1. A system (100) for automated cardiac anomaly detection from echocardiogram videos with real-time acquisition guidance, the system (100) comprising: an echocardiogram acquisition interface (102) integrated into an ultrasound imaging device (104) configured to receive live video streams, display real-time quality feedback and probe positioning guidance; a communication network (106) configured to transmit data between system components; a processing unit (108) connected to the echocardiogram acquisition interface (102) via the communication network (106), configured to perform cardiac anomaly detection and real-time quality assessment using machine learning, wherein the processing unit (108) further comprising: an input module (110) configured to receive echocardiogram video frames with associated metadata; a preprocessing module (112) configured to perform noise suppression, normalization, and frame alignment; a prediction module (114) configured to extract spatial features via convolutional neural networks, analyze temporal dynamics, and generate anomaly scores using deep learning models; a real-time quality evaluation module (116) configured to compute frame quality scores and generate positioning feedback with latency less than 200 milliseconds; a visualization module (118) configured to generate heatmap overlays highlighting anomalous regions; an optimization module (120) configured to balance diagnostic accuracy, computational efficiency, and processing latency using multi-objective optimization; a dynamic feedback module (122) configured to provide real-time probe positioning instructions, perform quality-gated frame selection, and trigger alerts during active acquisition; an adaptive learning module (124) configured to continuously update models using clinical feedback; an output module (126) configured to generate diagnostic reports with anomaly classifications and confidence scores.

2. The system (100) as claimed in claim 1, further comprising a database (128) configured to store historical patient studies, anomaly trajectories, prediction models, and clinical outcomes.

3. The system (100) as claimed in claim 1, wherein the prediction module (114) comprises convolutional neural networks, recurrent neural networks, variational autoencoders, or reinforcement learning agents.

4. The system (100) as claimed in claim 1, wherein the processing unit (108) calculates rate-of-change metrics from patient-specific histories, generates alerts for progressive deterioration, and outputs visualizations with temporal trend indicators.

5. The system (100) as claimed in claim 1, wherein the optimization module (120) employs Pareto optimization or evolutionary algorithms to optimize diagnostic accuracy, latency, and energy consumption.

6. The system (100) as claimed in claim 1, wherein the quality evaluation module (116) adapts processing depth based on frame quality thresholds.

7. The system (100) as claimed in claim 1, wherein the dynamic feedback module (122) performs view classification, calculates completeness scores, provides visual positioning guidance, and terminates acquisition when diagnostic adequacy is achieved.

8. A method for automated cardiac anomaly detection with real-time acquisition guidance, comprising: receiving live echocardiogram video frames via an input module (110); performing preprocessing via a preprocessing module (112); extracting spatial and temporal features to generate anomaly scores via a prediction module (114); computing real-time quality scores with sub-200 millisecond latency via a quality evaluation module (116); generating heatmap overlays via a visualization module (118); balancing diagnostic accuracy and efficiency via an optimization module (120); providing immediate probe positioning instructions, performing quality-gated selection, and calculating completeness scores via a dynamic feedback module (122); maintaining longitudinal anomaly databases and calculating rate-of-change metrics via the processing unit (108); updating models using clinical feedback via an adaptive learning module (124); generating diagnostic reports via an output module (126).

Description:FIELD OF DISCLOSURE
[0001] The present disclosure generally relates to the field of medical imaging and diagnostic systems, more specifically, relates to an adaptive cardiac anomaly detection system for echocardiogram analysis using deep learning with real-time acquisition guidance based on the integration of convolutional neural networks, temporal motion analysis, quality-driven feedback control, longitudinal trend monitoring, and automated probe positioning assistance for intelligent diagnostic support.
BACKGROUND OF THE DISCLOSURE
[0002] Cardiovascular diseases remain the leading cause of mortality worldwide, requiring accurate and timely diagnostic assessment through cardiac imaging. Echocardiography has become the primary non-invasive imaging modality for evaluating cardiac structure and function, enabling clinicians to detect wall motion abnormalities, valve defects, chamber enlargement, and myocardial dysfunction. Healthcare institutions increasingly depend on echocardiogram analysis to guide treatment decisions, monitor disease progression, and assess therapeutic effectiveness across diverse patient populations.
[0003] Traditional echocardiogram interpretation relies on manual visual assessment by trained cardiologists who analyze video sequences to identify anatomical and functional abnormalities. This conventional approach suffers from significant inter-observer variability, prolonged interpretation time, limited reproducibility, and subjective bias that cannot provide consistent quantitative measurements or standardized diagnostic criteria. Manual interpretation requires extensive training and experience, creating bottlenecks in healthcare delivery and limiting access to expert cardiac assessment in resource-constrained settings.
[0004] Current echocardiogram analysis solutions suffer from several critical limitations. First, they lack real-time quality assessment during image acquisition, resulting in suboptimal diagnostic studies that must be repeated, wasting valuable clinical time and patient resources. Sonographers have no immediate feedback on whether acquired views meet diagnostic quality standards, leading to incomplete or inadequate examinations that fail to capture all necessary cardiac segments. Second, existing automated systems analyze completed studies retrospectively without providing acquisition guidance, missing the opportunity to improve image quality at the point of capture.
[0005] Third, conventional diagnostic approaches perform single-timepoint analysis without tracking longitudinal changes in patient cardiac function over time. This prevents early detection of progressive deterioration and limits ability to identify subtle trends that may indicate developing pathology before overt abnormalities manifest. Fourth, existing computer-aided systems provide binary classification outputs without uncertainty quantification, offering no indication of prediction confidence or reliability to guide clinical decision-making.
