Embedded Heart Sound Classifier Using Onboard Neural Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current cardiac auscultation devices require connection to a mobile app or cloud for processing, limiting their use to real-time heart murmur detection and often miss subtle heart murmurs during examinations.
Innovation Solution
An embedded electronic device attached to a stethoscope that performs real-time processing using a trained deep neural network to detect heart murmurs, providing immediate feedback through visual or audio cues without the need for external connections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If heart murmur detection is performed using external processing devices (mobile app/cloud), then processing capability is improved, but real-time detection capability deteriorates
Solution Approach 1:
An embedded electronic device with onboard processing capabilities serves as an intermediary between the stethoscope and external computing resources. This intermediate device performs real-time heart murmur detection and classification using integrated neural network processors, eliminating the time delay associated with transmitting data to external devices while maintaining advanced processing capabilities.
Solution Approach 2:
The system segments the processing function from the data collection function. The embedded device handles real-time signal processing and murmur detection, while external devices can optionally receive summarized results. This segmentation allows real-time detection to occur independently without waiting for external processing resources.
2Measurement precision
If connection to mobile app and cloud is required for processing, then processing accuracy is improved, but ease of operation deteriorates
Solution Approach 1:
The embedded electronic device performs self-service by containing all necessary processing capabilities within the device itself. The onboard neural network processor executes heart murmur detection and classification algorithms independently, eliminating the need for users to connect to mobile apps or cloud services. This self-contained approach maintains processing accuracy while significantly improving ease of operation.
Solution Approach 2:
The embedded device integrates multiple functions into a single unit: signal acquisition, real-time processing, murmur detection, classification, and result display. This multi-functional design eliminates dependency on external devices and simplifies the user experience while maintaining comprehensive processing capabilities.
3Measurement precision
If intensive processing is performed on external devices, then detection accuracy is improved, but device complexity deteriorates
Solution Approach 1:
The embedded electronic device acts as an intermediary that consolidates complex processing functions within a single integrated unit. By incorporating onboard neural network processors and dedicated hardware accelerators, the device handles intensive signal processing internally, presenting a simplified interface to users while maintaining high detection accuracy.
4Ease of operation
If real-time processing is performed on embedded device, then ease of operation is improved, but processing capability deteriorates
Solution Approach 1:
The embedded device achieves self-service by integrating sufficient processing power directly into the device. The onboard neural network processors and dedicated hardware accelerators provide adequate processing capability for heart murmur detection and classification, eliminating the need for external computing resources while maintaining ease of operation.
Solution Approach 2:
The system replaces traditional mechanical connections (cables, physical interfaces) with wireless or integrated communication protocols. The embedded device communicates detection results wirelessly or through simple indicators, eliminating the need for physical connections to external devices while maintaining adequate processing capability for clinical use.
Data Source
AI summary
An automatic diagnostic apparatus and corresponding method is disclosed for recognizing heart sounds of interest, i.e., murmurs, detected in streaming audio data picked up by a stethoscope. Sensors included in the device capture audio data in real time during an auscultation exam performed by a physician. A feature vector that models the stream of audio data is created and supplied to a deep neural network stored on the diagnostic device. The deep neural network generates a probability for each of the heart sounds of interest. When the probability of detection exceeds a pre-established threshold value the device alerts the physician through visual and/or audio cues, enhancing the physician's diagnostic capability during routine examination.