Real-time device for predicting heart disease and supporting clinical decisions based on machine learning

Abstract

A system for real-time prediction of heart disease and to support clinical decisions; the system includes: a patient data acquisition module configured to receive multimodal inputs, including physiological signals selected from electrocardiographic waveforms, blood pressure readings, heart rate readings, and oxygen saturation readings, as well as demographic and lifestyle information, including age, gender, cholesterol levels, smoking habits, and family medical history; a preprocessing engine configured to perform data cleansing, imputation of missing values, categorical coding, and feature scaling, so that the input data is normalized and converted into a format suitable for machine learning; a processing core comprising a multi-core CPU / GPU system-on-chip operationally coupled with a secure storage unit, wherein the processing core is configured to execute a variety of pre-trained machine learning models, including Naive Bayes, Random Forest, Logistic Regression and Decision Tree classifiers; a model evaluation unit configured to calculate validation metrics such as precision, recall, F1 score and area under the curve for each of the models and dynamically select the model with optimal performance to generate real-time predictions for the risk of heart disease; a display interface configured to present predictive results in the form of risk probability values, confidence indices, and actionable recommendations that are mapped to clinical treatment guidelines; and a communication interface configured to transmit emergency alerts based on high-risk predictions to remote caregivers, hospitals, and emergency response systems via wireless communication protocols such as WLAN, Bluetooth, and 4G / 5G cellular connections.

Description

Technical field of the invention

[0001] The present invention relates to the field of medical devices and computer-aided health technologies and in particular to a system and device architecture that integrates machine learning techniques with clinical sensors, interfaces for patient data acquisition and decision support mechanisms for real-time prediction of the risk of heart disease and for guiding emergency treatment. Background of the invention

[0002] Cardiovascular diseases remain the leading cause of death worldwide. Early detection and accurate prediction of risk factors for heart disease are crucial for reducing morbidity and mortality. Conventional diagnostic systems rely on retrospective assessment methods such as the Framingham Risk Score, which cannot be adapted to different patient demographics and cannot integrate real-time monitoring data.

[0003] Recent advances in artificial intelligence (AI) and machine learning (ML) have demonstrated superior predictive capabilities compared to traditional models. However, existing AI-based systems are typically limited to software platforms and cannot be integrated with medical hardware for deployment in clinical, emergency, or rural healthcare settings.

[0004] Therefore, there is a need for a dedicated device and system architecture that combines demographic, clinical, physiological and real-time monitoring inputs with advanced, ML-based predictive models, providing medical professionals with actionable insights in real time while ensuring portability, reliability and ease of use in emergency care.

[0005] The challenge of predicting and treating cardiovascular disease has preoccupied medical researchers and clinicians for decades, as heart disease remains one of the leading causes of death worldwide. With the rapid growth of the global population and the increasing burden of lifestyle-related diseases, the need for accurate, real-time, and readily accessible diagnostic solutions is more urgent than ever. Historically, cardiovascular risk assessment relied heavily on statistical tools and clinical scoring systems derived from large epidemiological studies. One of the best-known examples is the Framingham Risk Score, which assesses a patient's likelihood of developing heart disease within a specific timeframe, taking into account factors such as age, cholesterol levels, blood pressure, and smoking habits.While this model is effective as a general guideline, it has inherent limitations because it relies on predefined statistical weights that do not adequately account for the heterogeneity of patient populations, evolving medical data, or region-specific factors. In many cases, patients outside the demographic group used in the original datasets—such as different ethnic groups or individuals with comorbidities—are poorly represented, leading to misclassification of risk. Furthermore, these tools tend to operate statically, providing a risk estimate at a specific point in time without the ability to dynamically adjust predictions based on continuous or real-time patient monitoring.

[0006] To address the shortcomings of purely statistical models, more sophisticated computational approaches have been introduced over the past two decades, particularly through the use of artificial intelligence and machine learning. Commercial solutions such as IBM Watson Health and Google DeepMind Health are attempting to integrate predictive analytics into healthcare, offering insights that go beyond the capabilities of traditional scoring systems. IBM Watson Health, for example, is being used in cardiovascular care by leveraging natural language processing and knowledge-based reasoning to synthesize patient records, published literature, and structured data. While Watson shows promise, it is often criticized for its reliance on massive, curated datasets, its high demands on computing infrastructure, and its limited adaptability to resource-poor or rural healthcare settings.Its use is largely limited to advanced hospital networks, restricting accessibility in regions that most urgently need cost-effective and portable diagnostic solutions.