[0006] While some progress has been made using convolutional neural networks for cardiac image classification or recurrent neural networks for temporal analysis, these approaches lack integrated intelligence combining real-time acquisition feedback, quality-based processing adaptation, longitudinal trend analysis, and confidence-aware diagnostic reporting. They cannot autonomously guide sonographers during active scanning, learn patient-specific baselines, or adapt computational resources based on input data quality.
[0007] Therefore, there is a critical need for an intelligent, self-adaptive cardiac diagnostic framework that not only analyzes cardiac structure and motion but also provides real-time acquisition guidance, evaluates prediction confidence, tracks longitudinal patient trends, and adapts processing strategies based on data quality. Such a solution must be autonomous, continuously learning, and capable of sophisticated decision-making to ensure optimal diagnostic accuracy, minimal scan time, consistent quality standards, and early disease detection across diverse healthcare environments.
SUMMARY OF THE DISCLOSURE
[0008] The following is a summary description of illustrative embodiments of the invention. It is provided as a preface to assist those skilled in the art to more rapidly assimilate the detailed design discussion which ensues and is not intended in any way to limit the scope of the claims which are appended hereto in order to particularly point out the invention.
[0009] According to illustrative embodiments, the present disclosure focuses on an adaptive cardiac anomaly detection system with real-time acquisition guidance using deep learning which overcomes the above-mentioned disadvantages and provides users with an intelligent, autonomous solution for efficient echocardiogram analysis and diagnostic support.
[0010] The present invention solves all the above major limitations of conventional cardiac imaging systems through intelligent AI-driven prediction, real-time quality assessment, acquisition guidance feedback, longitudinal trend monitoring, and confidence-aware diagnostic reporting with continuous learning capabilities.
[0011] The objective of the present disclosure is to provide an adaptive cardiac anomaly detection system that integrates deep learning-based spatial and temporal analysis, real-time quality evaluation, acquisition guidance feedback, and continuous learning mechanisms for long-term adaptability.
[0012] Another objective of the present disclosure is to enable intelligent cardiac feature extraction using convolutional neural networks and temporal analysis networks that extract morphological features and analyze dynamic motion patterns across cardiac cycles.
[0013] Another objective of the present disclosure is to implement real-time quality-driven acquisition control that evaluates frame quality and provides immediate positioning feedback to operators during active scanning with sub-200 millisecond latency.
[0014] Another objective of the present disclosure is to provide longitudinal trend monitoring that calculates rate-of-change metrics from patient-specific historical data and generates predictive alerts for progressive cardiac deterioration.
[0015] Yet another objective of the present disclosure is to employ multi-objective optimization algorithms that balance diagnostic accuracy, computational efficiency, and processing latency for optimal resource utilization.
[0016] Yet another objective of the present disclosure is to implement quality-adaptive processing depth that automatically adjusts analysis complexity based on input frame quality characteristics.
[0017] Yet another objective of the present disclosure is to establish diagnostic completeness scoring that tracks myocardial segment coverage and automatically terminates acquisition when adequacy thresholds are satisfied.
[0018] Yet another objective of the present disclosure is to develop a platform-agnostic system capable of integrating with diverse ultrasound imaging devices, PACS systems, and electronic health record platforms through vendor-independent interfaces.
[0019] In light of the above, in one aspect of the present disclosure, an adaptive cardiac anomaly detection system with real-time acquisition guidance using deep learning is disclosed herein. The system comprises an echocardiogram acquisition interface integrated into an ultrasound imaging device configured to receive live video streams, display real-time quality feedback and probe positioning guidance. The system also includes a communication network configured to transmit data between all components of the system. The system also includes a processing unit connected to the echocardiogram acquisition interface via the communication network, and configured to perform cardiac anomaly detection and real-time quality assessment using machine learning-driven adaptive analysis techniques, wherein the processing unit further comprises an input module configured to receive echocardiogram video frames with associated metadata, a preprocessing module configured to perform noise suppression, normalization, and frame alignment, a prediction module configured to extract spatial features via convolutional neural networks and analyze temporal dynamics to generate anomaly scores using deep learning models, a real-time quality evaluation module configured to compute frame quality scores and generate positioning feedback with latency less than 200 milliseconds, a visualization module configured to generate heatmap overlays highlighting anomalous regions, an optimization module configured to balance diagnostic accuracy, computational efficiency, and processing latency using multi-objective optimization, a dynamic feedback module configured to provide real-time probe positioning instructions, perform quality-gated frame selection, and trigger alerts during active acquisition, an adaptive learning module configured to continuously update models using clinical feedback, and an output module configured to generate diagnostic reports with anomaly classifications and confidence scores. The system also includes a database configured to store historical patient studies, anomaly trajectories, prediction models, and clinical outcomes.
[0020] In one embodiment, the prediction module comprises convolutional neural networks, three-dimensional convolutional networks, recurrent neural networks, variational autoencoders, or deep reinforcement learning agents for cardiac anomaly prediction.
[0021] In one embodiment, the processing unit maintains patient-specific anomaly databases, calculates rate-of-change metrics, classifies segments as improving, stable, or deteriorating, generates predictive alerts for progressive deterioration, and outputs visualizations with spatial heatmaps overlaid with temporal trend indicators.
[0022] In one embodiment, the optimization module employs Pareto optimization, genetic algorithms, or evolutionary algorithms to simultaneously optimize diagnostic accuracy, computational latency, and energy consumption.
[0023] In one embodiment, the quality evaluation module adapts processing depth based on frame quality, applying extended analysis when quality is high and conservative assessment with alerts when quality is low.
[0024] In one embodiment, the dynamic feedback module performs real-time view classification, calculates diagnostic completeness scores, generates visual positioning guidance with directional indicators, provides traffic-light quality feedback during acquisition, and automatically terminates acquisition when completeness threshold is satisfied.
