Heart Sound PH Detection Using Explainable A2/P2 Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting pulmonary hypertension (PH) are invasive, costly, or lack accuracy and explainability, making them unsuitable for widespread use in low- and middle-income regions, and existing automated methods do not provide sufficiently accurate results or explain which heart sound components contribute to the output.
Innovation Solution
An over-parameterized deep neural network is used to analyze heart sound recordings, with a pre-processing step to separate the aortic (A2) and pulmonary (P2) components, and optional compression, along with alternative training methods like fixed-weight networks and batch gradient descent, to achieve high accuracy and explainability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If right heart catheterization is used for PH detection, then measurement precision is improved, but device complexity and ease of operation worsen due to invasive nature and specialized team requirements
Solution Approach 1:
The patent replaces the mechanical invasive catheterization system with an acoustic sensing system using microphones or sensors to capture heart sounds. This substitution eliminates the need for specialized teams and invasive procedures while maintaining diagnostic capability through automated analysis of cardiac acoustic signals.
Solution Approach 2:
The patent introduces an intermediary automated analysis system that processes heart sound recordings to detect PH. This intermediary layer between the physical heart sounds and the diagnostic conclusion eliminates the need for specialized clinicians to perform invasive procedures, while still achieving accurate PH detection through algorithmic analysis.
2Measurement precision
If implantable pressure sensors are used, then measurement precision is improved, but device complexity and ease of operation worsen due to extreme cost and invasive nature
Solution Approach 1:
The patent replaces expensive implantable pressure sensors with non-invasive acoustic sensors that capture heart sounds externally. This substitution eliminates implantation surgery and extreme costs while deriving pulmonary pressure information indirectly from the acoustic characteristics of heart sounds, particularly the P2 component.
Solution Approach 2:
The patent extracts pulmonary pressure information from the acoustic signal without requiring physical implantation in the pulmonary artery. By analyzing the P2 heart sound component and its relationship with A2, the system extracts the necessary diagnostic information non-invasively, eliminating the need for implanted sensors.
3Measurement precision
If Doppler echocardiography is used for PH screening, then measurement precision is improved, but device complexity and ease of operation worsen due to trained technician requirements and expensive machinery
Solution Approach 1:
The patent replaces complex Doppler echocardiography machinery with simple acoustic sensors and automated analysis software. This substitution eliminates the need for expensive ultrasound equipment and trained technicians, while achieving comparable or superior PH detection accuracy through automated processing of heart sound recordings.
Solution Approach 2:
The patent implements self-service automated analysis that processes heart sound recordings without requiring trained technicians. The system automatically segments heart sounds, identifies A2 and P2 components, and detects PH based on predetermined criteria, making the diagnostic process independent of specialized human operators.
4Ease of operation
If automated PH detection using cardiac auscultation data is used, then ease of operation is improved, but measurement precision worsens due to low predictive performance
Solution Approach 1:
The patent segments the second heart sound (S2) into its A2 and P2 components and analyzes their temporal and amplitude relationships. By focusing on specific segments of the heart sound signal and their physiological relationships, the system achieves high predictive performance while maintaining ease of operation through automated processing.
Solution Approach 2:
The patent transitions from analyzing simple presence/absence of heart sounds to analyzing multiple dimensions including temporal relationships between A2 and P2, amplitude differences, and spectral characteristics. This dimensional expansion of the analysis enables high predictive performance while keeping the system easy to operate through automated multi-parameter assessment.
5Ease of operation
If existing automated methods are used, then ease of operation is improved, but measurement precision and reliability worsen due to low predictive performance and lack of explainability
Solution Approach 1:
The patent segments the heart sound signal into A2 and P2 components and analyzes their temporal and amplitude relationships. This segmentation enables both high predictive performance through physiologically grounded features and explainability by showing which specific signal segments contribute to the PH diagnosis.
Solution Approach 2:
The patent provides feedback through visualization of the analyzed heart sound components and their relationships. The system displays the segmented A2 and P2 waves, their temporal separation, and amplitude differences, giving clinicians feedback on the basis of the PH diagnosis and enhancing both reliability and explainability.
Data Source
AI summary
Disclosed is a computer-implemented method for non-invasive estimation of Pulmonary Hypertension, PH, from heart sound signals. Consistent with the disclosure, the method includes the steps of: receiving a sound signal acquired from a beating heart of a subject over a predetermined time period; generating one or more 2D feature maps comprising a 2D feature map with the received sound signal where a first axis of the map is arranged over time and a second axis of the map is arranged over individual heartbeats; applying a pre-trained neural network to relate the generated one or more 2D feature maps with a training dataset of previously acquired and generated training 2D feature maps of a PH subject group and a non-PH subject group, thus to obtain an indicator of the presence of Pulmonary Hypertension. Also disclosed is a training method of said neural network and a system.