Why the future of AI Healthcare needs Social Networks, underpinned by decentralization, collaboration, and data sharing.

In Dec, 2021, Dr Michelle Perugini presented the future of AI-Enhanced Healthcare at the AI Summit 2021 in Seoul. Michelle discussed how The Social Network model can enable AI healthcare that is scalable and therefore accessible and affordable for all.

Social networks use decentralization and the network effect, and have transformed many industries including media, video and now finance (i.e. DeFi). Presagen’s Social Network for Healthcare connects clinics and patients globally to enable collaboration and data sharing to access globally diverse datasets that are needed to create AI that is unbiased and equitable. The decentralized network democratizes the creation of AI products, promotes collaboration through incentives, and protects data privacy and ownership.

Michelle also provides a case study of AI healthcare product, Life Whisperer, that uses AI to improve pregnancy outcomes for IVF patients. The AI analyzes images of embryos during the IVF process to assess genetic integrity, as well as embryo viability and the likelihood of leading to a pregnancy.  Life Whisperer is being used by IVF clinics all over the world, from large, advanced clinics to small remote clinics.

Below is the transcript of the presentation and click here to download the presentation slides.

Slide 1: Hello everyone. Thank you for attending and thank you to the organizers for inviting me to speak today – and allowing me to speak virtually from my office in the San Francisco Bay Area. My name is Michelle Perugini and I am a Co-Founder and CEO of Presagen, an AI healthcare company.

Slide 2: As we all know, AI has the capacity to be able to improve global healthcare in terms of efficiency and outcomes. But this is not just for developed countries. AI provides an opportunity to support the needs of developing nations as well. For developed countries AI can drive efficiency for overloaded health care providers, by helping medical practitioners make critical decisions in a time poor environment.  For developing countries, there is a shortage of skilled experts. This is coupled with poorer quality health services, and general higher healthcare needs. Healthcare is a global problem. And when we look to AI, what we really need is a global equitable solution. That is, AI should be scalable, and unbiased, and therefore accessible and affordable to any clinic and any patient, any where in the world. Regardless of who you are, where you are, or how big or small or privileged you are.

Slide 3: However, there is a significant challenge in the AI healthcare industry where many of the AI solutions suffer from bias and lack of scalability. Recently one of the world experts in AI, Prof Andrew Ng from Stanford, said that they were able to develop AI using Stanford Hospital’s wonderful datasets, but then they struggle to get the AI to work in a clinic down the street. The problem is simple. Scalable and unbiased AI requires globally diverse dataset – one that represents different patient demographics and different clinical settings. The solution on the other hand is not so simple – it is actually very difficult because healthcare is highly regulated and protected, and requires a different innovative approach.

Slide 4: There are three major challenges with accessing and connecting globally diverse datasets in healthcare. First data privacy and sovereignty laws prevent medical data from being moved outside of the country or region it resides, which makes it difficult to bring together global datasets for AI training. So, we need a way to solve this. Collaboration is also key because data is often locked up in many different clinics distributed globally, and those clinics want to derive value from their data. And for patients, they want ultimate protection, control, and ownership over their data. Finally, Data quality is a huge problem in healthcare. Even 1% poor quality data can significantly impact AI scalability and accuracy, and clinical data is inherently poor quality due to subjectivity and uncertainty, so this needs to be addressed.

Slide 5: At Presagen, we have solved this problem by creating The Social Network for Healthcare, which uses decentralization and the network effect that has been successful in other industries like social media and decentralized finance. It’s a collaborative network that allows us to connect with clinics and patients globally, so that they can collaborate with us, and safely share data with us, so that we can build commercially scalable AI healthcare products. In return for collaborating with us, we provide royalties to clinics. For patients, we ensure they have control, protection and ownership of their personal medical data, and of course access to better low-cost healthcare services. The AI healthcare products that Presagen develops are delivered back to the same users that helped us build it, and others, that are connected to our global network. In this way, our network changes the way that clinics, patients and medical data from around the world are connected through artificial intelligence, and makes scalable and unbiased AI a reality. And our focus initially is women’s health, an under-served market that needs more attention. Later we will talk about our first case study for the fertility IVF sector.

