ORAiCLE

Cardiovascular disease (CVD) contributed to approximately 31% of deaths globally in 2019. It is shown (in 2020) that with a 4% reduction in CVD incidence, one-year direct medical cost savings were estimated at US$21.9 billion. CVD is the commonest cause of hospitalisation and premature death, and the single most expensive health condition to manage. Commensurate with the large CVD burden, the United States spends >$320 billion annually (15% of health care spending) managing and treating CVD and its main risk factors. However, current CVD risk prediction equations only have <65% specificity and require multiple blood tests and GP visits.
Current approaches to lowering CVD risk rely on accurate risk prediction for those who are at high risk of events, in order to appropriately target the increasingly expensive range of medications that lower CVD risk. Together with increasing cost, medications are increasing in number and efficacy, while current CVD risk assessment still has high false negative rates. i.e. a considerable number of CVD events still occur in those classified as low risk, who have not received risk modifying medications. Recent, single measurements of blood pressure, or glucose, lipids and kidney function typically used in risk prediction, do not accurately represent overall exposure to CVD risk. This is because CVD is a chronic disease resulting from long-term cumulative effects of multiple risk factors. The limitations of single measurements of individual risk factors, contributes to the poor specificity of calculated CVD, along with only identifying a subset of known risk factors that culminate in narrowing of the key blood vessels in these critical organs (heart, brain and kidneys). Accurate CVD risk analysis needs to account for longitudinal exposure to multiple parameters directly relevant to CVD.

The retina is the only part of the human vasculature that is directly visible by non-invasive means. Several studies have recently shown that an artificial intelligence (AI) can use retinal imaging for estimating CVD risk. Furthermore, a very recent and growing body of the literature demonstrates a direct link between disease of the eye and CVD-related mortality. We have created a novel solution which uses retinal imaging and our unique AI capabilities could significantly improve the accuracy and accessibility of medical triage of patients with high CVD risk. 

The Google AI research unit showed in 2018 that the CVD risk factors (HbA1c, blood pressure, age, …) can be predicted from retinal photographs, with moderate accuracy (AUC < 0.72). Although fundamental to our approach, there were several technical shortcomings in this study. The dataset used in this research had major data inconsistencies (e.g. missing CVD event records in more than half of the population). The researchers hence decided to use a single time measurement of CVD risk factors (e.g. one-time HbA1c record), which is clinically known to be unreliable. Consequently, they used a simpler AI architecture (Inception-V3) that suited their single-time laboratory measurements. Finally, they did not use a parametric representation of patient’s metadata as part of their CVD prediction. 

At Toku Eyes, we have improved upon this research and some of our achievements are available through our published papers, with more in-press. We have created the world's largest international dataset of 500,000 patients, to create ORAiCLE. 

In our clinical studies (in-press) ORAiCLE has proven to be highly accurate in predicting cardiovascular risk, using retinal images. 

At Toku, we are aiming to help the low and middle income communities that are most at risk of undiagnosed CVD risk, and early mortality. Our solution will specially assist those with CVD comorbidities, including diabetes , hypertension or high cholesterol.

More people will die prematurely from CVD than any other condition. Our proposal dramatically decreases the cost of screening services, by

a) identifying those at low-risk without requiring laboratory tests;

b) accurately detecting high-risk patients likely to experience CVD events within 5 years; 

 c) revealing the presence of individual CVD risk factors (hypertension, lipid levels, glucose levels and kidney function

This solution has benefits for patients, prescribers, and the health system. More accurate and more accessible CVD risk prediction information will benefit most adults. A more accurate, inexpensive, non-invasive and point-of-care CVD risk prediction, done at the same time that a retinal screening image is taken, could improve inequities through improved targeting of CVD risk-reducing medications, and saving those at low CVD risk from unnecessary drugs and possible side effects.  In practice, if proven to be effective, existing retinal screening providers could access the AI-CVD risk algorithm through the cloud, and the additional information about CVD risk would be provided as part of an automated report from the retinal photograph assessment by AI. This would have flow-on effects for the health system with cost savings.

The major social benefit will be rebalancing of equity in healthcare because our product will be easily accessible in rural and overburdened healthcare centres.  ORAiCLE will remove the need for multiple blood tests and GP visits and will significantly improve the speed, costs, accessibility and accuracy of CVD risk prediction. 

Associate Professor Ehsan Vaghefi is the founder\CEO of Toku Eyes and holds 4 patents around this technology. Through Toku Eyes, Ehsan has led its team to create ORAiCLE™, the most accurate medical AI designed for CVD screening via retinal imaging. As CEO of Toku Eyes, Ehsan has employed 14 FTEs in AI engineering, bioinstrumentation and regulatory (Toku is ISO13485 certified). Ehsan has supervised the roll out of Toku's products in New Zealand (as part of the national screening services) and India (as collaboration with Aravind - the largest eye hospital in the world). 