[0007] Another category of existing technologies focuses on image-based diagnostics, where machine learning is applied to CT scans, MRIs, and echocardiograms to detect cardiac abnormalities early. Companies like Zebra Medical Vision and Cardiologs offer AI-powered imaging and ECG interpretation. Zebra Medical Vision, for example, has trained deep learning models to identify cardiovascular diseases in image datasets, while Cardiologs uses AI to automate arrhythmia detection from ECG recordings. These solutions are valuable because they reduce the diagnostic workload for cardiologists and streamline workflows. However, outside of specialized clinical settings, they reach their limits. Image-based AI tools require advanced diagnostic equipment, stable internet or cloud integration, and tightly controlled data collection.Therefore, they are unsuitable for community health centers, outpatient clinics, or emergency departments. Furthermore, they focus on narrow modalities—either imaging or ECG interpretation—rather than combining multimodal data streams that more accurately reflect a patient's overall cardiovascular risk.

[0008] Mobile health applications have become another important component of existing solutions, often targeting consumers directly. Platforms integrated into wearable devices like Fitbit, Apple Watch, or Samsung Health continuously collect physiological data, including heart rate variability, activity levels, and sleep patterns. Some of these systems have basic anomaly detection techniques to alert users to potential arrhythmias or elevated risk markers. While these solutions are appealing in terms of accessibility and widespread adoption, they have several drawbacks. First, their predictive models are often proprietary and not medically validated to the same standards as clinical diagnostic tools. This raises concerns about reliability and the possibility of false positives or negatives.Second, wearable devices face challenges regarding data accuracy, as consumer sensors are not calibrated to the precision of medical devices. Third, while continuous monitoring is possible, these systems rarely integrate comprehensive demographic, clinical, and lifestyle variables, significantly reducing their diagnostic value. In practice, wearable predictions are, at best, advisory and cannot be used as standalone diagnostic systems in emergency rooms or hospitals.

[0009] Academic research has also demonstrated a wide range of machine learning applications for predicting heart disease, using datasets such as the Cleveland Heart Disease dataset or larger cardiovascular cohorts. Techniques such as decision trees, random forests, logistic regression, and Naive Bayes have been extensively studied for their ability to classify patient data into risk categories. These studies consistently demonstrate that machine learning outperforms traditional risk assessment methods in terms of predictive accuracy. For example, random forest models can process high-dimensional input data and capture nonlinear relationships between variables, while Naive Bayes classifiers are computationally efficient and robust even in the presence of noise. However, these research contributions are often limited to proof-of-concept demonstrations.Many models are trained on small or undiversified datasets, limiting their generalizability to real-world populations. Overfitting is a common problem, where models perform exceptionally well on training datasets but fail when applied to new patient data. Furthermore, most academic studies fail to integrate their techniques into deployable devices or clinical decision support systems, creating a persistent gap between research findings and practical healthcare applications.

[0010] Another limitation of many existing solutions is the lack of real-time integration of multimodal patient data. Clinical decisions in emergency situations, such as suspected myocardial infarction, require the rapid assimilation of various data sources, including vital signs, electrocardiograms, biochemical markers, demographic information, and medical history. Current AI-driven systems rarely integrate all of these simultaneously. Instead, they specialize in either structured datasets, image data, or data from wearable sensors, but do not harmonize them within a single predictive framework. This fragmentation reduces their effectiveness in emergency situations where time-critical, holistic insights are essential. Furthermore, many systems are cloud-dependent and require high-bandwidth internet connections to function properly.This poses a significant obstacle in rural or underdeveloped regions where health infrastructure is limited and patients are disproportionately affected by cardiovascular diseases.

[0011] Interpretability is another drawback of existing AI solutions for predicting heart disease. Black-box models, particularly deep learning systems, are notorious for their lack of transparency, which undermines physicians' trust and hinders adoption. Medical professionals need not only predictions but also explanations for them—for example, which patient characteristics contributed most to the risk classification. Without interpretability, physicians are reluctant to rely on technical results for life-or-death decisions, especially in emergency medicine. Although explainable AI is a growing field of research, few existing commercial systems possess robust interpretability capabilities, leaving a critical gap in clinical applicability.

