Palo Alto – May 22, 2019 – Well, a lot has been said and written about the topic of AI and healthcare, and there is no doubt in my mind that a lot will continue to be said and written about this topic in the future. But this has also created significant confusion around what AI really is, especially as it pertains to the topic of healthcare. I thought I would try and distill it down to digestible chunks. Many publications tend to replace the old jargon of “big data analytics” with “artificial intelligence”. The phrase “big data analytics” has almost disappeared from the technology lexicon in favor of AI. The concept of AI, however, is a lot more powerful than just that.
The term AI was coined in 1956 by John McCarthy at Dartmouth with the “conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it”. But over the last 60+ years, the term AI has taken on a life of its own (pun intended). Now, while there are several nuances, leading technology companies like Amazon, Google and Facebook, for whom AI has become a core tenet, describe it as a combination of “recognition” (data) and “intent” (context) of data to mimic human reasoning. And the type of data can generally be bucketed into five fairly broad categories: see (text, images, and videos), hear (the spoken word), recognize (understand), feel (sensory/physical), and analyze (contextual thought/analysis). One could make the argument that with the advent of scientific fields like Brain Computer Interface (BCI) that our “thought” itself will become a data source. Touch and taste as the other human senses are probably not too far behind, as the pace of AI research accelerates. The natural progression for AI is to recognize ? synthesize ? analyze ? optimize.
Within healthcare, the impact of AI is, and will be, absolutely massive across vast swaths of repositories where both structured and unstructured data reside. But my goal in this article is to focus on the use of AI (recognition and intent) across two key data categories (visual and audio) in disrupting healthcare. I am going to split up the blog in two parts — In this blog (Part 1), I will focus on the Visual component, which primarily deals with structured image data with heavy reliance on machine learning. Part 2, focused on Audio shall follow in the coming weeks. Zooming out from the healthcare focus, the promise of AI is to be more efficient and effective in virtually every aspect of our personal and business lives. I am also not going to discuss the ethical and moral dilemmas of AI. I will leave that to the likes of Elon Musk and others to debate. Right now, I am just glad that Alexa, Siri and Cortana, the new age AI golden girls, are not monitoring what I am actually typing (or maybe they are and I just don’t know about it☺)
I am going to start by inserting useful market maps of AI startups in general and Healthcare AI startups in particular. Figure 1 is a snapshot of AI proliferation across multiple verticals while Figure 2 is the more important bird’s eye view of AI startup’s impact on healthcare.
Figure 1: Overall AI Disruptors Across Industries
Figure 2: Startups Transforming Healthcare With AI
Going back to the topic at hand, Visual AI is one area that has perhaps received the most attention, especially having to do with image analysis, with most of the initial focus on cardiology and radiology. I have a very personal story where AI could have potentially saved my father in law’s life. He was a cancer survivor who, a few years ago, had a bout with what was diagnosed at a reputed Bay Area hospital as Pneumonia, based on his X-rays. What the technician and the physician missed was the recurrence of his Renal Carcinoma that had metastasized in his lungs. It was only after he was admitted to his hometown (Southern California) hospital post an incident of stroke a couple of weeks later that the specialist at that hospital pointed out the cancerous spots on the original X-ray. In his words, “most physicians would have missed the indication in that image. It requires a well trained pair of eyes to be able to catch something like that”. A computer-vision enabled AI platform would likely have been able to analyze and detect that, especially in the context of my father in law’s oncology history. That, by the way, brings up another related topic that AI or human-enabled medical diagnosis in a silo is much less effective or relevant than one that is taken in the context of the patient’s medical or family history, genotype and phenotype specifics, and existing EMR data. While there are several interesting health-AI solutions, there needs to be a lot more emphasis on comprehensive solutions that tie together AI’s potential in conjunction with other clinical data.
