Smart Stethoscope Neural Network Heart State Model Fitting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current electronic stethoscopes are limited in their ability to accurately analyze heart sounds, particularly when strong murmur signals are present, and few systems can provide real-time analysis capable of distinguishing between normal and abnormal heart conditions, which is crucial for early detection in both developed and developing countries.
Innovation Solution
A method and system that captures time series acoustic heart signal data, classifies it using a neural network into heart sound categories, and fits the data into heart state models with adjustable durations to determine the sequence of heart states, providing a confidence value for the model fit, thereby facilitating the identification of normal or abnormal heart conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional acoustic heart signal processing is used, then the system is simple and easy to implement, but the detection accuracy is insufficient particularly when strong murmur signals are present
Solution Approach 1:
The patent segments the heart signal processing into distinct phases (systole and diastole) with specific state sequences. By dividing the continuous signal into discrete temporal segments with defined characteristics, the system can apply targeted analysis to each phase, improving detection accuracy while maintaining manageable system complexity through structured organization of the processing workflow.
Solution Approach 2:
The patent implements dynamic state sequences that adapt to the temporal characteristics of heart sounds. The system transitions between different states (e.g., S1, systole, S2, diastole) based on the evolving signal characteristics, allowing the processing approach to dynamically adjust to the current phase of the cardiac cycle and the presence of murmurs, thereby improving accuracy without requiring a completely complex static system.
2Measurement precision
If complex neural network classification is applied, then the classification accuracy improves, but the processing time and computational resources increase
Solution Approach 1:
The patent applies preliminary segmentation of the heart signal into distinct phases (S1, systole, S2, diastole) before neural network classification. By pre-organizing the signal into temporally-defined segments with expected characteristics, the neural network receives pre-processed input that reduces the complexity of classification, thereby improving accuracy while reducing the computational time required for the neural network to process the data.
3Adaptability or versatility
If state sequence models with fixed durations are used, then the model is simple to implement, but the adaptability to varying heart rates and conditions is limited
Solution Approach 1:
The patent implements dynamic state sequence models where the duration and timing of states (S1, systole, S2, diastole) are not fixed but adapt to the detected heart rate and signal characteristics. The model dynamically adjusts state durations based on the temporal patterns observed in the input signal, allowing it to accommodate varying heart rates and cardiac conditions while maintaining a relatively simple underlying model structure that follows the natural physiology of the cardiac cycle.
Data Source
AI summary
An electronic stethoscope system comprising a device to capture an acoustic heart signal from a patient and a neural network to classify the data into heart sound categories to provide time series sound category data comprising, for each of a succession of time intervals, category probability data representing a probability of the acoustic signal falling into each of the categories. The stethoscope also includes one or more heart state models each having a sequence of heart cardiac cycle states, a system to fit the time series sound category data to the models and determine timing data for the sequence of heart states and a confidence value for the model fit, and an output to output one or both of a model fit indication dependent upon the confidence value and an indication of the timing data.