Dr David Squirell is the co-founder of Toku Eyes and has acted as its clinical lead for the past two years.  David is the Ophthalmology clinical lead for Lead Ophthalmologist for the New Zealand Diabetic Screening program. David has served as PI for a number of drug company-sponsored clinical trials, including Intravitreal Aflibercept (Eylea). 

Toku Eyes benefits from an internationally recognized scientific and executive board, as well as extremly capable AI engineering core. 

 https://tokueyes.com/our-team/

While ORAiCLE as a CVD screening tool is highly scalable, it's running cost (cloud GPU computational capacity) can make it prohibitve for low and middle income societies. 

We are hoping that through qualifying in this challenge, we could cover the cost of providing access to ORAiCLE in these communities.   

Cardiovascular disease (CVD) is the leading cause of mortality globally, and remains the commonest cause of hospitalisation, health loss and premature death for people with diabetes. While some are at low CVD risk and others are at extremely high CVD risk, people with diabetes are frequently treated as a homogenous high-risk group.

Current CVD risk prediction could be significantly improved. International CVD risk management guidelines recommend that treatment decisions (for people with and without diabetes) should be informed by their predicted CVD risk. CVD risk prediction equations derived from many population-based studies in many countries around the world. While these equations have progressively improved risk prediction compared to previous equations developed elsewhere, their accuracy remains modest. The main reason for this is that none of the currently available predictors are able to measure vascular disease directly. The retina is the only part of human vasculature that is directly visible without invasive measurement. We have shown to clinical studies that retinal imaging and AI-enhanced image analysis can provide a direct measure of vascular disease, could significantly improve current CVD risk prediction equations. 

Increased accuracy: An AI-generated retinal score is non-invasive and has the potential to provide similar CVD risk assessments at much lower cost, is more accessible and will be quick to assess. This solution could simplify, improve accuracy of and significantly increase the accessibility to CVD risk prediction. We believe that this research will be a step-change towards better understanding of “biological aging” and monitoring it through retinal imaging as it has been shown the large differences among one’s biological and chronological age is directly related to higher risk of mortality.

Point of care healthcare delivery: The other equally important aspect of this innovation is enabling the “point of care” CVD score calculation, which can lead to instant identification of high risk individuals and opportune engagement regarding improving their wellbeing. 

Optimized health budget allocation: Finally, in many health systems, it will be possible to target expensive new medications more cost-effectively to those who need them. 

Toku is using Amazon Web Services to capture data directly from centers that are equipped with ORAiCLE and creates up to date dashboards on 

• number of people screened
• level of risk
• nature of risk
• number of referrals 

These dashboards are directly accessed by clinicians, health workers and policy makers

We have developed world's first, exportable medical AI for cardiovascular disease (CVD) risk factor screening through retinal fundus imaging (ORAiCLE). 

ORAiCLE is an extremely high value-add medical product that can transform CVD risk factor screening, risk prediction and personalised risk management in NZ and internationally, to deliver equitable, environmentally and economically sustainable programmes.

Our team is a diverse and multi-disciplinary collaboration of leading scientists from diabetology, ophthalmology, cardiovascular disease epidemiology, medical AI and optical imaging, with >400 publications and several patents in related fields. 

By leveraging these core competencies, we will develop our ORAiCLE will even be available for smartphones that are capable of capturing reitnal images

A dataset of approximately 500,000 people who have also had retinal images in the past 20 years have been curated. This dataset includes CVD risk assessment data, every time primary health care practitioners completed a standardised CVD risk assessment. 

There is growing body of literature demonstrating a direct link with retinal diseases such as DR, DME and AMD and CVD related events and mortality. Toku Eyes have already created, clinically validated and published highly accurate AIs for detection of retinal diseases, that are now going through their FDA regulatory approval for market introduction. 

We have recently shown that creating an ‘ensemble neural network’ and using a combination of several inferences will create a much more accurate AI tool. The original design of the convolutional neural network is based on the INCEPTION-RESNET-V2 design, plus the multi-context design, since our experience shows that this approach has the advantage of faster convergence speed. This ‘ensemble’ design is then be further modified to enable the network to be trained on multiple images at the same time. Each imaging stream is set up to ‘learn’ a different feature (i.e. CVD risk factor). Finally, the parameterised patient metadata is added to this layer, for a comprehensive representation of retinal and CVD data. 

Healthcare workers that can capture retinal images

We are a multi-cultural team, at staff and management level, with a very high percentage of immigrants (Iranian, Chinese, Indian) as well as native (Maori) people of New Zealand 

We are already operating in low income communities in India, where our products are offered at cost. 

Toku is operating as a social enterprise, offering its services as SaaS across the world

Toku's products are going through FDA clearance process and soon will be available in the US market, where they will be elligible for Medicare and Medicaid adopted CPT codes.

Toku will become financially sustainable by the end of 2023 

Toku is already part of public health screening services in New Zealand and India. Due to the scalability of its solution, Toku will become sustainable after entering the US market