[0025] In light of the above, in another aspect of the present disclosure, a method for automated cardiac anomaly detection with real-time acquisition guidance is disclosed herein. The method comprises the steps of receiving live echocardiogram video frames via an input module, performing preprocessing via a preprocessing module, extracting spatial and temporal features to generate anomaly scores via a prediction module, computing real-time quality scores with sub-200 millisecond latency via a quality evaluation module, generating heatmap overlays via a visualization module, balancing diagnostic accuracy and efficiency via an optimization module, providing immediate probe positioning instructions and calculating completeness scores via a dynamic feedback module, maintaining longitudinal anomaly databases and calculating rate-of-change metrics via the processing unit, updating models using clinical feedback via an adaptive learning module, and generating diagnostic reports via an output module.
[0026] These and other advantages will be apparent from the present application of the embodiments described herein.
[0027] The preceding is a simplified summary to provide an understanding of some embodiments of the present invention. This summary is neither an extensive nor exhaustive overview of the present invention and its various embodiments. The summary presents selected concepts of the embodiments of the present invention in a simplified form as an introduction to the more detailed description presented below. As will be appreciated, other embodiments of the present invention are possible utilizing, alone or in combination, one or more of the features set forth above or described in detail below.
[0028] These elements, together with the other aspects of the present disclosure and various features are pointed out with particularity in the claims annexed hereto and form a part of the present disclosure. For a better understanding of the present disclosure, its operating advantages, and the specified object attained by its uses, reference should be made to the accompanying drawings and descriptive matter in which there are illustrated exemplary embodiments of the present disclosure.
BRIEF DESCRIPTION OF THE DRAWINGS
[0029] To describe the technical solutions in the embodiments of the present disclosure or in the prior art more clearly, the following briefly describes the accompanying drawings required for describing the embodiments or the prior art. Apparently, the accompanying drawings in the following description merely show some embodiments of the present disclosure, and a person of ordinary skill in the art can derive other implementations from these accompanying drawings without creative efforts. All of the embodiments or the implementations shall fall within the protection scope of the present disclosure.
[0030] The advantages and features of the present disclosure will become better understood with reference to the following detailed description taken in conjunction with the accompanying drawing, in which:
[0031] FIG. 1 illustrates a system architecture diagram of adaptive cardiac anomaly detection framework depicting core modules and their interactions throughout the diagnostic lifecycle, in accordance with an exemplary embodiment of the present disclosure; and
[0032] FIG. 2 illustrates a workflow diagram of AI-based cardiac diagnostic process showing sequential flow of data and decision-making for intelligent anomaly detection and real-time acquisition guidance, in accordance with an exemplary embodiment of the present disclosure.
[0033] Like reference numerals refer to like parts throughout the description of several views of the drawing.
[0034] The adaptive cardiac anomaly detection system is illustrated in the accompanying drawings, which like reference letters indicate corresponding parts in the various figures. It should be noted that the accompanying figure is intended to present illustrations of exemplary embodiments of the present disclosure. This figure is not intended to limit the scope of the present disclosure. It should also be noted that the accompanying figure is not necessarily drawn to scale.
DETAILED DESCRIPTION OF THE DISCLOSURE
[0035] The following is a detailed description of embodiments of the disclosure depicted in the accompanying drawings. The embodiments are in such detail as to communicate the disclosure. However, the amount of detail offered is not intended to limit the anticipated variations of embodiments; on the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the present disclosure.
[0036] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the embodiments of the present disclosure. It may be apparent to one skilled in the art that embodiments of the present disclosure may be practiced without some of these specific details.
[0037] Various terms as used herein are shown below. To the extent a term is used, it should be given the broadest definition persons in the pertinent art have given that term as reflected in printed publications and issued patents at the time of filing.
[0038] The terms “a” and “an” herein do not denote a limitation of quantity, but rather denote the presence of at least one of the referenced items.
[0039] The terms “having”, “comprising”, “including”, and variations thereof signify the presence of a component.
[0040] Referring now to FIG. 1 to FIG. 2 to describe various exemplary embodiments of the present disclosure. FIG. 1 illustrates a system architecture diagram of a system 100 for adaptive cardiac anomaly detection with real-time acquisition guidance using deep learning, in accordance with an exemplary embodiment of the present disclosure.
[0041] The adaptive cardiac anomaly detection system 100 may include an echocardiogram acquisition interface 102, an ultrasound imaging device 104, a communication network 106, a processing unit 108, and a database 128.
[0042] The echocardiogram acquisition interface 102 is integrated into an ultrasound imaging device 104 and configured to receive live video streams during active patient examination, display real-time quality feedback, probe positioning guidance, and diagnostic reports throughout the examination lifecycle. The interface 102 serves as the primary interaction point between the clinical operator and the intelligent diagnostic system, presenting actionable guidance that improves examination quality while reducing scan duration. The interface 102 displays visual indicators including color-coded quality metrics, directional arrows for probe positioning adjustments, and progress indicators showing which myocardial segments have been adequately captured.
[0043] In one embodiment, the echocardiogram acquisition interface 102 is further configured to display real-time frame quality indicators using traffic-light color schemes where green indicates optimal quality, yellow indicates acceptable but suboptimal quality requiring minor adjustments, and red indicates poor quality requiring immediate probe repositioning. The interface 102 also displays directional probe positioning arrows overlaid on the live ultrasound image, showing operators precisely which direction and angle to move the transducer for optimal cardiac window acquisition. Additionally, the interface 102 presents diagnostic completeness scores as percentage values or segmented graphical representations showing which of the 16 standard myocardial segments have achieved adequate visualization coverage, anomaly heatmap overlays superimposed on anatomical cardiac images highlighting regions of detected abnormality, and confidence metrics expressed as numerical scores or visual certainty indicators to inform clinical decision-making.