[0012] Another unresolved problem with current technologies for predicting heart disease is bias and fairness. Machine learning models are extremely sensitive to the data they are trained on. If the training dataset is focused on specific population groups, the resulting predictions can systematically disadvantage underrepresented groups such as women, ethnic minorities, or people from low-income families. This problem has already been documented in several AI applications in healthcare, where predictive models perform below average when applied to populations other than those represented in the training data. Particularly in the prediction of cardiovascular disease, this bias could lead to life-threatening discrepancies in diagnostic accuracy and treatment recommendations.Few existing systems explicitly address or mitigate such distortions, raising ethical concerns about their use.

[0013] Cost and scalability are further barriers to the adoption of current heart disease predictive systems. Advanced image-based AI tools or enterprise platforms like IBM Watson require significant investments in infrastructure, data storage, and specialist training. Therefore, they are best suited for tertiary care hospitals and well-funded healthcare networks, but impractical for small clinics, urgent care centers, or community health centers. Mobile health apps, while affordable and widely available, have limited medical credibility and integration with formal healthcare systems. A compromise solution that combines affordability, portability, and medical accuracy in a single device or system is lacking.

[0014] Finally, the clinical validation of existing solutions remains inconsistent. While academic prototypes demonstrate promising accuracy metrics such as precision, recall, and F1 scores, they are often not tested in rigorous real-world trials across various healthcare settings. Regulatory approval is a lengthy process, and many AI-based products remain in the pilot phase without widespread deployment. This results in promising technologies existing in isolation but lacking the robustness, trustworthiness, and integration required for routine clinical use.

[0015] In summary, existing solutions for predicting heart disease include statistical scoring systems, AI-powered enterprise platforms, image-based diagnostics, wearable health apps, and academic machine learning prototypes. While each category makes valuable progress, none offers a holistic, real-time, and accessible system that integrates multimodal data inputs into a medically validated, wearable device for emergency and routine care. Traditional statistical tools are outdated and inflexible, enterprise AI solutions are expensive and inaccessible, image-based systems are device-dependent, wearable devices lack clinical validity, academic models suffer from overfitting and poor applicability, and most systems lack interpretability, fairness, and real-time integration.These shortcomings underscore the urgent need for an invention like CardioAid, which fills these gaps by combining machine learning, multimodal data integration, portability, interpretability, and emergency-oriented design into a single, deployable device. Objectives of the invention

[0016] The invention aims to provide the following: 1. A machine learning device for real-time prediction of heart disease that integrates demographic data, lifestyle data, and physiological data of the patient. 2. A system that can embed multiple ML models, including Naive Bayes, Random Forest, Logistic Regression and Decision Tree, with automated model selection based on real-time validation metrics. 3. A device structure that includes data acquisition modules, a processing core, cloud connectivity, and an interactive display interface for clinicians. 4. Real-time integration with IoT-enabled wearable sensors for continuous monitoring and updating of predictive models. 5. A decision support mechanism that provides actionable recommendations for emergency care, thereby supporting rapid and accurate medical interventions. 6. Summary of the invention

[0017] The invention relates to a CardioAid device, a wearable device with integrated machine learning techniques for predicting heart disease. The device comprises a sensor acquisition unit that receives physiological data such as heart rate, ECG signals, and blood pressure; a user interface for inputting demographic and lifestyle parameters; and a computing engine with a trained Naive Bayes classifier optimized for real-time performance.

[0018] The device also features a model management module for comparing multiple machine learning models, secure local storage for patient records, and a cloud synchronization function for training large datasets. A display interface provides physicians with risk assessments, treatment recommendations, and confidence indices, while an emergency alert subsystem transmits real-time predictions to nursing staff and hospital systems.

[0019] This structure ensures portability and operational efficiency, enabling use in both hospitals and rural clinics.

[0020] The main objective of the present invention is to provide a machine learning-based device and system that enables real-time prediction of heart disease risk by integrating clinical, demographic, physiological, and lifestyle data into a unified computational model. The invention seeks to overcome the limitations of conventional risk assessment tools, such as statistical evaluation models, by employing adaptive machine learning classifiers that deliver dynamic, data-driven insights with significantly improved accuracy. A further objective of the invention is to provide a portable and robust device structure that can be deployed in a variety of healthcare settings, including hospitals, ambulances, and rural clinics where access to specialized cardiologists and advanced imaging systems may be limited.The invention also aims to establish a clinical decision support mechanism that not only predicts the risk of heart disease but also provides actionable recommendations and emergency alerts in order to reduce diagnostic delays and improve treatment outcomes.

