PregnaCare

A wearable, real-time pregnancy monitoring system that uses AI to flag abnormalities in maternal vitals before they become emergencies.

The Problem

Maternal mortality is a crisis that's stalled rather than improved. A woman dies every two minutes from preventable causes related to pregnancy and childbirth, and between 2015 and 2020 there was barely any progress on the global maternal mortality rate, the average annual rate of reduction over that period was only 0.04%. The 2030 target is 70 maternal deaths per 100,000 live births, which requires an 11% annual reduction from 2020 onward; at the current trajectory, that target is on track to be missed. A large share of these deaths are preventable in principle, the gap is in catching warning signs early enough to act on them, not in a lack of medical knowledge about what those warning signs are.

The Solution

PregnaCare is built to close that gap by making vitals monitoring continuous and low-friction rather than something that only happens during a hospital visit. It tracks maternal heart rate, respiratory rate, body temperature, and blood sugar (estimated via regression from respiratory rate and heart rate), then feeds that data into an AI model trained to flag abnormalities during pregnancy as they emerge. Results surface on a website with early predictions rather than waiting for a clinical visit to interpret them, and a Gemini AI-powered chatbot gives users a way to ask questions about their results directly.

How It Works

The core loop is a real-time monitoring cycle that repeats every 10 seconds. Sensors on the device read the mother's physiological signals (heart rate and SpO2 from a MAX30102 module, body temperature from a DS18B20 sensor) and send that data to a MongoDB database. From there, an AI model ingests the readings and classifies them into one of three condition classes. If the result indicates an abnormality, the system sends a warning and suggested next steps to both the mother and her doctor; if the reading is normal, the cycle simply continues to the next sampling window.

Hardware & Product Design

PregnaCare's form factor is a fingerless glove, chosen specifically so the device is wearable rather than something that has to be set up like clinical equipment. The MAX30102 sensor (heart rate and SpO2) sits at the fingertip to read pulse oximetry through light absorption in the blood vessels, while the DS18B20 temperature sensor reads body temperature with its built-in 12-bit ADC. Both sensors connect to an ESP32 microcontroller, which handles the WiFi connection and pushes sensor readings out over HTTP as JSON. Current builds are breadboard prototypes, sensors and wiring integrated into the glove and connected to the ESP32 for testing, ahead of a more finished enclosure.

Why PregnaCare

The pitch comes down to four things. It's portable: because the device is worn as a glove, it can be used anywhere and anytime rather than requiring a hospital visit. It's real-time: results show up on the website as soon as a reading completes, with no waiting period. It's effective and efficient, designed to be simple enough to use repeatedly without friction. And it's cost-effective, since routine monitoring no longer requires a trip to the hospital every time.

Technology & Dataset

The AI layer is built in Python, using multiclass classification to detect pregnancy abnormalities from the incoming sensor data, with the results website built in Streamlit. On the hardware side, an ESP32 (WiFi- and Bluetooth-enabled, built for wearable electronics) handles sensor communication and network transmission, paired with a MAX30102 pulse oximetry/heart-rate module and a DS18B20 temperature sensor. The classification model was trained on the Maternal Health Risk dataset from Kaggle, with methodology informed by a published reference paper on maternal risk classification (EMITTER, PENS).

Future Work

A few directions are already mapped out. Better sensors would mean better input data, more readable output, and a more reliable signal overall. Adding diastolic and systolic blood pressure monitoring is the next priority for improving the AI's prediction accuracy, since the current model only works from heart rate, SpO2, and temperature. And before any of that, the team wants to measure the deviation between the glove's sensor readings and medically approved devices, so the data itself can be calibrated and trusted.