[0044] The communication network 106 is configured to transmit data between all components of the system 100, ensuring seamless connectivity and secure data flow throughout the diagnostic pipeline. The network 106 may comprise wired connections such as Ethernet, USB, or fiber optic cables, wireless connections including WiFi, Bluetooth, or cellular networks, or hybrid configurations combining multiple transmission protocols for redundancy and reliability. The network 106 implements encryption protocols and authentication mechanisms to ensure patient data confidentiality and compliance with healthcare privacy regulations such as HIPAA or GDPR. The network 106 supports real-time data transmission with minimal latency to enable sub-200 millisecond feedback cycles essential for live acquisition guidance.
[0045] The processing unit 108 is connected to the echocardiogram acquisition interface 102 via the communication network 106, and configured to perform intelligent cardiac anomaly detection, real-time quality assessment, and autonomous optimization using machine learning-driven adaptive analysis techniques. The processing unit 108 may comprise one or more central processing units (CPUs), graphics processing units (GPUs) optimized for parallel neural network computations, tensor processing units (TPUs) specialized for deep learning inference, or field-programmable gate arrays (FPGAs) configured for low-latency real-time processing. The processing unit 108 executes trained deep learning models with inference speeds sufficient to analyze incoming video frames in real-time while maintaining diagnostic accuracy comparable to or exceeding human expert performance.
[0046] The processing unit 108 comprises multiple specialized modules: an input module 110, a preprocessing module 112, a prediction module 114, a real-time quality evaluation module 116, a visualization module 118, an optimization module 120, a dynamic feedback module 122, an adaptive learning module 124, and an output module 126. These modules operate in coordinated fashion, with data flowing sequentially through the processing pipeline while certain modules execute in parallel to minimize overall latency.
[0047] The input module 110 is configured to receive echocardiogram video frames with associated metadata including patient identifiers, view type, acquisition timestamp, frame quality indicators, and cardiac phase information. The input module 110 performs initial data validation to ensure frame integrity, checks for proper DICOM formatting compliance, extracts embedded metadata from medical imaging standards, synchronizes video frame timestamps with any concurrent electrocardiogram (ECG) signals when available, and buffers incoming frames to accommodate variable network transmission speeds while maintaining smooth data flow to subsequent processing stages.
[0048] The preprocessing module 112 is configured to perform speckle noise suppression, intensity normalization, boundary enhancement, and temporal frame alignment for predictive analysis, ensuring optimal data quality for accurate model performance. The preprocessing operations transform raw ultrasound video into standardized formats suitable for neural network analysis. Specifically, the preprocessing module 112 applies adaptive speckle reduction filters that suppress ultrasound-specific noise artifacts while preserving clinically relevant tissue boundaries and motion patterns. The module 112 performs intensity normalization using histogram equalization or contrast-limited adaptive histogram equalization (CLAHE) techniques to compensate for variable gain settings and tissue-dependent acoustic properties. Boundary enhancement algorithms sharpen endocardial and epicardial borders to improve myocardial wall detection accuracy. Temporal frame alignment uses optical flow estimation or feature-based registration to compensate for respiratory motion, patient movement, and probe drift that could degrade temporal analysis quality.
[0049] In one embodiment, the preprocessing module 112 further implements cardiac phase detection algorithms that automatically identify end-diastole and end-systole frames based on ventricular volume changes, even in the absence of concurrent ECG signals. This enables the system to perform phase-specific analysis and measurements that are clinically standardized to specific points in the cardiac cycle.
[0050] The prediction module 114 is configured to extract spatial features via convolutional neural networks, analyze temporal motion dynamics via temporal analysis networks, and generate anomaly scores for myocardial segments using deep learning models. The prediction module 114 represents the core artificial intelligence component of the system, implementing sophisticated neural network architectures trained on large datasets of annotated echocardiogram studies to recognize patterns associated with normal and abnormal cardiac function.
[0051] In one embodiment, the prediction module 114 comprises convolutional neural networks (CNNs) for spatial feature extraction that identify morphological characteristics including chamber dimensions, wall thickness, valve structure, and anatomical landmarks. The CNNs employ multiple convolutional layers with progressively increasing receptive fields to capture features at different spatial scales, from fine-grained tissue textures to global cardiac geometry. Three-dimensional convolutional networks extend this spatial analysis across temporal dimensions, processing video sequences as spatiotemporal volumes to capture motion patterns including wall motion, valve opening/closing dynamics, and blood flow characteristics. Recurrent neural networks including long short-term memory (LSTM) networks model long-range temporal dependencies that capture cardiac cycle periodicity, beat-to-beat variation, and sequential motion patterns that unfold over multiple frames. Variational autoencoders (VAEs) provide anomaly scoring via reconstruction error, where the network learns to reconstruct normal cardiac appearances and generates high reconstruction errors when presented with abnormal patterns that deviate from learned normality. Deep reinforcement learning agents can be employed for adaptive acquisition guidance, learning optimal probe manipulation strategies through trial-and-error interactions that maximize diagnostic information capture while minimizing examination duration.
[0052] The prediction module 114 generates segment-level anomaly scores for each of the 16 standard myocardial segments defined by American Society of Echocardiography guidelines, including basal anterior, basal anteroseptal, basal inferoseptal, basal inferior, basal inferolateral, basal anterolateral segments, mid-cavity anterior, mid-cavity anteroseptal, mid-cavity inferoseptal, mid-cavity inferior, mid-cavity inferolateral, mid-cavity anterolateral segments, apical anterior, apical septal, apical inferior, apical lateral segments, plus the apical cap. Each segment receives a quantitative anomaly score typically ranging from 0 (normal) to 1 (severely abnormal), with intermediate values indicating degrees of suspected pathology.
