Thanks for calling this out.
That was also my impression after having skimmed their paper: the only link to glucose monitoring is that the authors mention a few papers on the topic to motivate their research.
And looking at the papers they cite, I see little evidence that this approach could work in practice in the near future. Most of the citations [2, 15, 16] are to their own work, which did not look at glucose monitoring in the human body.
This is not my field of expertise, and maybe I am misunderstanding the papers. But it seems that there is little evidence that non-invasive glucose monitoring via measuring dielectric properties works reliably in practice. No in-the-wild studies, no investigation of potentially confounding factors.
Take for example citation 22 from the paper. A study where the authors propose a new antenna design. They seem to measure how the pancreas changes size during insulin production by monitoring its dielectric properties. IIUC, they look for a dip in the frequency spectrum caused by absorption of a certain frequency band.
But their measurements show an even larger effect when measuring on the thumb instead of the pancreas. This effect is not explained at all. (My guess: after having patients fast for 8-10 hours, giving them glucose will have an effect on the whole metabolism, resulting in higher blood flow, and that's what they measured).
Also, while they operate the antenna in the GHz range, they use a cheap USB soundcard (sampling rate 44.1 kHz) for capturing the signal. I did not understand this at all. They also repeatedly use the term "dielectric radiation". Seems to be a rather uncommon term?
The "machine learning algorithms" mentioned in the title seem to be a simple linear regression?
They claim an accuracy of ~90% and show some sample results. The complete study data is only available upon request, however.
[22] S.J. Jebasingh Kirubakaran, M. Anto Bennet, N.R. Shanker,
Non-Invasive antenna sensor based continuous glucose monitoring using pancreas dielectric radiation signal energy levels and machine learning algorithms,
Biomedical Signal Processing and Control, Volume 85, 2023, 105072,
https://doi.org/10.1016/j.bspc.2023.105072
I don’t have access to the full text, but I loved this part:
> Commercial CGM devices have certain drawbacks in diabetic measurement during daily activities such as food intake, sleeping, exercise and driving. The drawbacks are continuous radiations from devices
So they think a drawback of CGM is the (Bluetooth) radiation, and their alternative is to zap the pancreas with, um, magic dielectric radiation? Or magic radiation that results in “dielectric” backscatter?
I do find myself wondering whether a watch- or patch-sized object could get a usable NMR signal from glucose. Maybe a neodymium magnet and a very carefully shaped probe antenna to compensate for the horribly nonuniform magnetic field? Maybe an AC field with no permanent magnet at all? I found a reference suggesting that measuring glucose in blood outside the body by 1T NMR is doable but marginal, so this may be a lost cause.
The paper is full text, fyi. You won't get any extra info about actual glucose measurements. The paper is all about their device idea engineering. The press release dose purport to show a pic of a supposed sensor and a vague claim of clinical trials.
This is not my field of expertise, and maybe I am misunderstanding the papers. But it seems that there is little evidence that non-invasive glucose monitoring via measuring dielectric properties works reliably in practice. No in-the-wild studies, no investigation of potentially confounding factors.
Take for example citation 22 from the paper. A study where the authors propose a new antenna design. They seem to measure how the pancreas changes size during insulin production by monitoring its dielectric properties. IIUC, they look for a dip in the frequency spectrum caused by absorption of a certain frequency band.
But their measurements show an even larger effect when measuring on the thumb instead of the pancreas. This effect is not explained at all. (My guess: after having patients fast for 8-10 hours, giving them glucose will have an effect on the whole metabolism, resulting in higher blood flow, and that's what they measured).
Also, while they operate the antenna in the GHz range, they use a cheap USB soundcard (sampling rate 44.1 kHz) for capturing the signal. I did not understand this at all. They also repeatedly use the term "dielectric radiation". Seems to be a rather uncommon term?
The "machine learning algorithms" mentioned in the title seem to be a simple linear regression? They claim an accuracy of ~90% and show some sample results. The complete study data is only available upon request, however.
[22] S.J. Jebasingh Kirubakaran, M. Anto Bennet, N.R. Shanker, Non-Invasive antenna sensor based continuous glucose monitoring using pancreas dielectric radiation signal energy levels and machine learning algorithms, Biomedical Signal Processing and Control, Volume 85, 2023, 105072, https://doi.org/10.1016/j.bspc.2023.105072