Biomedical Projects in Bangalore for ECE

A single-lead ECG acquisition system uses the AD8232 analog front-end IC to amplify and filter the cardiac signal, which is sampled at 500 Hz by an Arduino Uno's ADC. The digitised ECG is transmitted over serial to a Python script that applies a Pan-Tompkins QRS detection algorithm to identify R-peaks and compute heart rate. Arrhythmia classification (normal sinus rhythm, bradycardia, tachycardia, atrial fibrillation) is performed using RR-interval analysis and a rule-based classifier. The system achieves QRS detection sensitivity of 98.2% and specificity of 97.8% on the MIT-BIH Arrhythmia Database.

A MAX30102 pulse oximeter and heart rate sensor module interfaces with an ESP32 over I2C. The ESP32 runs a peak detection algorithm on the red and infrared PPG signals to compute SpO₂ using the ratio-of-ratios method and heart rate from peak-to-peak intervals. Readings are displayed on a 0.96-inch OLED and transmitted over Bluetooth Low Energy to a smartphone app built with MIT App Inventor. The app logs readings with timestamps and plots a 60-second trend graph. Validation against a certified pulse oximeter shows a mean SpO₂ error of ±1.5% and heart rate error of ±2 bpm.

A two-class motor imagery BCI is implemented in MATLAB using EEG data from the BCI Competition IV Dataset 2a. Common Spatial Pattern (CSP) filters extract spatial features from 22-channel EEG recordings during left-hand and right-hand motor imagery tasks. Power spectral density features in the mu (8–12 Hz) and beta (13–30 Hz) bands are computed from CSP-filtered signals. A Linear Discriminant Analysis (LDA) classifier achieves a mean classification accuracy of 76.3% across 9 subjects, consistent with published benchmarks. The project demonstrates EEG signal processing, CSP feature extraction, and BCI classification methodology.

Surface EMG signals from a Myo armband are streamed over Bluetooth to a Raspberry Pi that classifies five hand gestures (open, close, pinch, point, thumbs-up) using a k-NN classifier trained on 500 samples per gesture per user. Classified gestures are sent over serial to an Arduino Mega controlling five servo motors in a 3D-printed prosthetic hand. Classification latency is 45 ms, enabling natural-feeling control. The system achieves 91% gesture recognition accuracy in a 10-fold cross-validation. The project demonstrates EMG signal acquisition, gesture classification, and servo-based prosthetic actuation.

A near-infrared spectroscopy system uses three NIR LEDs (850 nm, 940 nm, 1050 nm) and a TSL2591 photodiode to measure the optical absorption of the fingertip at three wavelengths. An Arduino Mega samples the photodiode output and transmits absorbance values to a MATLAB script that applies a partial least squares (PLS) regression model trained on 80 subjects to estimate blood glucose concentration. The model achieves a mean absolute relative difference (MARD) of 14.2% against a reference glucometer, placing it in the Clarke Error Grid Zone A+B for 89% of predictions. The project demonstrates optical biosensing and chemometric modelling.

A MATLAB image processing pipeline analyses retinal fundus images from the DRIVE dataset to detect signs of diabetic retinopathy. The pipeline includes green-channel extraction, CLAHE contrast enhancement, blood vessel segmentation using a matched filter, and microaneurysm detection using morphological top-hat transform. A rule-based grading system classifies images as no DR, mild, moderate, or severe based on lesion counts. The system achieves a vessel segmentation sensitivity of 0.72 and specificity of 0.97, consistent with published results. The project demonstrates medical image analysis and ophthalmological screening algorithm design.

An MPU6050 IMU worn on the waist samples acceleration and gyroscope data at 100 Hz. An Arduino Nano runs a threshold-based fall detection algorithm that identifies the characteristic free-fall phase (acceleration < 0.6 g for > 80 ms) followed by an impact phase (acceleration > 3 g) and a post-fall inactivity phase (tilt angle > 60° for > 5 seconds). On fall detection, a SIM800L GSM module sends an SMS with the GPS coordinates (from a NEO-6M module) to the caregiver's number within 10 seconds. The system achieves 94% sensitivity and 96% specificity on a dataset of 200 simulated falls and 300 activities of daily living.

A continuous-wave Doppler ultrasound circuit operating at 2 MHz detects fetal heart motion using a transmit transducer and a receive transducer held against the maternal abdomen. The Doppler shift signal is amplified, bandpass filtered (100–600 Hz), and sampled by an Arduino Due at 10 kHz. A zero-crossing detector algorithm extracts the fetal heart rate from the Doppler waveform. The computed FHR is displayed on an LCD and compared against a reference cardiotocograph. The system achieves FHR measurement accuracy of ±3 bpm in the 120–160 bpm range. The project demonstrates ultrasound electronics, Doppler signal processing, and obstetric monitoring.