This graph shows we can repeatedly distinguish different levels of concentration. The graph on the right is very similar to the graph on the left, but it’s based on clinical data. Many technologies work in an in-vitro environment, but lose their efficiency when moved into a clinical setting. However, for our technology, there’s still a very clear distinction between the different glucose concentrations in the body. This is the foundation of our technology. Our work now is focused on mapping all of these frequency signatures for different blood glucose levels. That is why scaling data collection is so important to complete R&D and algorithm development and ultimately achieve commercialization. We still have a lot of data collection and testing to be done, but we’ve achieved a lot in the last 12 months.
A timeline shows we have methodically approached the task and achieved significant improvement. As our data sets increase, we have experienced a clear improvement in the algorithm’s accuracy. This allowed us to go from an above 20% MARD to an MARD of 11. 27%. Documents covering each of these studies can be found on the research and validation page of our website. As we prepare to use Generation 2 with study participants in 2024, we estimate we will have tens of billions of observations to process. We are confident that our algorithm performance will improve as we get more raw data collected with our sensor. This is standard procedure in machine learning and will hopefully lead to a higher accuracy level or lower MARD figures. We will present the results of our current study at ATTD in Florence, Italy.
This study protocol allows us to test up to 100 participants with diabetes and pre-diabetes and uses venous blood as the reference point. This is the type of study we need to test real-world applications and to continue developing an algorithm for commercialization. This slide details the protocol for our current clinical study. For every participant, a test or what we call a data set takes at least three hours to complete. The data is subsequently processed and our glucose readings and accuracy compared to our reference device is determined. In general, the test starts with a participant resting his or her arm on our device for around 30 minutes, which allows the device to capture a baseline reading. After this baseline stage, the participant ingests 75 grams of glucose solution, which should spike his or her blood glucose levels.
As a reaction to the glucose increases, insulin should be released in the bloodstream, bringing blood glucose levels back to baseline values. Going through this glucose tolerance test allows our device to measure the participant’s changes in glucose values. Reference data through IV and a CGM is collected every five minutes. This step is critical to train our algorithms. But keep in mind that based on current setting parameters, our sensor completes a sweep every 22 seconds. This means the new glucose prediction is completed every 22 seconds. This also means while we get one reading from an IV sample or CGM, our technology delivers 13 different readings. This could be a very powerful help for diabetes management and in mitigating latency issues.
This is a very simple slide. I simply wanted to show you how we collect data outside the laboratory. This is a real life use case with that Gen 1 allowed us to implement. Users collect data either by resting their forearm or palm on the device. We have a committed and passionate team at Know Labs. We achieved a lot, and we learned a lot. Part of inventing is discovering new things. Our RF spectroscopy sensing technology was first disclosed in 2018. It sits atop sensor inventions that we first began working on in 2007. We’ve spent the last five years primarily focused on miniaturizing the technology and testing its capability, taking it from an exploratory concept to proof-of-concept, and lastly, to the feasibility stage. The miniaturization work has allowed us to reveal the Gen 1 prototype device and is the foundation for the Generation 2 prototype device.
The sensor being used in the Generation 2 device is the same one that was originally used in the exploratory studies and used in the Gen 1 device. So far, the validation has involved in-vitro testing and clinical testing with small cohorts, but it has allowed us to build the foundational knowledge to scale testing. We learned in more detail what needs to be accomplished before presenting our technology and devices to the FDA and before launching a medical device in the global marketplace. We learned there’s more work to do to conform with all the regulatory requirements, especially in topics like data accuracy across all glycemic ranges. More than 80% of our current data is still concentrated in the normal glucose range, which is 80 to 150 milligrams per deciliter.
We need to expand data collection to a wider range, including hypo and hyperglycemic scenarios to be successful in front of the FDA. These will also help us determine the intended use of our glucose monitoring devices in people with Type 1 and Type 2 diabetes or other subsegments. Patient physiological characteristics. We need to expand our data collection efforts to include a more diverse set of patients with different skin pigmentation, skin thickness, and the presence of other elements that could cause interference such as air, sweat, intense movement, and other substances, and environment and human factors. Our tests are currently conducted in a controlled laboratory environment. Ultimate regulatory approval requires us to have a deep understanding of real world environmental conditions, such as air pressure, temperature and humidity, other substances, noise, and human interaction with the technology, which may interfere with our signal and ultimately impact its accuracy.