Artificial Intelligence for Congenital Heart Disease Screening
A study on large-scale CHD screening using AI-based heart sound analysis, transmitting data via STM32 and W5100 Ethernet for deep learning classification.
0
Project description
π Overview
This project is a study on large-scale screening for congenital heart disease (CHD) using an AI-based heart sound analysis system. Heart sound data collected with an electronic stethoscope are transmitted from an embedded system based on STM32 + Ethernet (W5100) to an upper-level computing platform, where diseases are automatically classified using deep learning models (RNN/CNN). The system aims to provide a low-cost, highly scalable diagnostic support solution that can be deployed even in regions with limited medical infrastructure.
π Features
- AI-Based Heart Sound Diagnosis
- Automatic classification of major CHD types such as ASD, VSD, and PDA
- Achieves up to approximately 90% accuracy using an RNN-based model
- Large-Scale Screening Validation
- Real-world field screening conducted on 518,258 individuals
- Low-Cost Hardware Configuration
- STM32 microcontroller combined with an Ethernet network
- Field-Oriented Design
- Portable, low-power operation with real-time data acquisition and transmission
π System Architecture


The system adopts a clearly separated two-layer architecture, consisting of a lower embedded acquisition system (Lower Computer) and an upper processing and visualization system (Upper Computer).
This structure is designed to appropriately distribute responsibilitiesβensuring stable medical signal acquisition, reliable data transmission, and high-performance AI processingβaccording to each layerβs purpose.
Lower Computer
- Sensor Layer
- Heart Sound Sensor
- EMS Sensor
β Collect heart sounds and other biosignals from the human body in analog form
- Signal Preprocessing and Conversion
- Sensor signals are conditioned using a ThinkLabs module
- Multi-channel analog signals are converted into digital data using the AD7606 ADC
- Main Control MCU
- STM32F103ZET6
- Handles ADC data acquisition and packet construction
- Manages data processing for Ethernet transmission
- Network Interface
- W5100 Ethernet Controller
- Connected to the STM32 via SPI
- Performs TCP/IP communication at the hardware level
- Reliably transmits collected medical data to the upper system
Upper Computer
- APSoC-Based Processing Platform
- ZedBoard (Zynq APSoC)
- Composed of a Processing System (PS) and Programmable Logic (PL)
- Data Reception and Buffering
- Receives data from the lower system via Ethernet
- Stores and shares data in DDR3 memory
- AI Processing and Application Execution
- Runs on a Linux (Linaro)βbased environment
- Executes Qt-based UI and data acquisition/analysis applications
- Performs heart disease classification and diagnostic assistance using AI models
- Display Output
- ADV7511 HDMI transmitter
- Connects to LCD drivers and HDMI displays
- Visualizes real-time waveforms and analysis results
- Internal Interfaces
- Data exchange between PS and PL via AXI interconnects
This architecture clearly separates medical signal acquisition, transmission, and analysis, ensuring scalability and system reliability.
π Role and Application of the W5100
In this study, the W5100 Ethernet controller plays a core networking role.
- Handles wired Ethernet data transmission on the STM32-based embedded board
- Reliably delivers real-time heart sound and ECG data to the upper-level computer
- Implements TCP/IP communication while minimizing MCU resource usage
- Contributes to the high reliability and low latency required in medical environments
π WIZnet Strategic Value
This project is a representative example of an Ethernet-based medical IoT system.
WIZnetβs W5500 hardware TCP/IP Ethernet chip can provide stable and reliable data transmission for STM32- and ESP32-based medical and healthcare devices, and it can serve as a practical reference for AI-assisted diagnostic devices and remote healthcare platforms like the system presented in this study.
π Summary
This study is a successful example of large-scale screening for congenital heart disease by combining AI-based heart sound analysis with an Ethernet-based embedded system.
The architecture using STM32 and Ethernet (W5100) serves as a strong reference model for low-cost, high-reliability medical IoT design.
WIZnet Ethernet solutions have significant potential to expand as core network infrastructure in next-generation smart healthcare devices.