[0021] A further objective of the invention is to ensure the scalability and adaptability of the system by supporting both standalone and cloud-integrated operating modes. This enables the device to operate reliably in resource-constrained environments while simultaneously allowing for continuous improvement of predictive performance through cloud-based updates, provided the infrastructure permits. The invention also aims to facilitate the seamless integration of wearable and IoT-enabled patient monitoring devices, enabling real-time data streams to be incorporated into the predictive models and thus ensuring continuous monitoring of high-risk individuals. Another important objective of the invention is interpretability through the integration of mechanisms that explain the contribution of individual patient characteristics to the predictive outcome.This strengthens the confidence of doctors and ensures that the system complements medical assessment rather than replacing it.

[0022] The invention is also intended to contribute to the equitable use of artificial intelligence in healthcare by integrating techniques for detecting and mitigating biases in predictive models, thereby ensuring fairness for different patient groups. A further objective is to reduce healthcare costs and optimize resource allocation in hospitals through improved early detection and the minimization of unnecessary interventions and misdiagnoses. Finally, the invention aims to create a system that is not only medically sophisticated but also accessible and user-friendly, seamlessly integrating into existing healthcare workflows. This will maximize its acceptance and impact in both modern and resource-poor healthcare facilities. BRIEF DESCRIPTION OF THE FIGURE

[0023] These and other features, aspects, and advantages of the present invention will be better understood if the following detailed description is read with reference to the accompanying drawing, in which the same symbols consistently represent the same parts. The following applies: Fig. Figure 1 shows a block diagram of a quantum-enhanced system for detecting fake news.

[0024] Experts will also recognize that the elements in the drawing are shown for the sake of simplicity and are not necessarily to scale. For example, the flowcharts illustrate the process by highlighting the main steps to enhance understanding of the aspects of this disclosure. Furthermore, with regard to the design of the device, one or more components of the device may be represented in the drawing by conventional symbols, and the drawing may show only the specific details relevant to understanding the embodiments of this disclosure, so as not to clutter the drawing with details that are readily apparent to those skilled in the art after reading this description. Detailed description of the invention

[0025] For a better understanding of the inventive principles, reference is made below to the embodiment shown in the drawing, which is described in specific terminology. However, this does not limit the scope of the invention. Changes and further modifications of the illustrated system, as well as further applications of the inventive principles, are possible, as would normally occur to a person skilled in the art in the field of invention.

[0026] It is clear to the person skilled in the art that the preceding general description and the following detailed description are exemplary and explanatory of the invention and are not intended as a limitation of it.

[0027] References in this specification to “an aspect”, “another aspect”, or similar expressions mean that a particular feature, structure, or property described in connection with the embodiment is included in at least one embodiment of the present disclosure. Therefore, occurrences of the expressions “in one embodiment”, “in another embodiment”, and similar expressions in this specification may all refer to the same embodiment, but need not.

[0028] The terms "includes," "include," or other variations thereof are intended to cover non-exclusive inclusion, such that a process or method that includes a list of steps may not only contain those steps but may also include other steps not expressly listed or inherent in such process or method. Likewise, the statement "includes..." in the case of one or more devices, subsystems, elements, structures, or components does not, without further limitations, preclude the existence of other devices, subsystems, elements, structures, components, or additional devices, subsystems, elements, structures, or components.

[0029] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as understood by a person skilled in the art in the field of the invention. The systems, methods, and examples provided herein serve only for illustration and are not to be construed as limitations.

[0030] Embodiments of the present disclosure are described in detail below with reference to the attached drawing.