[0053] The real-time quality evaluation module 116 is configured to compute frame quality scores, assess diagnostic adequacy of acquired views, determine acquisition completeness metrics, and generate immediate positioning feedback with latency less than 200 milliseconds from frame capture. The module 116 operates in parallel with the prediction module 114, performing rapid quality assessment that does not depend on completion of full diagnostic analysis. This enables the system to provide instantaneous feedback to operators even as more computationally intensive anomaly detection proceeds in the background.
[0054] Quality assessment algorithms evaluate multiple factors including image contrast measured by histogram distribution metrics, endocardial border clarity quantified by edge strength and continuity, acoustic shadowing detected by analyzing intensity dropout regions, field-of-view adequacy determined by checking whether all required cardiac structures are visible within the imaging sector, and temporal stability assessed by analyzing frame-to-frame consistency and absence of excessive motion artifacts.
[0055] In one embodiment, the quality evaluation module 116 is configured to adapt processing depth based on frame quality, applying extended analysis with increased temporal windows when quality is high and conservative assessment with quality alert notifications when quality is low. When high-quality frames are detected, the module 116 signals the prediction module 114 to apply computationally intensive deep learning models with full temporal context windows spanning multiple cardiac cycles, enabling more accurate but slower analysis. When quality falls below acceptable thresholds, the module 116 triggers conservative processing modes that avoid wasting computational resources on unreliable data while simultaneously generating real-time alerts to guide operators toward improved acquisition technique.
[0056] The visualization module 118 is configured to generate gradient-weighted heatmap overlays highlighting anomalous cardiac regions and display real-time acquisition guidance indicators on the ultrasound interface. The visualization module 118 transforms abstract numerical anomaly scores into intuitive graphical representations that clinicians can rapidly interpret during patient examinations and subsequent diagnostic review.
[0057] The module 118 implements Gradient-weighted Class Activation Mapping (Grad-CAM) techniques that identify which spatial regions in the input echocardiogram images most strongly influenced the neural network's anomaly predictions. This produces heatmaps where colors ranging from blue (low contribution) through green and yellow to red (high contribution) are overlaid on the original cardiac images, clearly showing which myocardial regions exhibit abnormal patterns. These heatmaps can be displayed as static overlays on key frames, animated temporal sequences showing how abnormalities evolve throughout the cardiac cycle, or side-by-side comparisons with prior studies to visualize disease progression or treatment response.
[0058] In one embodiment, the visualization module 118 further generates three-dimensional reconstructions from multiple two-dimensional views, creating volumetric cardiac models with color-coded anomaly scores mapped onto anatomically accurate myocardial surfaces. This provides comprehensive spatial understanding of abnormality distribution across the entire heart.
[0059] The optimization module 120 is configured to balance diagnostic accuracy, computational efficiency, energy consumption, and processing latency using multi-objective optimization algorithms and dynamically adapt analysis depth based on real-time quality assessments. The module 120 manages computational resource allocation across competing objectives, ensuring the system delivers maximum clinical value while respecting practical constraints of deployment environments including battery-powered portable ultrasound devices, cloud-based processing services with usage-based costs, or hospital networks with limited bandwidth.
[0060] In one embodiment, the optimization module 120 employs Pareto optimization techniques that identify non-dominated solutions representing optimal trade-offs where no objective can be improved without degrading another. Genetic algorithms evolve populations of candidate resource allocation strategies over multiple generations, selecting superior configurations that balance all objectives. Particle swarm optimization simulates collective behavior of decentralized agents exploring the solution space to converge on global optima. Evolutionary algorithms apply mutation, crossover, and selection operations to iteratively refine allocation strategies. These algorithms simultaneously optimize diagnostic accuracy maximization by allocating sufficient computational resources for high-quality analysis, computational latency minimization by prioritizing time-critical operations and skipping unnecessary computations, and energy consumption reduction by dynamically scaling processing intensity based on available power budget and thermal constraints.
[0061] The optimization module 120 continuously monitors system performance metrics and adjusts processing strategies in real-time. For example, when operating on battery power in portable devices, the module 120 may reduce frame sampling rates, employ lighter neural network architectures, or defer non-critical analysis to conserve energy. When connected to wall power with ample computational resources, the module 120 enables maximum-depth analysis using ensemble models and extended temporal contexts to achieve highest possible accuracy.
[0062] The dynamic feedback module 122 is configured to provide real-time probe positioning instructions to operators during active acquisition, trigger automatic view classification, perform quality-gated frame selection, and initiate preemptive alerts when diagnostic completeness falls below adequacy threshold. The module 122 represents the key innovation enabling real-time acquisition guidance that distinguishes this system from conventional post-acquisition analysis approaches.
[0063] In one embodiment, the dynamic feedback module 122 is configured to perform real-time view classification identifying current standard echocardiographic projection from among apical 4-chamber, apical 2-chamber, apical long-axis, parasternal long-axis, parasternal short-axis at mitral valve level, parasternal short-axis at papillary muscle level, parasternal short-axis at apex, subcostal 4-chamber, subcostal short-axis, suprasternal, and other specialized views. View classification employs trained convolutional neural networks that recognize distinctive anatomical landmarks and spatial relationships characteristic of each standard view, achieving classification accuracy exceeding 95% within 100 milliseconds of frame capture.