Slide 6: Collaboration is key to enable globally connected data and the development of AI that is based on global diversity. This ensures that the AI that is developed is equitable, affordable, and accessible for all. Collaboration also democratizes the development of AI healthcare products. No longer is AI only for the large medical companies or prestigious hospitals like Stanford with the large datasets and large budgets and technical expertise to develop the AI. Collaboration allows any clinic of any size, anywhere in the world to contribute data to the development of AI healthcare products that will benefit all, and not just the privileged. And our collaborative approach does not exclude the large medical companies or hospitals, and is beneficial for them too. Because developing AI at a single institute level will not be scalable, nor is it commercially or technically viable. It is very expensive and will be biased to that patient population meaning it will need to be retrained or calibrated for every clinic.

Slide 7: Our network solves key technical healthcare challenges around data access and AI development. We use global cloud infrastructure to ensure diverse data from around the world is stored locally to comply with data privacy laws. Coupled with our own decentralized federated AI training algorithm, which allows us to train on data distributed all throughout the world without having to move or physically see the data. Finally, we manage data quality, which in healthcare is huge problem, with an algorithm that automatically detects errors in the data. This algorithm was recently published in Nature Scientific Reports, if you want to learn more.

Slide 8: Presagen’s Social Network brings together the whole healthcare ecosystem, enabling clinics and patients to safely connect, share data, use AI services, and get rewarded. This approach allows our team at Presagen to access diverse data, distributed globally, without having to move, own, or even physically see the data. This allows us to build scalable AI healthcare products. For clinics, they can leverage their data, and monetize their data without any commercial or technical cost or risk. They also receive royalties for their data based on clinical contributions and access high quality products that make them more competitive. And patients gain the benefit of better healthcare at a low cost, whilst maintaining control and ownership of their data. This is all possible with a Social Network approach – and the power of decentralization and the network effect.

Slide 9: The first case study we have applied our Social Network approach to is in the fertility IVF sector – to develop our first product Life Whisperer. Life Whisperer uses AI to improve pregnancy outcomes for IVF patients. Life Whisperer was built collaboratively with IVF clinics around the world, and is now being used by clinics globally, from advanced clinics in Europe, Australia, Canada and SE Asia, to remote clinics in Trinidad and Tobago, and less advanced clinics in India. It is a truly scalable AI product – demonstrated to work for different demographics in different counties, different clinical settings, and even widely different camera systems.

Slide 10: The idea for Life Whisperer was born out of an unmet medical need in the fertility sector. In IVF, selecting the best embryo during IVF procedures is critical to a successful pregnancy outcome. Current standard of care for evaluating embryo quality involves visual assessment of embryo morphology to determine embryo viability, that is, the likelihood of implantation. This is performed using morphological grading systems, of which the most common is the Gardner scoring system. A complementary method that is being increasingly used is using genetic testing, or PGT-A, to evaluate the genetic integrity of the embryo. It is well established that genetically normal embryos are more likely to have positive clinical outcomes. Embryo viability and embryo genetics, together, are the key determinants of overall embryo quality.

Slide 11: Life Whisperer uses AI to both assess embryo viability, which is the likelihood of the embryo of leading to a pregnancy, and embryo genetics, which is the likelihood of the embryo being genetically normal. This information helps inform the embryologist as to which embryo is healthiest and should be transferred to the patient, with the ultimate aim of improving the pregnancy outcome for the IVF patient.

Slide 12: The AI does both of the viability and genetic assessments non-invasively, using just a single image of the embryo. The embryologist simply drags and drops the embryo image, which is already taken in most IVF labs, onto the web-based application, and they get an instant result as to the embryo quality. Essentially, the AI assesses the morphology of the embryo, and based on what it learnt from thousands of embryo images, is able to identify morphological features that indicate the embryo viability and genetic integrity. If you go to our LifeWhisperer.com web site, you can see a demo video of the Life Whisperer software in action.

