DEEP LEARNING EMPOWERED CONTINUOUS BLOOD GLUCOSE MONITORING FOR DIABETES BASED ON PPG SIGNAL
Diabetes is a life-long and life threatening disease which can lead to severe complications and affect all of the organs that are vital to the body. Today, diabetes is a very common disease and is increasing day by day across the globe, which can be alarming. Monitoring of glucose is very critical t
2025-06-28 16:26:05 - Adil Khan
DEEP LEARNING EMPOWERED CONTINUOUS BLOOD GLUCOSE MONITORING FOR DIABETES BASED ON PPG SIGNAL
Project Area of Specialization Software EngineeringProject SummaryDiabetes is a life-long and life threatening disease which can lead to severe complications and affect all of the organs that are vital to the body. Today, diabetes is a very common disease and is increasing day by day across the globe, which can be alarming. Monitoring of glucose is very critical to diabetes management. The techniques and methods which are used for ever increasing diabetic people is invasive and is often painful, also they are time consuming and seems to cost a lot of budget. The non-invasive way of monitoring glucose overcomes these constraints, for which this topic of research proves to be an exciting and pleading topic for market and companies. Photoplethysmography (PPG) is considered to be one of the most popular and favorable technologies in the last decade for monitoring the physiological conditions of a patient. PPG has been largely used and is accepted by market because of its convenience and capacity to record continuous readings. It is applied to personal portable devices and pulse oximetry. Additionally, this signal provides information about both the respiratory and cardiovascular systems. So we came with an idea of a deep learning based approach for an effective continuous blood glucose monitoring for management of diabetes based on PPG signals.
Project ObjectivesA diabetic patient has to check their blood glucose level on daily basis often two times a day. Previously, painful invasive methods were used for this purpose but this has been replaced by non invasive ways. Modern problems require modern solutions so, this paper presents a non invasive way i.e. Photoplethysmography which uses infrared rays to measure the blood glucose level of diabetic patient after some interval of time. This has no contact with the skin. Here different deep learning algorithms and methodologies will be implemented in order to train the model. The main objectives include:
-
To overcome the unease of the diabetic patient they might have using invasive method of pricking a finger.
-
To check the blood glucose level of a diabetic patient continuously.
-
Emphasis on the regular monitoring and measurement of blood glucose level.
-
Checks the blood glucose level without any interaction with the skin.
A flowchart has been created to explain the suggested method for noninvasively estimating blood glucose levels using a smartphone video. Video data was collected from the individuals' fingertips and transformed into the appropriate PPG waveform. The distortion and movement artefacts in this signal were removed throughout the preprocessing processes. The precompiled signals were examined for relevant features. We created a deep artificial neural network approach to determine diabetes from arterial signals known as PPG (photoplethysmography), which can be monitored with existing equipment in smart devices such as smartphones and fitness activity trackers. Many smart devices and clinics utilize the same technology to assess heart rate. From data recorded with a basic smartphone camera, we constructed a deep neural network learning algorithm that can detect persons with diabetes. To check the accuracy, we will be testing and training our model with two deep learning models. To detect frequent diabetes from cellphone PPG readings, we created a 39-layer convolutional deep neural network (DNN). The DNN receives only a PPG signal as input and returns a score between 0 and 1, with excessive high scores reflecting a higher chance of diabetes. PPG waveforms of a duration of less than 21 seconds are accepted by the DNN, but long or short durations can be buffered or chopped. The Azumio Immediate Pulse Rate mobile application was used to capture PPG waveforms by putting the tip of index finger on the camera. Pulsatile blood capacity change is deduced from changes in reflecting light intensity captured by the cellphone camera. Either at 100 or 120 Hz, PPG waveforms were recorded.
Finally, RFR—a methodology based on employing many decision trees to forecast a continuous value—was employed to employ ensemble techniques. Decision Tree Regression (DTR) entails gradually building a decision tree from smaller portions of the training data and can be used to map nonlinear relationships relatively well. To produce an ensemble approximation of the anticipated value, the outputs from numerous decision trees are blended together. RFR has been shown to be effective in predicting biological parameters from ECG or PPG data in previous research, typically surpassing other regression techniques.
-
To overcome the unease of the diabetic patient they might have using invasive method of pricking a finger.
-
To check the blood glucose level of a diabetic patient continuously.
-
Emphasis on the regular monitoring and measurement of blood glucose level.
-
Checks the blood glucose level without any interaction with the skin.
We will be developing a smartphone application which will take measurements from the subject's fingertip in the form of PPG signals. ppg sensors would be required for the measurement of PPG signals.
Final Deliverable of the Project Software SystemCore Industry HealthOther Industries IT , Medical Core Technology Wearables and ImplantablesOther Technologies OthersSustainable Development Goals Good Health and Well-Being for PeopleRequired Resources| Item Name | Type | No. of Units | Per Unit Cost (in Rs) | Total (in Rs) |
|---|---|---|---|---|
| Total in (Rs) | 20000 | |||
| PPG Sensors | Equipment | 4 | 4000 | 16000 |
| Stationery | Miscellaneous | 2 | 2000 | 4000 |