The most visible and talked about application of AI visual analytics capability is in the field of early cancer detection. But AI is really the catalyst for analysis, accuracy and throughput on a much larger scale across image data, whether it’s in digital pathology or more specific radiology, cardiology, CT or something else. Machine learning in conjunction with AI will continue to push the technical boundaries and free up specialists for more anomalous cases or second opinions where human intervention is necessary. According to Dr. Paras Lakhani of Jefferson University Hospital, “neural networks are approaching human performance in terms of detecting malignant or benign growths in images such as mammograms”. The sheer throughput and accuracy of AI for image analysis in Radiology has increased multiple-fold in the past few years. In a recent study published in the journal Radiology, researchers at the Icahn School of Medicine at Mount Sinai reported that, “through a study conducted in 2018 comprising of over 96,000 tomography reports covering head CT scans, the AI machine was able to successfully identify radiological concepts with 91% accuracy”. This finding is remarkable and provides a clear indication of the ability of future AI machines to identify cardiovascular irregularities.
The impact of AI is truly global. In emerging markets, where per capita availability of physicians and specialists is the lowest in the world, AI will be a game changer and truly save countless lives. The AI health market as a whole is expected to reach $6.6B by 2021. Another related data point, perhaps not surprisingly, is the fact that healthcare AI companies have raised increasingly more capital in each of the last six years (Figure 3). Collectively healthcare AI companies have raised $7.5B across 691 transactions since 2013, more than any other AI vertical including Marketing Technology, FinTech, and Cybersecurity (Figure 4)
Figure 3: Healthcare AI Funding 2013 – 2018
Figure 4: Share By Vertical of $7.5B In Funding For AI Companies Since 2013
This tsunami of AI health investments has played a role in creating meaningful regulatory change. FDA has implemented “breakthrough device designation” for several AI algorithms, especially for clinical imaging and diagnostics applications, resulting in expedited 510(k) device and De Novo clearances. While FDA is relying primarily on its 510(k) regulatory pathway for approval, the fact that AI based algorithmic solutions are so new and different from anything that has ever come across the FDA, the agency is granting increasing numbers of De Novo clearances to get products to market. Listed below are a few examples of both De Novo and 510(k) clearances provided by the FDA recently across a broad spectrum of digital health solutions, as aggregated by mobihealthnews.com.
Viz.ai’s Contact is a clinical decision support (CDS) software platform that uses an AI algorithm to scan CT images for indications of stroke, and then notifies a neurovascular specialist if it identifies a potential large vessel blockage. Because the tool alerts the specialist during the time a first-line provider is reviewing the images, patients may receive attention from a specialist earlier than they would normally.
IDx’s IDx-DR is an AI software system for the autonomous detection of “more than mild diabetic retinopathy” in adults who have diabetes. The algorithm analyzes images taken with the Topcon NW400 retinal camera and uploaded to a cloud server. Within minutes the software provides doctors with a binary result, either indicating that more than mild diabetic retinopathy is present and that the patient should be referred to an eye care professional, or that the screen is negative and should be repeated in 12 months. The software is notable in that it was the first AI-based diagnostic system to be authorized by the FDA for commercialization in the US that can provide a screening decision without the need for clinician interpretation.
MaxQ AI’s Accipio Ix in an AI workflow tool designed to help clinicians prioritize adults patients likely presenting with acute intracranial hemorrhage. Cleared late last year, the algorithm automatically retrieves and processes non-contrast CT images to provide a case-level indicator, which is used to triage cases most in need of expert review and diagnosis.
Imagen’s OsteoDetect uses an AI algorithm to scan X-ray images for a common type of wrist bone fracture, known as distal radius fracture. The software can be fed images of adult wrists in the posterior-anterior and medial-lateral position, and using these highlights regions with potential fracture. OsteoDetect — which received De Novo clearance in May, 2018 — is intended for use by primary, emergency, urgent and specialty care practitioners alike, but should be accompanied by a standard clinical review.
The cloud-based DreaMed Advisor Pro is a diabetes treatment decision support product that analyzes data from continuous glucose monitors, insulin pumps and self-monitoring to determine an insulin delivery recommendation. Through an event-based learning process, the software incorporates a number of components into its recommendations, including basal rate, carbohydrate ratio and correction factor. Dosage recommendations are delivered directly to the monitoring clinician, who can push the adjustment to a patient’s diabetes management devices with the click of a button. FDA granted DreaMed a De Novo clearance in June, 2018.
