No movie series (Star Wars, LOTR, Nolan’s Batman) is complete without a trilogy, and here is my Part III, the conclusion to our trilogy - AI in Healthcare.
In the last post, we looked at some interesting and important use cases across different sectors in Digital Health including Payor and Provider administration, Real World Data, and Real World Evidence. In this post, I want to focus on more niche use cases that might span one or more sectors we explored in the previous two posts.
Note:
The last post mentioned that the US healthcare industry is worth $800 billion. I received a few follow-ups on this figure and wanted to clarify that this is more or less the figure for the healthcare industry in a traditional sense (providers, payors, IT, medtech, pharma). This valuation figure is likely an underestimate once you start to include more Digital Health, Telemedicine, up and coming Medtech unicorns. The spending in the industry is a lot higher, $4.3 trillion is the closest estimate. More on this here.
Medical Device (Hybrid) - Cleerly
Although we covered software-only medical devices aka Software As a Medical Device (SAMD) in Part I, we need to spend some real estate discussing traditional or hybrid medical devices using AI. One very exciting project in this space is Cleerly. When I say hybrid, Cleerly still utilizes some hardware to collect data that feeds into its revolutionary platform. Cleerly has built a digital care pathway to enable more efficient diagnoses, personalized treatments, and tracking for heart disease. Its platform combines data from a Coronary Computer Tomography Angiography (CCTA) scan, which is easy to perform in many clinical facilities, with its AI technology. An example of how this works is that CCTA converts a scan into a color-coded image that depicts not just the size of an arterial blockage but also the type of plaque. These AI algorithms then generate a 3D model tracking a patient’s arteries and their measurements of plaque, blockage, etc: The holy grail is the staging system that can longitudinally track and predict plaque buildup for a patient. The plaque volume is analyzed by the machine learning algorithm (approved by the FDA) to create a more comprehensive picture of the patient’s condition.
Heart disease is the leading cause of death and surprisingly, there is essentially no staging system to make these measurements on a longitudinal scale. Staging systems only exist for Cancer where a patient is given a prognosis (usually Level I to Level V). With Cleerly’s platform, heart disease can be quantified more to manage patients’ treatments and overall health better.
As we looked at in the earlier posts, medical devices are going through an exciting phase. There is still innovation in the traditional hardware-focused medical devices space and rapid innovation and investment in software medical devices. In addition to the exciting SAMD startups, the interesting developments will be in how medical devices utilize the existing technology in clinics and hospitals and combine it with AI to produce more comprehensive and groundbreaking results. The investments in startups are already there but a more accurate indicator might be the validation of the science behind these companies, especially since there is a lot of noise concerning which ones have results to back their proposed methods. One way to track this progress is through green lights from the FDA which shows that we are heading in the right direction as far as science is concerned.
Drug Discovery (De Novo Antibody Developments) - Absci
Another exciting company that spans multiple sectors that we looked at in previous posts (Large Language Models (LLMs) and Real World Data) is Absci. We need to look at Absci on its own because there is a ton of work being done with Machine Learning and AI for clinical trials such as tracking and managing clinical trials better to make them more efficient and faster, understanding patient outcomes to improve future trials in an area but very few companies are using AI, let alone generative AI, in the science and biology of drug development. Absci’s primary focus lies in de novo antibodies. De novo antibodies refer to antibodies that develop exclusively after organ transplantation. However, this concept has revolutionized clinical trials, where the integration of artificial intelligence (AI) can reduce the number of required human participants by half. By leveraging AI to generate diverse antibody sets, Absci aims to create an artificial but accurate and comprehensive clinical trial pool. The company employs Zero-Shot learning techniques to facilitate this process. Unlike conventional Machine Learning algorithms that can only classify objects within their trained classes, Zero-shot learning enables algorithms to classify objects based on contextual information. For instance, by incorporating contextual data about wild cats into the algorithm, it may even be capable of accurately classifying a cheetah, despite not being explicitly trained to do so. In the biotech world, zero-shot generative AI can generate antibodies to bind to specific targets without using any training data of antibodies that bind to those targets. More details on the specifics of antibody binding and how Absci achieves this are here.
