 [AfA2026-PhysicalAI] - Smart Rural Triage Station

Grove - Servo

Modulino™ Buzzer

Arduino-CLI

Modulino™ Thermo

Arduino App Lab

Modulino™ Knob

Arduino IDE

Modulino™ Movement

Grove - OLED Display 0.96"

<html><body><h3>Smart Rural Triage Station</h3>Transform your Arduino UNO Q into an AI-powered medical screening device that helps rural healthcare workers identify heart and lung abnormalities in real-time, with complete privacy and zero internet dependency. <h3>Components and Supplies </h3><figure><img src="https://projects.arduinocontent.cc/f5272f32-514a-4308-918d-65c75b16d681.png"/><figcaption></figcaption></figure> <h3>Tools and Machines</h3><ol><li>3D Printer (Optional - for enclosure) </li><li>Soldering Iron (Optional) </li></ol><h3>Apps and Platforms</h3><ol><li>Arduino App Lab </li><li>TensorFlow Lite </li><li>Python 3.x </li><li>Flask Web Framework </li></ol><h3>Project Description</h3>Imagine a rural health clinic where a single nurse serves hundreds of patients without access to specialist doctors or advanced diagnostic equipment. This project transforms the Arduino UNO Q into a Smart Rural Triage Station an AI-powered screening device that listens to heart and lung sounds to detect abnormalities instantly. By combining a contact microphone with TensorFlow Lite machine learning models, the system processes audio in real-time directly on the device. When a patient places the microphone on their chest, the AI analyzes the sounds and provides immediate feedback through servo animations, buzzer alerts, and an OLED display showing the triage result. This application runs entirely on the Arduino UNO Q . The AI inference happens locally on the board, ensuring zero latency and complete privacy no patient audio is ever stored or transmitted. <h3>Build the Triage Station</h3>The Arduino UNO Q acts as the central hub, running both the AI processing on its Linux core and the real-time hardware control on its MCU core. <h3>Hardware Lineup</h3> <figure><img src="https://projects.arduinocontent.cc/ad02ad58-7561-4ada-9f27-ef11d262f41c.png"/><figcaption></figcaption></figure><h3>The Dashboard</h3>The application includes a web-based dashboard hosted directly on the Arduino UNO Q, accessible from any device on the same local network at <code class="code">http://192.168.7.2:5000 </code>. Dashboard Features <ol><li>Mode Selection : View current mode (HEART/LUNG/CALIBRATION) based on knob position </li><li>Live Sensor Data : Real-time temperature, audio level, and movement detection </li><li>Examination Control : Start/Reset examination with one click </li><li>Triage Results : Clear visual indication of risk level (LOW/MEDIUM/HIGH) </li><li>Connection Status : Shows which serial port is connected to MCU </li></ol><h3>Risk Level Indicators</h3> <figure><img src="https://projects.arduinocontent.cc/2f4f406a-e9ed-411e-9e1b-45a9515b983a.png"/><figcaption></figcaption></figure> <h3>The AI Models</h3>The heart and lung classification is powered by TensorFlow Lite models optimized for edge deployment. <h3>Heart Sound Classification</h3><ol><li>Input : 64x128 mel-spectrogram </li><li>Architecture : CNN (3 conv blocks + dense layers) </li><li>Output Classes : Normal, Murmur, Extrasystole </li><li>Accuracy : &gt;85% </li><li>Model Size : &lt;2MB quantized </li></ol><h3>Lung Sound Classification</h3><ol><li>Input : 64x128 mel-spectrogram </li><li>Architecture : CNN (3 conv blocks + dense layers) </li><li>Output Classes : Normal, Wheeze, Crackle, Both </li><li>Accuracy : &gt;96% </li><li>Model Size : &lt;2MB quantized </li></ol>The models are trained on the PhysioNet and ICBHI medical datasets, then converted to TensorFlowLite format for efficient on-device inference. <h3>Software Architecture Overview</h3>The software leverages the Arduino UNO Q's dual-processor architecture to separate high-level AI processing from real-time hardware control. <pre>Arduino UNO Q </pre><pre>├── Linux Side (QRB2210) - Python </pre><pre>│ ├── Audio Processing &amp; Feature Extraction </pre><pre>│ ├── TensorFlow Lite Inference </pre><pre>│ ├── Flask Web Dashboard </pre><pre>│ └── Serial Communication to MCU </pre><pre>│ </pre><pre>└── MCU Side (STM32U585) - Arduino Sketch </pre><pre>	 ├── Real-time Sensor Polling (Modulino) </pre><pre>	 ├── OLED Display Updates </pre><pre>	 ├── Servo Motor Control </pre><pre>	 └── Buzzer &amp; LED Control </pre><h3>Communication</h3>The two cores communicate via Serial Bridge at 115200 baud using JSON messages. The MCU sends sensor data, and the Linux side sends