Slide 13: There are three primary benefits Life Whisperer provides to clinics. First is Objectivity of embryo assessment. Embryologist can be subjective with their embryo grading, and even the same embryologist may give the same embryo a different score on a different day. The AI score is always the same for the same embryo image, and the score is evidence based – based on the learnings from thousands of images of embryos. Second benefit is ease of use. Embryologist just need to drag and drop images that they are already taking onto the web application. There is no annotation required, and they get an instant assessment in seconds. The third benefit is cost effectiveness. Some AI algorithms require an expensive piece of hardware – a time-lapse machine. Our AI works with both camera types – a standard microscope camera or a time-lapse camera. This ensures scalability and that the AI works in all IVF clinics, and not just the prestigious clinics that can afford the expensive camera equipment.

Slide 14: The benefit for the patient, outside of better IVF outcomes, is about transparency. Patients love the Life Whisperer patient report that is automatically generated which explains the AI score relating to their embryo assessment.

Slide 15: One of the amazing things about AI in this use case is the ability to learn and identify things that are either not yet known about embryos, or complex morphological features that are not easily visible to the human eye. In a recent study we have shown that the AI is correlated to the embryologist manual grading method, the Gardner Score, however the AI actually performs better when looking at pregnancy outcomes. This suggests that the AI is detecting features that the manual assessment is not.

Slide 16: Life Whisperer Viability, which assess embryos for their likelihood of leading to a pregnancy, has been extensively tested internationally, as well as used in-market for almost two years. Clinical studies have shown that the accuracy of Life Whisperer Viability for predicting clinical pregnancy is up to 25% better than when using manual visual assessment alone. We have also shown that ranking of embryos using Life Whisperer Viability can decrease time to pregnancy by 15%, which will result in tangible clinical benefits to both patients and IVF clinics alike. As well as the reduced physical and emotional burden on the patient as a result of requiring fewer cycles to reach pregnancy, the average patient savings is estimated to be $1200 per patient.  Clinic revenue is also expected to increase by an average of $4,600 (per patient), due to the ability to service more first cycle higher value patients.

Slide 17: Life Whisperer Genetics has also shown incredible results for a non-invasive procedure that only looks at images of the embryos, the morphology, in comparison to a procedure which is invasive, requires a biopsy to cut out a piece of the patient embryo for genetic testing, and in many countries either not available or very expensive. This slide shows the results of the AI score, but put simply, when we used the Life Whisperer score to rank embryos in a simulated cohort study, Life Whisperer Genetics was able to rank the genetically normal embryo as the top embryo 82% of the time. It should also be noted that due to the nature of the biopsy approach, PGT-A genetic test is not 100% accurate. The embryo cell or cells that are cut from the embryo during the biopsy may not be representative of the whole embryo. Life Whisperer Genetics was not developed as a replacement for the invasive and costly PGT-A genetic test. It can be used to pre-screen embryos prior to undertaking a genetic test – saving the patient cost of unnecessary PGT-A tests. Or it could be used as an alternative for PGT-A for patient not comfortable with undertaking a biopsy of their embryo – in other words removing a part of the embryo that may eventually become their child. However, in this case it is important for the patient to understand that the AI is not as accurate as the PGT-A test, and the patient report which explains simply the graph shown on this slide, helps manage expectation of the AI result’s accuracy and likely outcome.

Slide 18: To summarize, our Life Whisperer product has demonstrated the power of our scalable AI approach, which is using The Social Network for Healthcare, a collaborative approach of building AI Healthcare products with and for clinics and patients globally, who are the ultimate end users and beneficiaries. The power of sharing and collaboration can achieve amazing results – as we have shown for our first two Life Whisperer applications with great performance and benefits for clinics and patients globally. As a business, as well as continuing to create more products in IVF, which we have underway two new IVF applications, we are now ready to expand beyond IVF and apply the same model to other areas of women’s health, such as OBGYN, and beyond.

Slide 19: Our vision at Presagen is to create the largest global network of clinics, patients and medical data to make AI-Enhanced healthcare accessible and affordable for all. And we will achieve this by democratizing AI development for the healthcare sector by allowing anyone to contribute valuable diverse data needed to build the AI, facilitating sharing and collaboration through incentives and rewards, and allowing decentralized data access and product development without owning patient or clinic data, which is key.

Slide 20: Thank you for listening. If you have any questions, please reach out to me.