[0031] In Fig.Figure 1 shows a block diagram of a CardioAid device that uses real-time machine learning to predict heart disease and support clinical decision-making. The System 100 includes a patient data acquisition module (102) configured to receive multimodal inputs, including physiological signals selected from electrocardiographic waveforms, blood pressure readings, heart rate values, and oxygen saturation values; demographic and lifestyle information, including age, sex, cholesterol levels, smoking habits, and family medical history; and a preprocessing engine (104) configured to perform data cleansing, imputation of missing values, categorical coding, and feature scaling, thus normalizing the input data and converting it into a format suitable for machine learning.a processing core (106) comprising a multi-core CPU / GPU system-on-chip coupled with a secure storage unit (106a) in an operational state, wherein the processing core is configured to execute a variety of pre-trained machine learning models, including Naive Bayes, Random Forest, logistic regression, and decision tree classifiers; a model evaluation unit (108) configured to calculate validation metrics such as precision, recall, F1 score, and area under the curve for each of the models and dynamically selects the model with optimal performance to generate real-time risk predictions for heart disease; a display interface (110) configured to present prediction results in the form of risk probability values, confidence indices, and actionable recommendations that correspond to clinical treatment guidelines;and a communication interface (112) configured to transmit emergency alerts corresponding to high-risk predictions to remote caregivers, hospitals and emergency response systems via wireless communication protocols, including Wi-Fi, Bluetooth and 4G / 5G cellular connections.

[0032] In one embodiment, the preprocessing engine (104) also includes a domain-specific feature engineering pipeline that generates derived features, including age group categorization, blood pressure variability indices, cholesterol ratio metrics, and heart rate variability derived from ECGs, wherein the features are optimized for machine learning-based prediction of cardiovascular risk.

[0033] In one embodiment, the processing core (106) also includes a real-time inference accelerator implemented using parallelized vectorized operations on the GPU subsystem, enabling the execution of machine learning models with a latency of less than 200 milliseconds, thus making predictions available in real time during clinical emergency workflows.

[0034] In one embodiment, the model evaluation unit (108) is also configured to perform a dynamic weighting of the outputs of multiple classifiers using a soft-voting ensemble method, with classifier-specific weights being updated in real time based on sliding power windows, thereby improving prediction stability for heterogeneous patient inputs.

[0035] In one embodiment, the display interface (110) is also configured to generate visualizations for interpretability, including feature importance charts, contribution heatmaps, and Shapley-based explanations, so that the clinician can verify which patient parameters contributed most significantly to risk prediction.

[0036] In one embodiment, the communication interface (112) also includes a secure data encryption module that implements AES-256 or higher-grade cryptographic protocols for encrypting patient data during wireless transmission, thus ensuring compliance with healthcare data protection regulations, including HIPAA and GDPR.

[0037] In one embodiment, it also includes a dual-mode operating framework, wherein in a standalone mode the system runs locally stored, embedded-in-the-memory machine learning models without external connectivity, and in a cloud-integrated mode the system synchronizes with a cloud-based training pipeline to continuously retrain and update the models with aggregated patient datasets.

[0038] In one embodiment, the patient data acquisition module (102) also includes connectivity to wearable IoT-enabled biosensors configured to stream continuous physiological data at sampling rates of at least 200 Hz for ECG signals and at least 1 Hz for blood pressure and heart rate values, the data being time-synchronized and merged with static demographic inputs to improve the temporal accuracy of predictions.

[0039] In one embodiment, the predictive results generated by the system are also linked to a clinical decision support knowledge database containing treatment protocols for cardiovascular emergencies, so that the system provides context-specific recommendations, including immediate referral, pharmacological intervention, or guidance on emergency transport.

[0040] In one embodiment, the secure storage unit (106a) also includes a modular audit trail subsystem that stores prediction results, model versions, and protocols of interaction with clinicians in immutable datasets anchored by blockchain-based cryptographic hashing, thereby ensuring the traceability and accountability of diagnostic predictions in clinical audits.

[0041] The present invention describes a system and a device for the real-time prediction of heart disease. It is based on the integration of multimodal patient data with advanced machine learning techniques and is designed for use in both hospitals and emergency care settings. The system is based on the patient data acquisition module, which acts as the primary input portal. This module is configured to acquire a variety of data sources, including physiological measurements from biosensors such as ECG electrodes, non-invasive blood pressure monitors, and heart rate sensors, as well as wearable IoT-enabled devices capable of continuously transmitting real-time data. In addition to physiological signals, the acquisition module also processes demographic and lifestyle data, including age, sex, cholesterol levels, blood glucose levels, smoking habits, and family medical history.These datasets form the basis for the system's predictive models. The data acquisition unit incorporates time synchronization mechanisms to ensure that continuous data streams from wearable devices align with static demographic data, thereby improving the accuracy of the predictive framework.