control commands. <h3>A Quick Look into the Code</h3>The Linux Side: Running the Examination On the Linux side ( <code class="code">app.py </code>), the Flask server manages the examination flow: <pre>def run_exam(): </pre><pre>	 system_state["mode"] = "EXAMINING" </pre><pre>	 send_mcu_command({"progress": 0, "result": 90}) </pre><pre>	 </pre><pre>	 # Progress loop with movement detection </pre><pre>	 for i in range(0, 101, 2): </pre><pre>	 if system_state["movement"] == 1: </pre><pre>	 send_mcu_command({"buzzer": 2000}) # Alert if moving </pre><pre>	 system_state["diagnosis"] = f"Recording... {i}%" </pre><pre>	 send_mcu_command({"progress": int(i * 1.8)}) </pre><pre>	 time.sleep(0.1) </pre><pre>	 # Analyze and determine risk </pre><pre>	 score = calculate_risk_score() </pre><pre>	 if score &gt;= 2: </pre><pre>	 system_state["risk_level"] = "HIGH" </pre><pre>	 send_mcu_command({"result": 180, "buzzer": 1000}) </pre><pre>	 # ... handle other levels </pre> The MCU Side: Display &amp; Actuators On the MCU side (main.ino), the firmware handles the UI and hardware: <pre>void updateDisplay() { </pre><pre>	 display.clearDisplay(); </pre><pre>	 </pre><pre>	 // Header Bar (Inverted text for mode) </pre><pre>	 display.setTextColor(SSD1306_BLACK, SSD1306_WHITE); </pre><pre>	 String modeStr = (currentSensorData.knobMode == 0) ? " HEART MODE " : </pre><pre>	 (currentSensorData.knobMode == 1) ? " LUNG MODE " : " CALIBRATION"; </pre><pre>	 display.println(modeStr); </pre><pre>	 </pre><pre>	 if (currentState == STATE_EXAMINING) { </pre><pre>	 display.setTextSize(2); </pre><pre>	 display.println("SCANNING"); </pre><pre>	 // Animated progress bar </pre><pre>	 display.drawRect(10, 45, 108, 10, SSD1306_WHITE); </pre><pre>	 int fill = (millis() / 250) % 20; </pre><pre>	 display.fillRect(12, 47, fill * 5, 6, SSD1306_WHITE); </pre><pre>	 } </pre><pre>	 </pre><pre>	 display.display(); </pre><pre>} </pre><h3>JSON Communication Protocol</h3>MCU → Linux (Sensor Data): <pre>{ </pre><pre>	 "knob": 450, </pre><pre>	 "mode": 1, </pre><pre>	 "temp": 36.5, </pre><pre>	 "movement": 0, </pre><pre>	 "mic": 512 </pre><pre>} </pre> Linux → MCU (Commands): <pre>{ </pre><pre>	 "progress": 90, </pre><pre>	 "result": 180, </pre><pre>	 "buzzer": 1000, </pre><pre>	 "state": "EXAMINING" </pre><pre>} </pre><h3>Privacy &amp; Security</h3><ol><li>No Audio Storage : Sounds are processed in real-time and never saved </li><li>Offline Operation : No internet required, no cloud dependencies </li><li>Local Processing : All AI inference happens on-device </li><li>Anonymized Logs : Only timestamps and results (no patient data) </li><li>Secure Updates : Model updates via USB, not network </li></ol><h3>Clinical Impact</h3><ol><li>Early Detection : Identifies abnormalities missed by basic stethoscopes </li><li>Standardization : Consistent screening across different skill levels </li><li>Referral Optimization : Reduces unnecessary referrals, catches critical cases </li><li>Training Support : Helps build local healthcare capacity </li><li>Data Sovereignty : Keeps patient data in local control </li></ol><h3>Code</h3>GitHub Repository <a href="https://github.com/Raja-89/Arduino-Triage.git">Raja-89/Arduino-Triage </a>An AI-powered medical screening device for Arduino UNO Q that uses TensorFlow Lite models and multi-sensor fusion to provide instant triage decisions for rural healthcare. <h3>Schematics</h3><blockquote>Pin Mapping (STM32U585 MCU Side) </blockquote><h3></h3><figure><img src="https://projects.arduinocontent.cc/98959cfc-04cf-4fe6-ac60-dd70178e6bd6.png"/><figcaption></figcaption></figure> <h3>Quick Start</h3>1. Clone the repository <pre>git clone https://github.com/Raja-89/Arduino-Triage.git </pre><pre>cd Arduino-Triage </pre>2. Install Python dependencies <pre>pip install -r requirements.txt </pre>3. Deploy to Arduino UNO Q <pre>scp -r * root@192.168.7.2:/opt/triage-station/ </pre>4. Start the application <pre>ssh root@192.168.7.2 "cd /opt/triage-station &amp;&amp; python3 final_deployment/app.py" </pre>5. Access the dashboard Open http://192.168.7.2:5000 in your browser <h3>License</h3><a href="https://github.com/Raja-89/Arduino-Triage/blob/main/LICENSE.txt">Mozilla Public License 2.0 (MPL-2.0) </a><h3>Acknowledgments</h3><ol><li>Arduino x Qualcomm for the AI for All Hackathon platform </li><li>PhysioNet and ICBHI for open medical datasets </li><li>TensorFlow Lite team for edge AI framework </li><li>Rural healthcare workers who inspired this solution </li></ol> </body></html>

Smart Rural Triage Solution

Links