[0064] The module 122 calculates diagnostic completeness scores indicating percentage of myocardial segments adequately visualized by tracking which segments have been captured with sufficient quality across acquired views. The system maintains a running tally of segment coverage, updating in real-time as each new frame is processed. When completeness falls below target thresholds, such as when fewer than 80% of segments have been adequately imaged, the module 122 generates alerts prompting operators to acquire additional views.
[0065] The module 122 generates visual positioning guidance including directional indicators for probe adjustment by analyzing the current view geometry, comparing it to ideal reference standards, and computing the geometric transformations required to achieve optimal imaging planes. For example, if the apical 4-chamber view shows foreshortened ventricles indicating excessive angulation, the system displays an arrow indicating "tilt probe posteriorly" with suggested angle adjustments. If endocardial borders are poorly visualized on one side of the image, the system may suggest "rotate probe 15 degrees clockwise" to improve acoustic window alignment.
[0066] The module 122 provides immediate operator feedback with traffic-light quality indicators during active acquisition, using green lights to signal "optimal quality - maintain current position," yellow lights to indicate "acceptable quality - minor adjustments recommended," and red lights to warn "poor quality - reposition probe immediately." These intuitive visual cues enable even less experienced operators to achieve diagnostic-quality acquisitions by following system guidance.
[0067] The module 122 automatically terminates acquisition when completeness threshold is satisfied, displaying messages such as "All segments adequately visualized - examination complete" once the system confirms sufficient diagnostic information has been captured. This prevents unnecessary prolonged scanning that wastes time and patient tolerance while ensuring no critical anatomical regions are inadvertently missed.
[0068] The adaptive learning module 124 is configured to continuously update machine learning models using clinical outcome feedback data and verified diagnostic results, implementing online learning mechanisms for continuous improvement. The module 124 ensures the system does not remain static after initial deployment but rather evolves and adapts to local patient populations, institutional imaging protocols, equipment characteristics, and emerging disease patterns.
[0069] The module 124 collects outcome feedback including confirmed diagnoses from follow-up catheterization procedures, surgical findings, autopsy results, or clinical course observations that verify or refute initial automated predictions. This ground-truth data is used to fine-tune model parameters through techniques including transfer learning where pre-trained models are adapted to local data distributions, incremental learning where models are updated with new examples without catastrophic forgetting of previous knowledge, and reinforcement learning where the system receives rewards for accurate predictions and penalties for errors, gradually improving decision policies through trial and error.
[0070] In one embodiment, the adaptive learning module 124 implements federated learning approaches that enable the system to benefit from collective experience across multiple institutions while preserving patient privacy. Local model updates are computed on-site using institution-specific data, and only anonymized model parameter adjustments are shared with central servers for aggregation. This allows the system to leverage large-scale diverse datasets for robust model training while maintaining strict data governance and regulatory compliance.
[0071] The module 124 operates autonomously without requiring manual intervention, automatically detecting when sufficient new training data has accumulated to warrant model retraining, executing optimization procedures during low-usage periods to avoid interfering with clinical operations, validating updated models on held-out test sets before deployment, and rolling back to previous versions if performance degradation is detected.
[0072] The output module 126 is configured to generate diagnostic reports including anomaly classifications, confidence scores, segment-level assessments, heatmap visualizations, and completeness metrics for clinical review. The module 126 produces comprehensive documentation suitable for integration into electronic health records, clinical decision support systems, and medicolegal documentation requirements.
[0073] Generated reports include structured data fields containing categorical classifications such as "normal," "mild abnormality," "moderate abnormality," or "severe abnormality" for overall cardiac function and each myocardial segment. Quantitative metrics include ejection fraction estimates, wall motion scores, valve function assessments, and chamber dimension measurements. Confidence scores expressed as percentages or probability distributions indicate the system's certainty in each prediction, enabling clinicians to appropriately weight automated findings alongside other clinical information.
[0074] Reports incorporate visual elements including representative key frames showing normal and abnormal regions, annotated heatmap overlays highlighting areas of concern, temporal graphs showing cardiac motion patterns over time, and comparison views displaying current findings alongside prior studies to track disease progression or treatment response. Completeness metrics document which views were acquired and which myocardial segments achieved adequate visualization, providing quality assurance for the examination.
[0075] The database 128 is connected to the processing unit 108 via the communication network 106 and configured to store historical patient echocardiogram studies, temporal anomaly score trajectories, prediction model weights, quality assessment parameters, and clinical validation outcomes. The database 128 implements secure storage infrastructure compliant with healthcare data protection regulations, employing encryption for data at rest and in transit, access control mechanisms limiting data retrieval to authorized personnel, audit logging to track all data access for security monitoring, and backup systems ensuring data durability and disaster recovery capabilities.
[0076] The database 128 maintains longitudinal patient records linking multiple studies over time, enabling temporal trend analysis and patient-specific baseline establishment. Stored data includes raw echocardiogram video files, extracted features and intermediate processing results, final anomaly scores and classifications, quality metrics for each frame and examination, operator feedback and manual overrides when clinicians disagree with automated predictions, and outcome data including subsequent diagnoses, treatments, and clinical events.
[0077] The database 128 also stores machine learning model parameters including neural network weights, normalization constants, and hyperparameter configurations, enabling model versioning and reproducibility. Quality assessment thresholds and optimization parameters can be customized per institution or equipment type and stored for consistent application across examinations.
[0078] In one embodiment, the processing unit 108 is further configured to maintain longitudinal patient-specific anomaly score databases, calculate rate-of-change metrics by comparing current segment scores to historical baselines normalized by time interval, classify segments into improving, stable, or deteriorating categories based on velocity thresholds, generate predictive alerts for segments exhibiting progressive deterioration even when current scores remain within normal range, and output dual visualizations comprising spatial heatmaps overlaid with temporal trend indicators showing direction and magnitude of change per myocardial segment.