[0042] Once the data is received, it is forwarded to the preprocessing engine, which ensures that the raw data is converted into a standardized format suitable for machine learning. The preprocessing engine performs the imputation of missing values ​​using statistical or probabilistic methods, such as k-nearest neighbor-based imputation for numerical data and mode-based substitution for categorical data. The system also performs categorical coding using one-hot or ordinal encoders and converts non-numerical attributes, such as chest pain type or smoking status, into representations suitable for machine learning models. Feature scaling and normalization routines are applied to fit heterogeneous features, such as cholesterol and blood pressure levels, to comparable numerical ranges, thus minimizing biases during model training and inference.Beyond basic preprocessing, the engine employs domain-specific feature engineering. It generates derived features, including cholesterol ratios, heart rate variability indices calculated from ECG signals, and age group categorization bands. These technical features enrich the dataset and enhance the models' discriminatory power. They ensure that the predictive techniques account for subtle risk factors that might not be apparent in the raw data alone.

[0043] The preprocessed data is then forwarded to the processor core, which forms the computational heart of the invention. This core is based on a high-performance CPU / GPU system-on-chip optimized for parallelized machine learning inference tasks. Its secure memory integrates several pre-trained classifiers, including Naive Bayes, Random Forest, Logistic Regression, and Decision Tree methods. Each classifier interprets the processed dataset independently and determines its own probability value for the likelihood of heart disease. The Naive Bayes model is based on the principle of conditional independence and calculates posterior probabilities by applying Bayes' Theorem to the entire feature set. This provides computational efficiency and robustness against noise.Logistic regression calculates probabilities using the logistic sigmoid function and is valued for its interpretability and suitability for binary classification tasks. The decision tree classifier uses recursive partitioning to divide the feature space into decision nodes, thus creating a set of if-then rules for classification. Random Forest, an ensemble method, generates multiple decision trees and aggregates their results by majority vote, reducing the risk of overfitting and improving overall accuracy.

[0044] Once the classifiers deliver their results, these predictions are forwarded to the model evaluation unit, which validates and selects the optimal predictive model. The evaluation unit continuously monitors the performance of each classifier using moving performance windows of current patient data. Validation metrics are calculated for each technique, including accuracy, precision, recall, F1 score, and area under the ROC curve (AUC). Based on these metrics, the evaluation unit either selects the single best-performing model or applies a dynamically weighted ensemble strategy. In this ensemble approach, weights are assigned to the classifiers proportional to their validation performance, and a weighted average of their probability scores is calculated to achieve a stable, robust prediction.This dynamic model selection ensures that the system is adaptable and can optimize predictions in real time, even with different and evolving patient profiles.

[0045] The system's predictive results are transmitted to the display interface, which is designed for use by physicians in emergency or routine care. The interface presents a clear risk probability score along with a confidence index and assigns these results to clinically relevant categories such as low, medium, or high risk. To enhance interpretability and physician confidence, the display interface includes visualization features such as feature importance charts, contribution heatmaps, and Shapley score-based explanations. These visualizations provide insight into the most influential features affecting the prediction, highlighting, for example, whether elevated cholesterol levels, abnormal ECG patterns, or family history contributed most to the classification.This transparency ensures that the system supports medical judgment rather than replacing it, and enables doctors to make informed treatment decisions.

[0046] In addition to the predictive display, the system features an emergency decision support layer. This layer is integrated with a clinical knowledge base containing treatment guidelines for cardiovascular events. If the system identifies a high-risk patient, the decision support module not only highlights the risk but also generates actionable recommendations such as immediate pharmacological interventions, referral for further diagnostics, or transfer to an emergency department. Furthermore, the emergency alert subsystem transmits these findings via secure wireless communication to external caregivers, hospital networks, and emergency medical services. To ensure compliance with healthcare data privacy regulations, all transmitted data is encrypted using state-of-the-art cryptographic standards such as AES-256. This guarantees patient confidentiality and data integrity during transmission.

[0047] The invention also offers dual-mode functionality, ensuring adaptability to different healthcare infrastructures. In standalone mode, the device operates with embedded models stored locally in its secure memory. This enables predictions without internet access—a crucial feature in rural or resource-constrained environments. In cloud-integrated mode, the system synchronizes with external servers, enabling continuous training of the models using large, aggregated datasets. This improves predictive accuracy and ensures current clinical relevance. This dual capability guarantees independent operation of the device in emergencies and continuous improvement in well-connected hospital networks.