AliveCor received 510(k) clearance back in 2014 for the AF algorithm, an app-based algorithm that works with the company’s smartphone ECG device to detect atrial fibrillation. Being a consumer-focused tool, users who receive a positive result are encouraged to print out and confirm their results with a board-certified cardiologist. Less than a year later, AliveCor also received clearance for two more related algorithms – the Normal Detector, which assures patients that their personal ECG is free of abnormalities; and the Interference Detector, a tool that automatically detects whether interference could be compromising their ECG test.
On the subject of atrial fibrillation, Apple made waves a few months back when it announced that its latest ECG-equipped smartwatch would also come with an algorithm to detect irregular heart rates. The software, which required the consumer tech company to seek a De Novo marketing approval, can monitor the user’s heart rate behind its various other functions, and alerts the wearer when it notices cause for concern.
Bay Labs’ EchoMD AutoEF software assists cardiologists by using an algorithm to automatically review relevant digital video clips collected from an echocardiography study, rates their quality and then selects the best to calculate ejection fraction, a key measure of cardiac function. Of note, the software can be integrated into a cardiologist’s routine diagnostic workflow to better assist decision-making.
The Coronary Calcium Scoring algorithm from Zebra Medical Vision offers a coronary artery calcification score from a patient’s ECG-gated CT scan. Clinicians can use this score to flag patients at high risk of cardiovascular disease sooner, thereby allowing for quicker and more effective care. The July, 2018 510(k) clearance was the first for the Israel-based company, which also holds a number of algorithm clearances in the EU.
In February, medical imaging software company Arterys Inc. touted 510(k) clearance for its Artyrys Oncology AI suite, a web-based platform that helps clinicians analyze ARIs and CT scans for signs of potential liver and lung cancer. The tool uses deep learning algorithms to expedite interpretation of these images.
What the examples above indicate is that AI is having and has the potential of having a tremendous impact on healthcare, especially when it comes to image detection and analysis across specialties. 70% of all healthcare decisions are made based on pathology results. These results can be far more precise through AI. The power of AI is two-fold: the amount of data that can be analyzed (throughput), and the increasing accuracy of the analysis. In this blog I have focused on images, but AI-based analytics can be just as powerful with respect to structured EHR data. And that the FDA is taking notice of the AI efficacy and creating expedited pathways for approvals, especially where it perceives the risk to be minimal, is truly remarkable.
But as with every other technology related advance, there is a human element that needs to be considered. In the case of AI and healthcare, the key question to ask is whether physicians broadly embrace the new innovation wave and embed the technology in their existing workflow, or feel threatened by it. Initial indications are positive since AI enables physicians to be able to see more patients by “outsourcing” the up-front analysis to their AI assistants. Personally, I am really excited about the prospects of AI-enabled diagnostics not only in developed markets, but more importantly the emerging markets where the technology can truly help create a healthcare leapfrog at massive scale. In addition to throughput and accuracy, which have been mentioned earlier, affordability is a nut that will have to be cracked for the broader global population to reap the benefits of AI in healthcare. Given the pace and magnitude of progress being made, I have no doubt that we will get there. And we are just scratching the surface☺
References:
https://www.cbinsights.com/research/healthcare-ai-startup-boom-expert-research/
For more information please contact:
Vini Jolly
Executive Director
Woodside Capital Partners International LLC
[email protected]
(650) 513-2755
About Woodside Capital Partners
Woodside Capital Partners is a global, independent investment bank that delivers world-class strategic and financial advice to emerging growth companies in the technology and healthtech sectors. With a strong track record in M&A, strategic partnerships and private placements, Woodside Capital Partners has been providing worldwide investment banking services since 2001 with leading domain experience in software, Internet services, electronic communications and materials, and healthcare. Woodside Capital Partners is headquartered in Silicon Valley. Securities offered through Woodside Capital Securities LLC, member FINRA/SIPC. For more information, please visit www.woodsidecap.com . NOTICE: No information herein is a recommendation that any particular individual should purchase or sell any particular security in any amount or at all, and is not a solicitation of any offer to purchase or sell from or to any particular individual.