[0079] This longitudinal analysis capability enables early detection of subtle disease progression that might not be apparent from single-timepoint analysis. For example, if a patient's anterior wall anomaly score has increased from 0.3 (normal range) at baseline to 0.45 at 6-month follow-up and now 0.65 at current examination, the system calculates a deterioration velocity of approximately 0.06 units per month and extrapolates that clinical abnormality threshold (typically 0.8) will be crossed within approximately 2.5 months if the trend continues. This triggers a predictive alert: "Anterior wall showing progressive deterioration - recommend increased surveillance or intervention consideration" even though the current score of 0.65 remains technically within borderline-normal range.
[0080] Trend indicators are visualized as arrows overlaid on segment diagrams, with arrow direction showing improvement (upward green), stability (horizontal yellow), or deterioration (downward red), and arrow thickness or length encoding the magnitude of change rate. This provides clinicians with intuitive at-a-glance assessment of disease trajectory across all myocardial segments simultaneously.
[0081] FIG. 2 illustrates a workflow diagram depicting the AI-based cardiac diagnostic process, showing the sequential flow of data collection, quality assessment, feature extraction, anomaly detection, real-time feedback, and adaptive learning that enables autonomous echocardiogram analysis with acquisition guidance.
[0082] The method 200 may include, at step 202, receiving live video frames via input module 110 where incoming echocardiogram video streams are captured and validated, at step 204, performing preprocessing via preprocessing module 112 where noise reduction, normalization, and alignment operations prepare data for analysis, at step 206, extracting features and generating anomaly scores via prediction module 114 where deep learning models identify morphological and functional abnormalities, at step 208, computing real-time quality scores via quality evaluation module 116 where frame adequacy is assessed within sub-200 millisecond latency, at step 210, generating heatmap overlays via visualization module 118 where anomaly locations are visualized for clinical interpretation, at step 212, balancing accuracy and efficiency via optimization module 120 where computational resources are allocated to maximize diagnostic value, at step 214, providing positioning instructions and completeness scoring via dynamic feedback module 122 where real-time guidance improves acquisition quality, at step 216, maintaining longitudinal databases and calculating trend metrics where patient-specific histories enable progressive change detection, at step 218, updating models via adaptive learning module 124 where continuous improvement occurs through outcome feedback, and at step 220, generating diagnostic reports via output module 126 where comprehensive findings are documented for clinical use.
[0083] In the best mode of operation, the adaptive cardiac anomaly detection system 100 operates through coordinated functioning of all components to deliver intelligent, autonomous diagnostic support with real-time acquisition guidance and minimal human intervention.
[0084] Upon system initialization, when echocardiogram video frames arrive at input module 110 from the ultrasound device during active patient scanning, the preprocessing module 112 immediately applies speckle noise suppression, intensity normalization, and boundary enhancement to ensure optimal data quality. Simultaneously, the real-time quality evaluation module 116 analyzes each incoming frame within 200 milliseconds, computing quality scores based on acoustic shadow artifacts, endocardial boundary clarity, and temporal stability.
[0085] The prediction module 114 employs trained convolutional neural networks to extract spatial features capturing myocardial structure, wall thickness, and chamber geometry, while temporal analysis networks model inter-frame motion dynamics across cardiac cycles. These features are combined to generate segment-level anomaly scores for each of the 16 standard myocardial segments, with variational autoencoders calculating reconstruction error to identify deviations from learned normal patterns.
[0086] Based on quality scores from module 116, the optimization module 120 dynamically adapts processing depth, applying extended temporal window analysis when frame quality is high and conservative assessment when quality is low. The dynamic feedback module 122 continuously performs real-time view classification, determining whether the current acquisition represents apical 4-chamber, parasternal long-axis, or other standard projections, and calculates diagnostic completeness scores showing what percentage of myocardial segments have been adequately visualized.
[0087] When quality scores fall below threshold or when specific myocardial segments remain inadequately captured, module 122 generates immediate visual feedback on the ultrasound display, providing directional arrows indicating optimal probe positioning adjustments and traffic-light indicators showing acquisition status. If diagnostic completeness falls below adequacy threshold, the system triggers alerts prompting the sonographer to acquire additional views. Once all segments reach quality thresholds and completeness criteria are satisfied, the system automatically signals acquisition termination, minimizing unnecessary scanning duration.
[0088] Throughout all operations, the processing unit 108 accesses the patient's historical echocardiogram studies from database 128, calculating rate-of-change metrics for each myocardial segment by comparing current anomaly scores to previous examinations normalized by time interval. Segments exhibiting progressive deterioration are classified and flagged with predictive alerts, even when current scores remain within normal range, enabling early detection of developing pathology.
[0089] The visualization module 118 generates gradient-weighted heatmap overlays highlighting cardiac regions contributing most to anomaly classifications, with dual visualizations showing both current spatial anomaly distribution and temporal trend arrows indicating direction and velocity of change per segment. These visualizations are displayed on the acquisition interface 102 in real-time during scanning and included in final diagnostic reports.
[0090] The adaptive learning module 124 continuously collects clinical outcome feedback including verified diagnoses, treatment responses, and follow-up assessments, using this data to iteratively refine prediction models through incremental learning algorithms. The learning process occurs autonomously without manual intervention, ensuring the system adapts to institutional-specific imaging protocols, patient population characteristics, and emerging disease patterns.
[0091] All diagnostic results, quality metrics, completeness scores, trend analyses, and learning progress are stored in database 128 for historical tracking and model training, while output module 126 generates comprehensive diagnostic reports with anomaly classifications, confidence scores, segment-level assessments, heatmap visualizations, and temporal trends that are transmitted through communication network 106 for clinical review.