π Overview
λ³Έ νλ‘μ νΈλ AI κΈ°λ° μ¬μ(Heart Sound) λΆμ μμ€ν μ μ΄μ©ν΄ **μ μ²μ± μ¬μ₯μ§ν(CHD)**μ λκ·λͺ¨λ‘ μ€ν¬λ¦¬λνλ μ°κ΅¬μ΄λ€. μ μ μ²μ§κΈ°λ‘ μμ§ν μ¬μ λ°μ΄ν°λ₯Ό STM32 + Ethernet(W5100) κΈ°λ° μλ² λλ μμ€ν μμ μμ μ»΄ν¨ν νλ«νΌμΌλ‘ μ μ‘νκ³ , λ₯λ¬λ(RNN/CNN) λͺ¨λΈμ ν΅ν΄ μ§νμ μλ λΆλ₯νλ€. μ΄ μμ€ν μ μλ£ μΈνλΌκ° λΆμ‘±ν μ§μμμλ νμ© κ°λ₯ν μ λΉμ©Β·κ³ νμ₯μ± μ§λ¨ 보쑰 μ루μ μ λͺ©νλ‘ νλ€.
π Features
- AI μ¬μ μ§λ¨
- ASD, VSD, PDA λ± μ£Όμ CHD μ ν μλ λΆλ₯
- RNN κΈ°λ° λͺ¨λΈλ‘ μ΅λ μ½ 90% μ νλ λ¬μ±
- λκ·λͺ¨ μ€ν¬λ¦¬λ κ²μ¦
- 518,258λͺ λμ μ€μ νμ₯ μ€ν¬λ¦¬λ μν
- μ λΉμ© νλμ¨μ΄ ꡬμ±
- STM32 λ§μ΄ν¬λ‘컨νΈλ‘€λ¬ + Ethernet λ€νΈμν¬
- νμ₯ μΉνμ μ€κ³
- μ΄λνΒ·μ μ λ ₯Β·μ€μκ° λ°μ΄ν° μμ§ λ° μ μ‘
π System Architecture


ν΄λΉ μμ€ν
μ νμ μλ² λλ μμ§ μμ€ν
(Lower Computer) κ³Ό μμ μ²λ¦¬Β·νμ μμ€ν
(Upper Computer) μΌλ‘ λͺ
νν λΆλ¦¬λ 2-κ³μΈ΅ ꡬ쑰λ₯Ό κ°μ§λ€.
μ΄ κ΅¬μ‘°λ μλ£ μ νΈμ μμ μ μμ§, μ λ’°μ± μλ μ μ‘, κ³ μ±λ₯ AI μ²λ¦¬λ₯Ό λͺ©μ μ λ§κ² λΆλ΄νλλ‘ μ€κ³λμλ€.
νμ μμ€ν (Lower Computer)
- μΌμ κ³μΈ΅
- μ¬μ μΌμ(Heart Sound Sensor)
- EMS μΌμ
β μΈμ²΄μμ λ°μνλ μ¬μ λ° μ체 μ νΈλ₯Ό μλ λ‘κ·Έ ννλ‘ μμ§
- μ νΈ μ μ²λ¦¬ λ° λ³ν
- ThinkLabs λͺ¨λμ ν΅ν΄ μΌμ μ νΈ μ ν
- AD7606 ADCλ₯Ό μ¬μ©ν΄ λ€μ±λ μλ λ‘κ·Έ μ νΈλ₯Ό λμ§νΈ λ°μ΄ν°λ‘ λ³ν
- λ©μΈ μ μ΄ MCU
- STM32F103ZET6
- ADC λ°μ΄ν° μμ§ λ° ν¨ν· ꡬμ±
- Ethernet μ μ‘μ μν λ°μ΄ν° μ²λ¦¬ λ΄λΉ
- λ€νΈμν¬ μΈν°νμ΄μ€
- W5100 Ethernet 컨νΈλ‘€λ¬
- STM32μ SPIλ‘ μ°κ²°
- TCP/IP ν΅μ μ νλμ¨μ΄ λ 벨μμ μ²λ¦¬
- μμ§λ μλ£ λ°μ΄ν°λ₯Ό μμ μ μΌλ‘ μμ μμ€ν μΌλ‘ μ μ‘
μμ μμ€ν (Upper Computer)
- APSoC κΈ°λ° μ²λ¦¬ νλ«νΌ