[0048] To ensure the traceability and verifiability of diagnostic predictions, the secure storage unit of the processing core also includes a blockchain-based audit trail subsystem. All prediction results, the version of the machine learning model used, and the clinician's interactions with the interface are stored in immutable datasets secured by cryptographic hashing. This not only ensures transparency during clinical audits but also protects against manipulation and guarantees compliance with legal regulations for AI-powered healthcare devices.

[0049] In summary, the system disclosed in this invention offers a comprehensive framework that integrates multimodal patient data acquisition, robust preprocessing, machine learning-based prediction, dynamic model evaluation, interpretable outcome visualization, secure communication, and emergency decision support. By combining these elements in a single device, the invention addresses the shortcomings of existing solutions, which are fragmented, opaque, or inaccessible. The system's technical foundation, particularly the integration of multiple classifiers with real-time validation and interpretability mechanisms, ensures both technical robustness and clinical applicability.The invention thus achieves the crucial objectives: improving diagnostic accuracy, supporting rapid emergency decisions, reducing healthcare costs, and extending advanced cardiovascular risk prediction capabilities to environments where conventional solutions are impractical or unavailable. System architecture

[0050] The CardioAid device integrates hardware and software components into a unified structure. The hardware subsystem includes: • Module for recording patient data: Interfaces to clinical instruments such as ECG electrodes, non-invasive blood pressure monitors and optional portable devices. • Processing core: A high-performance CPU / GPU system-on-chip (SoC) optimized for ML inference, running pre-trained models stored in secure local memory. • Communication interfaces: Wireless modules (Wi-Fi, Bluetooth, 4G / 5G) for integration with hospital networks, cloud servers and IoT wearables. • User interaction interface: A medical-grade touchscreen display for entering patient data and viewing prognoses.

[0051] The software subsystem includes: • Preprocessing engine: Cleans and standardizes patient input data and performs missing value treatment and feature transformations. • Model evaluation unit: Runs Naive Bayes, Random Forest, Logistic Regression and Decision Tree classifiers in parallel and selects the optimal model based on real-time validation metrics such as precision, recall and F1 score. • Prediction module: Outputs probability values ​​for heart disease risk with confidence levels that are assigned to a scale to support clinical decisions (e.g., low risk, medium risk, critical). • Emergency decision support level: Triggers warnings, recommendations and links to hospital databases for rapid triage. Device design

[0052] The CardioAid device is designed as a portable diagnostic station with a robust housing for clinical and emergency services. The housing features compartments for sensor connections, a modular battery unit for field use, and cooling systems to maintain processing stability.

[0053] The touchscreen interface guides clinical staff through data entry and sensor connections, while the backend processing core automatically analyzes the data using trained machine learning models. The device features dual-mode operation: 1. Standalone mode: Provides local forecasts without internet access using embedded models. 2. Cloud-integrated mode: Synchronizes with cloud servers for model updates and more comprehensive data aggregation.

[0054] The present invention falls within the technical fields of medical technology, computer-aided health technologies, and machine-learning diagnostic systems. It relates in particular to the design and development of a hardware- and software-integrated system for real-time prediction of the risk of cardiovascular diseases and for providing clinical decision support in emergency and routine care. The invention encompasses advances in the acquisition of biosensor data, the application of machine learning techniques on embedded processors, the secure communication of health data, and interpretable artificial intelligence for medical diagnostics.It also lies at the interface of health informatics, biomedical engineering and emergency medicine, and addresses the urgent need for precise, portable and understandable diagnostic devices that function reliably in both modern hospital infrastructures and resource-constrained rural health systems.

[0055] The drawing and the preceding description show examples of embodiments. Those skilled in the art will recognize that one or more of the described elements can be combined to form a single functional element. Alternatively, certain elements can be divided into several functional elements. Elements of one embodiment can be added to another embodiment. For example, the sequence of the processes described here can be changed and is not limited to the manner described here. Furthermore, the actions of a flowchart need not be implemented in the sequence shown; nor does it necessarily have to be performed by all actions. Actions that are not dependent on other actions can also be performed in parallel with the other actions. The scope of the embodiments is in no way limited by these specific examples.Numerous variations are possible, whether explicitly stated in the specification or not, such as differences in structure, dimensions, and material use. The range of embodiments is at least as broad as specified in the following claims.

