Know Labs, Inc. (NYSE American: KNW) today announced the results of a new study titled, “Algorithm Refinement in the Non-Invasive Detection of Blood Glucose Using Know Labs’ Bio-RFID Technology.” The study demonstrates that algorithm optimization using a light gradient-boosting machine (lightGBM) machine learning model improved the accuracy of Know Labs’ Bio-RFID™ sensor technology at quantifying blood glucose, demonstrating an overall Mean Absolute Relative Difference (MARD) of 12.9% – which is within the range of FDA-cleared blood glucose monitoring devices. Bio-RFID is a novel technology platform that uses electromagnetic energy in the form of radio waves to non-invasively capture molecular signatures and convert them into meaningful information.
Like all previous Know Labs clinical studies, this study was designed to assess the ability of the Bio-RFID sensor to non-invasively and continuously quantify blood glucose, using the Dexcom G6® continuous glucose monitor (CGM) as a proxy for the measurement of blood glucose. Unique from previous studies, Know Labs tested new data science techniques and trained a lightGBM model to predict blood glucose using 1,555 observations – or reference device values – from over 130 hours of data collection across five healthy participants. Using this model, Know Labs was able to predict blood glucose in the test set – the dataset that provides a blind evaluation of model performance – with a MARD of 12.7% in the normoglycemic range and 14.0% in the hyperglycemic range.
“This is a transformational time for Know Labs. We are constantly uncovering new learnings in our research, and in this case found that the lightGBM model is well-suited for these early datasets given the amount of data available,” said Steve Kent, Chief Product Officer at Know Labs. “In our previous technical feasibility study we utilized a neural network, and as is best practice when developing algorithms, our data science team is constantly refining our machine learning models to understand and optimize system performance and accuracy. This positive development is another critical step in our data collection, algorithm refinement, and technical development.”
This study, which was peer-reviewed by Know Labs’ Scientific Advisory Board, builds upon recently released peer-reviewed research. In February, Know Labs published a proof-of-concept study that examined the efficacy of the Bio-RFID sensor using one participant, resulting in a MARD of 19.3%. Earlier this month, Know Labs also released study results validating the technical feasibility of Bio-RFID using a neural network (NN) model to predict readings of the Dexcom G6® as a proxy for blood glucose, which resulted in a MARD of 20.6%. The techniques used to analyze the data differed from previous analyses among the same (N=5) participant population, including: approach to feature reduction, stratification of the data by glycemic range and only from the arm corresponding to the reference device, and a different machine learning model. The improved accuracy as measured by a MARD of 12.9% achieved in
this study is comparable to other independently validated MARD values reported for today’s FDA-cleared, commercially available CGM devices.
“A MARD of 12.9% at this stage in our development is a truly remarkable feat. Our whole team is thrilled by these findings and the improved accuracy of our Bio-RFID technology as we continue to refine our approach,” said Ron Erickson, CEO and Chairman at Know Labs. “Our goal with these ongoing clinical studies is to develop large volumes of data to enable further model development, which is a critical step in our goal to bring the first FDA-cleared non-invasive glucose monitoring device to the market so that millions of people can manage their diabetes more efficiently.”
The full manuscript of this study will be submitted to a peer-review journal as Know Labs continues to prioritize external validation of the Bio-RFID technology. To view Know Labs’ growing body of peer-reviewed research, visit www.knowlabs.co/research-and-validation.