[0092] This integrated operation of real-time quality assessment, acquisition guidance feedback, spatial and temporal feature extraction, longitudinal trend monitoring, confidence-aware reporting, and continuous adaptive learning enables the system 100 to achieve superior diagnostic accuracy, minimized scan duration, consistent quality standards, early disease detection, and improved clinical workflow efficiency across diverse healthcare environments while continuously improving performance through autonomous learning from operational experience.
[0093] While the invention has been described in connection with what is presently considered to be the most practical and various embodiments, it will be understood that the invention is not to be limited to the disclosed embodiments, but on the contrary, is intended to cover various modifications and equivalent arrangements included within the scope of the appended claims.
[0094] A person of ordinary skill in the art may be aware that, in combination with the examples described in the embodiments disclosed in this specification, units and algorithm steps may be implemented by electronic hardware, computer software, or a combination thereof.
[0095] The foregoing descriptions of specific embodiments of the present disclosure have been presented for purposes of illustration and description. They are not intended to be exhaustive or to limit the present disclosure to the precise forms disclosed, and many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described to best explain the principles of the present disclosure and its practical application, and to thereby enable others skilled in the art to best utilize the present disclosure and various embodiments with various modifications as are suited to the particular use contemplated. It is understood that various omissions and substitutions of equivalents are contemplated as circumstances may suggest or render expedient, but such omissions and substitutions are intended to cover the application or implementation without departing from the scope of the present disclosure.
[0096] Disjunctive language such as the phrase “at least one of X, Y, Z,” unless specifically stated otherwise, is otherwise understood with the context as used in general to present that an item, term, etc., may be either X, Y, or Z, or any combination thereof (e.g., X, Y, and/or Z). Thus, such disjunctive language is not generally intended to, and should not, imply that certain embodiments require at least one of X, at least one of Y, or at least one of Z to each be present.
[0097] In a case that no conflict occurs, the embodiments in the present disclosure and the features in the embodiments may be mutually combined. The foregoing descriptions are merely specific implementations of the present disclosure, but are not intended to limit the protection scope of the present disclosure. Any variation or replacement readily figured out by a person skilled in the art within the technical scope disclosed in the present disclosure shall fall within the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure shall be subject to the protection scope of the claims.
, Claims:I/We Claim:
1. A system (100) for automated cardiac anomaly detection from echocardiogram videos with real-time acquisition guidance, the system (100) comprising:
an echocardiogram acquisition interface (102) integrated into an ultrasound imaging device (104) configured to receive live video streams, display real-time quality feedback and probe positioning guidance;
a communication network (106) configured to transmit data between system components;
a processing unit (108) connected to the echocardiogram acquisition interface (102) via the communication network (106), configured to perform cardiac anomaly detection and real-time quality assessment using machine learning, wherein the processing unit (108) further comprising:
an input module (110) configured to receive echocardiogram video frames with associated metadata;
a preprocessing module (112) configured to perform noise suppression, normalization, and frame alignment;
a prediction module (114) configured to extract spatial features via convolutional neural networks, analyze temporal dynamics, and generate anomaly scores using deep learning models;
a real-time quality evaluation module (116) configured to compute frame quality scores and generate positioning feedback with latency less than 200 milliseconds;
a visualization module (118) configured to generate heatmap overlays highlighting anomalous regions;
an optimization module (120) configured to balance diagnostic accuracy, computational efficiency, and processing latency using multi-objective optimization;
a dynamic feedback module (122) configured to provide real-time probe positioning instructions, perform quality-gated frame selection, and trigger alerts during active acquisition;
an adaptive learning module (124) configured to continuously update models using clinical feedback;
an output module (126) configured to generate diagnostic reports with anomaly classifications and confidence scores.
2. The system (100) as claimed in claim 1, further comprising a database (128) configured to store historical patient studies, anomaly trajectories, prediction models, and clinical outcomes.
3. The system (100) as claimed in claim 1, wherein the prediction module (114) comprises convolutional neural networks, recurrent neural networks, variational autoencoders, or reinforcement learning agents.
4. The system (100) as claimed in claim 1, wherein the processing unit (108) calculates rate-of-change metrics from patient-specific histories, generates alerts for progressive deterioration, and outputs visualizations with temporal trend indicators.
5. The system (100) as claimed in claim 1, wherein the optimization module (120) employs Pareto optimization or evolutionary algorithms to optimize diagnostic accuracy, latency, and energy consumption.
6. The system (100) as claimed in claim 1, wherein the quality evaluation module (116) adapts processing depth based on frame quality thresholds.
7. The system (100) as claimed in claim 1, wherein the dynamic feedback module (122) performs view classification, calculates completeness scores, provides visual positioning guidance, and terminates acquisition when diagnostic adequacy is achieved.
8. A method for automated cardiac anomaly detection with real-time acquisition guidance, comprising:
receiving live echocardiogram video frames via an input module (110);
performing preprocessing via a preprocessing module (112);
extracting spatial and temporal features to generate anomaly scores via a prediction module (114);
computing real-time quality scores with sub-200 millisecond latency via a quality evaluation module (116);
generating heatmap overlays via a visualization module (118);
balancing diagnostic accuracy and efficiency via an optimization module (120);
providing immediate probe positioning instructions, performing quality-gated selection, and calculating completeness scores via a dynamic feedback module (122);
maintaining longitudinal anomaly databases and calculating rate-of-change metrics via the processing unit (108);
updating models using clinical feedback via an adaptive learning module (124);
generating diagnostic reports via an output module (126).