This repository documents an open hardware/software experiment in noninvasive blood-glucose sensing using optical measurements. It includes embedded data acquisition, electronics, mechanical designs, sensor logs, analysis scripts, and the design history of multiple prototype concepts.
The first prototype was built and tested rather than stopping at a concept. The most important result is also a negative one: Glucosan 1 did not demonstrate a useful correlation between its optical measurements and reference blood-glucose values. The repository preserves the hardware, software, raw data, analysis, observed failure modes, and follow-on design work so the experiment can be inspected rather than presenting an unsuccessful hypothesis as a successful product.
Not a medical device. The designs and data in this repository have not been evaluated or approved by the FDA and are not intended to diagnose, treat, cure, prevent, or manage any disease or medical condition.
The project spans several disciplines:
- embedded C/C++ sensor acquisition and data logging
- optical emitter/detector selection
- analog and digital electronics
- schematic and electronics development
- 3D-printed mechanical fixtures
- repeatable sensor logging
- MATLAB/post-processing analysis
- experimental iteration based on measured failure modes
Repository areas include:
src/— embedded software/library sourceelectrical/— electrical schematics and follow-on design workmechanical/— mechanical designs and prototype imageslog_data/— experimental sensor/reference measurements and analysisparts_list/— component informationexamples/— development and test-harness examples
The repository includes pseudonymized measurements from two participants, identified in the dataset only by participant number. The published measurements include reference blood-glucose and SpO2 values used to compare the experimental optical sensor output with conventional measurements.
The participants knowingly consented to these measurements being released publicly as part of this open-source experiment. No participant names are encoded in the published measurement index.
The data should still be treated as experimental human physiological data rather than as a validated medical dataset.
Glucosan 1 used infrared/visible optical sensing to investigate whether a low-cost portable device could detect a useful signal correlated with blood glucose.
| Item | Glucosan 1 |
|---|---|
| Method | IR Spectrometry |
| Core Platform | Adafruit Feather M0 Adalogger |
| Emitters | 1 x 617nm, 1 x 940nm |
| Reflectance Detectors | Phototransistors 1 x 630nm, 1 x 940nm (Si) |
| Transmittance Detectors | Phototransistors 1 x 630nm, 1 x 940nm (Si) |
| ADC | bypass caps, 12 bit resolution |
| Display | 128x64 OLED |
| SD Card | CSV or Binary Logging |
| Wireless | None |
| IMU | None |
| Temperature | None |
| Published data | more than 150 log sessions across two participants |
| Analysis | post-processing, correlation analysis, filtering, and polynomial fitting |
| Observed correlation | weak to none |
| Listed prototype component cost | $48.33 |
The embedded acquisition code configures 12-bit ADC reads, cycles the 617 nm and 940 nm emitters, samples four optical channels, and supports CSV, binary, and serial logging. Binary records include a CRC-8 for integrity checking.
The first design produced weak to no useful correlation with reference glucose measurements. Larger datasets also exposed an apparent signal-integrity problem: multiple sensor channels sometimes showed very similar photoplethysmogram waveforms with small offsets or shifts, suggesting coupling or another hardware failure rather than independent sensor information.
Several experimental limitations became apparent:
The flexible finger clamp improved light sealing but allowed emitter/sensor geometry to change between readings. Small position changes could produce signal changes larger than the effect the experiment was attempting to measure.
A better design should constrain optical geometry much more tightly or measure relative position so it can be included in the model.
The first prototype did not independently measure motion or vibration. Movement can dominate small optical effects. Later concepts therefore include inertial sensing so unstable readings can be identified or rejected.
Measurements were taken in varying environments without strong control of ambient conditions. Better experimental controls are needed to distinguish sensor behavior from environmental variation.
Review of the larger dataset revealed apparent duplicated/coupled waveform behavior across channels in some later logs. That data is retained, but those logs should not be treated as independent clean sensor measurements.
These problems are why the repository does not claim that Glucosan 1 measured blood glucose successfully.
The second design was intended to address limitations observed in the first prototype and explore a wider spectral range.
| Item | Glucosan 2 concept |
|---|---|
| Method | IR Spectrometry |
| Core Platform | Adafruit Feather RP2040 Adalogger |
| Emitters | 1 x 670nm, 1 x 850nm, 1 x 950nm, 1 x 1300nm, 1 x 1550nm |
| Reflectance Detectors | None |
| Transmittance Detectors | Photodiodes, 1 x 950nm (Si), 1 x 1650nm (InGaAs) |
| ADC | analog front end IC, 14 bit resolution |
| Display | 128x64 OLED |
| SD Card | CSV or Binary Logging |
| Wireless | None |
| IMU | 2 x Accel, 2 x Gyro |
| Temperature | 1 x Temp |
| Current listed component cost | $190.33, excluding as-needed consumables |
| Status | design in progress / not validated |
The second concept should be understood as follow-on engineering work, not as a validated sensor. The repository currently contains detector-side schematic work and a component list rather than a completed Glucosan 2 build.
The Glucosan 1 dataset is under log_data/Glucosan1/:
data/Index.csv— maps groups of sensor logs to participant number and pre/post reference measurementsInitialLogData.csv/.mat— large sensor datasetUserTwoLogData.csv— additional dataset used to broaden the observed glucose rangeanalysis/EstimateBG.m— filtering, polynomial fitting, correlation, MARD, and RMSE explorationanalysis/ResultsOverview.m— correlation overview across sensor/finger configurationsanalysis/Clarke_EGA.png— retained analysis output
The raw files are intentionally retained so analysis can be checked against the underlying measurements.
The purpose of publishing this work is not to claim a medical breakthrough. It is to make the engineering experiment inspectable:
- publish enough design information to reproduce the hardware
- collect reference and experimental measurements
- test whether the hypothesized relationship appears in the data
- document failure modes when it does not
- use those failures to drive the next hardware iteration
That cycle is more useful than reporting attractive accuracy metrics from an inadequately controlled or narrow dataset.
The repository was originally created around an open design competition intended to encourage reproducible noninvasive-glucose experiments. The competition did not receive entries, so the repository evolved into a record of the prototype designs and experimental results.
A $1,750 project fund was originally set aside to encourage work in this area. For questions about the historical project or the designs, contact noninvasiveglucometer@gmail.com.
An early build/progress video is available here:
- Hina A, Saadeh W. Noninvasive Blood Glucose Monitoring Systems Using Near-Infrared Technology—A Review. Sensors. 2022;22(13):4855. doi:10.3390/s22134855.
- Javid B, Fotouhi-Ghazvini F, Zakeri FS. Noninvasive Optical Diagnostic Techniques for Mobile Blood Glucose and Bilirubin Monitoring. J Med Signals Sens. 2018;8(3):125-139.
- Meter G. Open Source Non-Invasive Glucose Meter. YouTube, 2015.
- Bhuyan M. Design and Implementation of an NIR-Technique Based Non-Invasive Glucometer using Microcontroller. 2020.
- Oxford Instruments. Measuring glucose concentration: NIR absorption spectroscopy.
- Naresh M, Nagaraju VS, Kollem S, Kumar J, Peddakrishna S. Non-invasive glucose prediction and classification using NIR technology with machine learning. Heliyon. 2024;10(7):e28720.
See AUTHORS for the contributors credited by this repository.
See LICENSE.