Drug Discovery (Accelerating clinical trials) - Insitro
Insitro is novel in that it is using machine learning to more efficiently identify and design drugs and treatments. Traditional drug developments spend tons of capital and time trying to find the right drug for the target and usually, these projects run until they find a match. However, Insitro uses machine learning and advanced computation to develop a new type of model - Insitro - a combination of understanding diseases in vitro systems with in silico machine learning models. While the specific type of Machine Learning tools Insitro is using is not yet public, their models allow them to quickly differentiate between cell states with much more detail and identify clinical traits that are more relevant to the development of a particular treatment.
Insitro revolutionizes and speeds up the first three stages in the traditional model, which uses a rote trial and error method to find the right combination of treatment and effect, before a clinical trial and more importantly, identifies new molecular compounds that might not be surfaced in traditional clinical studies. Insitro methods, analyzing and understanding the target site (such as an antibody, protein etc:) can use ML to essentially predict what the ideal compound aka drug might be.
If Insitro’s models can start to deliver results, drug discovery could be heading toward this completely new AI-based model. Given that this is a new model, Insitro will be faced with challenges such as a lack of available data to train its models and scaling its pipeline enough so that it can actually compete with traditional drug discovery methods prevalent in the commercial space. The company is starting with developing a treatment for Amyotrophic Lateral Sclerosis (Lou Gerig’s) and Frontotemporal Dementia.
Medical Imaging (Rapid Cancer Detection) - Behold AI
My previous posts did not cover medical imaging because medical imaging is how AI started in healthcare. While machine learning and imaging are nothing new, it is still necessary to look at some of the exciting projects in this space - one is Behold AI. Lung Cancer is one of the leading causes of death globally and Behold AI is reducing the time to detect lung cancer from weeks to minutes. Most Lung Cancer detection pathways today make patients wait weeks to months for a CT scan and more time in providing an initial diagnosis. Using deep learning, Behold’s platform, Red Dot, significantly reduces the time between a patient undergoing an initial X-ray to a CT and suspected SLC (Suspected Lung Cancer). The highlight for Behold AI is that it has already demonstrated that RedDot can reduce an average of 1-3 weeks to detect cancer down to 3 days.
Red Dot’s impact in splitting patients into suspicious v non-suspicious groups is significant. The platform has demonstrated an error rate of 0.33% compared to 13.5% by human radiologists. While imaging has been integrated more and more into radiology workflows, imaging technologies need to continue to demonstrate effectiveness at scale as Behold AI has been.
Medical Imaging and Administration (Dental AI) - Overjet
Overjet is a very cool AI application and one of its kind specifically focused on transforming dentistry. Overjet uses AI to empower both payers and providers. What Overjet does well and that sets it apart from other medical imaging applications, especially for dentists, is its comprehensiveness and presentation. The machine learning algorithms detect calculus, enable longitudinal comparisons and bone level quantification, and present all this to the customers with high-quality visualization. The platform is a one-stop shop for dental practices to not only efficiently run the clinical side of things but to also manage their administrative workflow - the company also uses AI to identify overutilization and underutilization as well as coordinate appointments and patient flow.
Overjet also provides payers with AI recommendations on the flip side. Using the clinical AI results, Overjet is able to tell payers whether treatment is viable or not. This has helped a payer save up to $4M and up to 790% in ROI for certain dental procedures. Overjet’s business is a lesson for other AI companies to build more comprehensive and one-stop shop solutions. Managing the clinical and the administrative sides of the business can enable more coordinated solutions, and workflows between different healthcare entities involved and ultimately improve patient care significantly. Just take a look at Overjet’s success so far.
Summary
This concludes a 3-week summary of AI in Healthcare. Although we covered many sectors in healthcare and some very interesting projects, this is just the beginning for AI in Healthcare. AI’s impact in healthcare will be dictated by how companies set standards for the industry as we saw for Real World Evidence in my last post and Insitro for drug discovery and more importantly, how regulation will shape the industry in many aspects going forward. Additionally, there is a lot more to be seen for AI’s potential in generating revenue, financial success for companies, and its place in Healthcare as a business. One thing is for sure though, AI is starting to impact every sector of healthcare and setting up for an exciting next 5 years.
See you next week with more healthcare stuff!