AI. However, I do not always agree they are doing a great job on things. I was irked by the craze with which the orphan applications were being reviewed by the FDA. The FDA, in the past, has gotten too cozy with the industry and Congress, with terrible consequences. Those who remember the Menaflex incident (reference below) remember how quickly the public’s trust in the FDA eroded, and how the organization took years to recover from that. Therefore, there is cause for concern here, and one hopes there will be a balancing act in making sure that the approval processes remain rigorous enough.

Conclusion: The current “breakthrough” designation and fast track review show great promise, but one hopes the FDA will balance it out with an equal dose of caution.

Subscribe and Support, Please! 

Did you enjoy this post? Please subscribe for more updates, using the sidebar. Have ideas or blog posts you’d like to see here? Contact me at yamanoor at gmail dot com.

References:

  1. The Press Release on IDx: https://www.eyediagnosis.net/single-post/2018/02/05/FDA-to-expedite-review-of-IDx-DR-a-breakthrough-AI-diagnostic-system?lipi=urn%3Ali%3Apage%3Ad_flagship3_feed%3B6Bf9xSfCRQSJXa4nvXjScA%3D%3D
  2. On what happened with Menaflex: https://www.fda.gov/downloads/NewsEvents/PublicHealthFocus/UCM183642.pdf
  3. Image, Courtesy, Pexels: https://www.pexels.com/photo/people-face-child-eye-32267/

Personalized Medicine might be the true ticket for AI’s march on Healthcare

by Srihari Yamanoor

I came across this interesting piece on siliconANGLE  focused on how big data is coming to cardiovascular medicine, via a precision medicine initiative. I found the read to be fascinating, and I recommend you read the article, provided in the references section below.

In summary, the American Heart Association (AHA) is collaborating with Amazon Web Services to build a data analytics platform. Details are thin, so I am unsure of how much is hype and how much is reality. I assume, as time progresses, it will indeed turn out to be very useful. The article describes the platform as a “marketplace” (yes, concerning, in more ways than one) for various forms of patient data:

  1. Personal Data
  2. Clinical Trial Data
  3. Pharmaceutical Data
  4. Hospital Data, and apparently, other data.

The goal here is to use this to personalize treatments for patients. Read more in the article. I believe that personalized medicine and AI share a deep connection. I want to share some of my own thoughts of how such efforts could give way to lasting influence by AI on healthcare:

  1. Personalized medicine needs knowledge, which comes through data. Similarly, learning requires data. Data helps you glean patterns, which is what learning is. Thus, AI and Personalized medicine are the two snakes of Caduceus, which in this case is the data itself. Thus, efforts to bring about personalized medicine should be expanded vastly.
  2. Knowledge is supreme. When it comes to personalized medicine and such, you hear all this hype of n=1. Unfortunately, for medicine to be effective, first you need large quantities of data, and with that the ability to glean actionable information from the data, to understand diseases, treatments, and their effectiveness. Whether you see this on a gross scale, or all the way down to a Single Nucleotide Polymorphism (SNP), this type of knowledge needs to be the horse in front of the cart. Without actionability, the patterns recognized through the data might not be useful. This knowledge doesn’t exist, so the hype cycle might cause enough damage and reduce funding levels later on. Realistic expectations ought to be set so that this doesn’t happen.
  3. There are two problems with healthcare data that come to the fore. First, we don’t really know that is being collected is appropriate. There is a lot of day dreaming about genetic data, racial data, physiological parameters (one company claims to be measuring 300 parameters!)  and other measurands and the hope that these will somehow magically transform themselves into diagnoses and treatments. With enough research, they will, but this will take expensive effort and time. Low hanging fruits such as breast cancer diagnoses, etc., are pushing people to imagine that the same level of rapid success will be seen throughout the healthcare continuum. There is no way to know if this will be the case.
  4. The second problem with healthcare data is about ownership and walls around the data. For example, will AHA share the data with the general public? If not, why not? Terms like “marketplace” give me pause. They might very well be just terms used to explain things colloquially. On the other hand, it is important to think about the issue of data sitting behind paywalls, or blocked of with HIPAA as a convenient excuse (more posts to come on all this). Data Democracy is a critical need for the progress of medicine and for Artificial Intelligence, eventually.
  5. What if the data is not free? What about the insights? Do we suddenly now start paying for the insights? Or, would CMS start reimbursing Doctors for personalized medicine? For the use of AI tools? There are several questions here, and only time will answer them.

Conclusion: 

Data will give us clinical insights. These insights will eventually fragment into meaningful personalized healthcare. The same data will lead to even more advances with AI. However, unanswered questions over data ownership and access remain.

 

References:

  1. https://siliconangle.com/blog/2018/01/22/can-precision-medicine-break-chokehold-on-healthcare-big-data-reinvent-womenintech/
  2. Image, Courtesy Pexels: https://www.pexels.com/photo/interior-of-office-building-325229/