- ZedBoard (Zynq APSoC)
- PS(Processing System)μ PL(Programmable Logic)λ‘ κ΅¬μ±
- λ°μ΄ν° μμ λ° λ²νΌλ§
- Ethernetμ ν΅ν΄ νμ μμ€ν μΌλ‘λΆν° λ°μ΄ν° μμ
- DDR3 λ©λͺ¨λ¦¬μ λ°μ΄ν° μ μ₯ λ° κ³΅μ
- AI μ²λ¦¬ λ° μ ν리μΌμ΄μ
μ€ν
- Linux(Linaro) κΈ°λ° νκ²½
- QT κΈ°λ° UI λ° λ°μ΄ν° μμ§Β·λΆμ νλ‘κ·Έλ¨ μ€ν
- AI λͺ¨λΈμ ν΅ν μ¬μ₯ μ§ν λΆλ₯ λ° λ³΄μ‘° μ§λ¨ μν
- λμ€νλ μ΄ μΆλ ₯
- ADV7511 HDMI νΈλμ€λ―Έν°
- LCD λλΌμ΄λ² λ° HDMI λμ€νλ μ΄ μ°κ²°
- μ€μκ° νν, λΆμ κ²°κ³Ό μκ°ν
- λ΄λΆ μΈν°νμ΄μ€
- AXI μΈν°μ»€λ₯νΈλ₯Ό ν΅ν΄ PSβPL κ° λ°μ΄ν° κ΅ν
μ΄ κ΅¬μ‘°λ μλ£ μ νΈ μμ§βμ μ‘βλΆμμ λͺ νν λΆλ¦¬ν΄ νμ₯μ±κ³Ό μμ μ±μ ν보νλ€.
π Role and Application of the W5100
λ³Έ μ°κ΅¬μμλ W5100 Ethernet 컨νΈλ‘€λ¬κ° ν΅μ¬ λ€νΈμν¬ μν μ μννλ€.
- STM32 κΈ°λ° μλ² λλ 보λμμ μ μ Ethernet λ°μ΄ν° μ μ‘ λ΄λΉ
- μ€μκ° μ¬μΒ·ECG λ°μ΄ν°λ₯Ό μμ μ»΄ν¨ν°λ‘ μμ μ μΌλ‘ μ λ¬
- MCU μμμ μ΅μλ‘ μ¬μ©νλ©΄μ TCP/IP ν΅μ ꡬν
- μλ£ νκ²½μμ μꡬλλ μ λ’°μ±Β·μ§μ° μ΅μνμ κΈ°μ¬
π WIZnet Strategic Value
μ΄ νλ‘μ νΈλ Ethernet κΈ°λ° μλ£ IoT μμ€ν
μ λν μ¬λ‘λ€.
WIZnetμ W5500 νλμ¨μ΄ TCP/IP Ethernet μΉ©μ STM32Β·ESP32 κΈ°λ° μλ£Β·ν¬μ€μΌμ΄ μ₯λΉμμ μμ μ μΈ λ°μ΄ν° μ μ‘μ μ 곡ν μ μμΌλ©°,
λ³Έ μ°κ΅¬μ κ°μ AI μ§λ¨ 보쑰 μ₯μΉΒ·μ격 μλ£ νλ«νΌμμ μ€μ§μ μΈ λ νΌλ°μ€λ‘ νμ© κ°λ₯νλ€.
π Summary
λ³Έ μ°κ΅¬λ AI μ¬μ λΆμ + Ethernet μλ² λλ μμ€ν
μ κ²°ν©ν΄ μ μ²μ± μ¬μ₯μ§νμ λκ·λͺ¨λ‘ μ€ν¬λ¦¬λν μ±κ³΅μ μΈ μ¬λ‘λ€.
STM32μ Ethernet(W5100)μ νμ©ν ꡬ쑰λ μ λΉμ©Β·κ³ μ λ’° μλ£ IoT μ€κ³μ μ’μ μ°Έκ³ λͺ¨λΈμ΄ λλ€.
WIZnet Ethernet μ루μ
μ μ΄λ¬ν μ°¨μΈλ μ€λ§νΈ ν¬μ€μΌμ΄ κΈ°κΈ°μμ ν΅μ¬ λ€νΈμν¬ μΈνλΌλ‘ νμ₯ κ°λ₯μ±μ΄ ν¬λ€.