[0056] Advantages, further benefits, and problem solutions have been described above with reference to specific embodiments. However, the advantages, benefits, problem solutions, and all components that can lead to an advantage, benefit, or solution occurring or becoming more apparent are not to be construed as critical, necessary, or essential features or components of individual or all claims. REFERENCES 100 Cardioaid: Real-time device for predicting heart disease and supporting clinical decisions based on machine learning. 102 Patient Data Collection Module 104 Preprocessing Engine 106 Processing core 106a Secure Storage Unit 108 Model evaluation unit 110 Display interface 112 Communication interface

Claims

[1] A real-time prediction system for heart disease and clinical decision support, the system includes: a patient data acquisition module configured to receive multimodal inputs, including physiological signals selected from electrocardiographic waveforms, blood pressure readings, heart rate readings, and oxygen saturation readings, as well as demographic and lifestyle information, including age, gender, cholesterol levels, smoking habits, and family medical history; a preprocessing engine configured to perform data cleansing, imputation of missing values, categorical coding, and feature scaling, so that the input data is normalized and converted into a format suitable for machine learning; a processing core comprising a multi-core CPU / GPU system-on-chip operationally coupled with a secure storage unit, wherein the processing core is configured to execute a variety of pre-trained machine learning models, including Naive Bayes, Random Forest, Logistic Regression and Decision Tree classifiers; a model evaluation unit configured to calculate validation metrics such as precision, recall, F1 score and area under the curve for each of the models and dynamically select the model with optimal performance to generate real-time predictions for the risk of heart disease; a display interface configured to present predictive results in the form of risk probability values, confidence indices, and actionable recommendations that are mapped to clinical treatment guidelines; and a communication interface configured to transmit emergency alerts based on high-risk predictions to remote caregivers, hospitals, and emergency response systems via wireless communication protocols such as WLAN, Bluetooth, and 4G / 5G cellular connections. [2] System according to claim 1, wherein the preprocessing engine further comprises a domain-specific feature engineering pipeline that generates derived features, including age group categorization, blood pressure variability indices, cholesterol ratio metrics and heart rate variability derived from ECGs, wherein the features are optimized for machine learning-based prediction of cardiovascular risk. [3] System according to claim 1, wherein the processing core further comprises a real-time inference accelerator implemented using parallelized vectorized operations on the GPU subsystem, enabling the execution of machine learning models with a latency of less than 200 milliseconds, so that predictions are available in real time during clinical emergency workflows. [4] System according to claim 1, wherein the model evaluation unit is further configured to perform a dynamic weighting of the outputs of multiple classifiers using a soft voting ensemble method, classifier-specific weights being updated in real time based on sliding power windows, thereby improving the predictive stability for heterogeneous patient inputs. [5] System according to claim 1, wherein the display interface is further configured to generate visualizations for interpretability, including feature importance charts, contribution heatmaps and explanations based on the Shapley score, so that the clinician can verify which patient parameters have contributed most significantly to risk prediction. [6] System according to claim 1, wherein the communication interface further comprises a secure data encryption module that implements AES-256 or higher cryptographic protocols for encrypting patient data during wireless transmission, thus ensuring compliance with healthcare data protection regulations, including HIPAA and GDPR. [7] System according to claim 1, further comprising a dual-mode operating framework, wherein in a standalone mode the system executes locally stored, embedded-in-the-memory machine learning models without external connectivity and in a cloud-integrated mode the system synchronizes with a cloud-based training pipeline to continuously retrain and update the models with aggregated patient datasets. [8] System according to claim 1, wherein the patient data acquisition module further comprises connectivity to wearable IoT-enabled biosensors configured to stream continuous physiological data with sampling rates of at least 200 Hz for ECG signals and at least 1 Hz for blood pressure and heart rate values, wherein the data are time-synchronized and merged with static demographic inputs to improve the temporal accuracy of the predictions. [9] System according to claim 1, wherein the predictive results generated by the system are also linked to a clinical decision support knowledge base containing treatment protocols for cardiovascular emergencies, so that the system provides context-specific recommendations, including immediate referral, pharmacological intervention or guidance on emergency transport.

Images