FORM 2
THE PATENT ACT 1970
(39 OF 1970)
&
The Patents Rules, 2003
PROVISIONAL/COMPLETE SPECIFICATION

1. TITLE OF THE INVENTION
Machine Learning-Driven System and Method for Predictive Diagnosis of Heart Disease
and Risk Assessment
2. APPLICANT(S): 1. Mrs. Swetha S
(a) NATIONALITY: Indian
(b^ ADDRESS: Research Scholar, Department of Computer Science and Engineering,
UVCE, Bengaluru 560001.
APPLICANT(S): 2. DrSHManjula
a) NATIONALITY: Indian
b) ADDRESS: Professor, Department of Computer Science and Engineering. UVCE,
Bengaluru 560001.

3. PREAMBLE TO THE DESCRIPTION
COMPLETE
The following specification particularly describes
the invention and the manner in which it is to be
performed.

4.
DESCRIPTION
03-Dg(C-2024/143304/202441095000/Form 2(Title Page)
FIELD OF INVENTION
The present invention relates to the field of medical diagnostics, specifically focusing on the predictive diagnosis and risk assessment of cardiovascular diseases. This invention is centered on utilizing advanced machine learning algorithms and data processing techniques to develop a robust, reliable and accurate system for identifying heart disease risk. The system integrates multiple sources of health data, including Electronic Health Records (EHR), wearable device data, medical imaging and public health datasets, creating a comprehensive and holistic approach to cardiovascular risk prediction. This field combines advancements in artificial intelligence, data science and healthcare technology to address the growing need for preventive healthcare solution, particularly in managing chronic conditions like heart disease. This invention operates at the intersection of predictive analytics, machine learning and clinical decision support systems (CDSS). The goal is to leverage cutting-edge machine learning methodologies, including deep learning architectures and ensemble learning techniques, to identify and interpret risk factors associated with heart disease. By utilizing multiple data streams, the system provides an accurate prediction of heart disease risk, enabling healthcare professionals to proactively identify high-risk patients and implement early intervention strategies.

In this field, machine learning models are tailored for clinical application, emphasizing interpretability, accuracy and real-time integration within hospital information systems (HIS) and electronic medical records (EMR). The system supports clinical workflows by delivering actionable risk scores and diagnostic recommendations, which facilitate timely clinical decisionmaking. This approach enables healthcare providers to shift from reactive treatment to proactive prevention, improving patient outcomes and reducing healthcare costs associated with advanced stages of heart disease. The field of invention thus encompasses healthcare technologies aimed at enhancing predictive accuracy, facilitating preventive care and enabling personalized treatment strategies within cardiology. It leverages developments in artificial intelligence, data interoperability and secure healthcare data management, ultimately contributing to the ongoing transformation of healthcare systems into data-driven, proactive and personalized models of care.

BACKGROUND OF THE INVENTION:
Cardiovascular diseases (CVDs), including heart disease, are among the leading causes of death globally, with millions of lives lost annually due to late detection and inadequate risk assessment. Traditional methods for diagnosing heart disease, such as clinical evaluations, blood tests, electrocardiograms (ECGs) and imaging studies, often identify the disease only after symptoms become apparent or significant damage has already occurred. This reactive approach leads to higher treatment costs, increased patient suffering and limited scope for preventive interventions. In recent years, the integration of data-driven technologies in healthcare, particularly artificial intelligence (Al) and machine learning (ML), has created new opportunities for predictive diagnostics. Machine learning algorithms have shown considerable promise in identifying subtle patterns and correlations in large datasets, which are often missed by traditional diagnostic methods. Predictive analytics applied to heart disease has emerged as a valuable tool to address limitations in early detection and enable proactive treatment strategies. However, existing predictive systems face challenges, including data integration from multiple sources, high dimensionality and the need for model interpretability in clinical settings. Additionally, the presence of imbalanced datasets—where cases of heart disease may be underrepresented compared to non-cardiovascular health data—often leads to biased predictions that limit accuracy and reliability. With the widespread use of electronic health records (EHRs).
\ *
wearable health devices and public health datasets, a wealth of patient-specific information is now accessible. However, harnessing this data effectively to deliver accurate and actionable insights requires robust preprocessing, feature extraction and predictive modeling. Furthermore, the data must be cleaned, normalized and standardized across sources to ensure consistency and usability. An optimized system that incorporates data from wearable devices (e.g., heart rate monitors), imaging (e.g., MRI scans) and clinical records (e.g., demographics, lifestyle, medical history) could drastically improve the timeliness and accuracy of heart disease risk assessments.

Despite significant advancements, challenges remain in integrating diverse data types, developing models that balance complexity with clinical interpretability and ensuring secure, real-time access to predictive insights within healthcare environments. Existing solutions often lack scalability and compatibility' with hospital information systems (HIS) and electronic medical records (EMR), limiting their adoption in real-world clinical sellings. Moreover, traditional machine learning models may overlook the intricacies of individual health factors, reducing the potential for personalized recommendations and tailored interventions. The present invention addresses these limitations by introducing an optimized machine learning-based system and method for predictive diagnosis of heart disease. This invention is designed to collect and process multi-source data, extract and prioritize key risk factors and generate an interpretable risk score that assists healthcare providers in identifying high-risk patients. The system employs
advanced algorithms, such as meta-classifier ensembles, deep learning models like CNN-BiLSTM for ECG analysis and U-Net for image-based diagnostics, to enhance predictive accuracy while maintaining clinical relevance. Moreover, the system is integrated with HIS and EMR platforms, enabling real-time application in clinical environments and ensuring data privacy and security. This invention fills a critical gap in current healthcare systems by providing a comprehensive, scalable and interpertable diagnostic tool for early heart disease prediction. It empowers healthcare providers with actionable insights that facilitate early interventions, personalized treatment plans and better patient outcomes. By shifting from a reactive to a proactive approach, this invention aims to transform cardiovascular care and address the global health burden associated with heart disease.

SUMMARY OF THE INVENTION
The present invention provides a novel system and method for predictive diagnosis of heart disease, utilizing optimized machine learning algorithms to offer reliable, interpretable, and actionable risk assessments in clinical environments. The system is designed to collect, process, and analyze multi-source patient data to identify individuals at elevated risk of heart disease, facilitating early intervention and personalized treatment plans. The invention features a Data Collection Layer that aggregates patient information from diverse sources, including Electronic Health Records (EHR), wearable device data, MR! scan reports, and public health datasets. This multi-faceted data intake enables a comprehensive view of each patient's health status, incorporating demographics, lifestyle factors, physiological metrics, and medical history, which are critical in assessing heart disease risk accurately. A Data Preprocessing Layer is then employed to clean, normalize, and balance the collected data. This includes removing noise, standardizing data scales, and applying techniques such as Synthetic Minority Over-Sampling Technique (SMOTE) or Random Under-Sampling with Boosting (RUSBoost) to manage class imbalances. This standardized dataset forms a reliable foundation for the system’s predictive
analysis. To optimize model performance, the invention includes a Feature Extraction and Selection Layer. This layer uses dimensionality reduction techniques, such as Principal Component Analysis (PCA), to identify and retain the most relevant risk factors. By focusing on key features, the system enhances predictive accuracy and efficiency while minimizing computational complexity. The invention further comprises a Model Training and Optimization Layer that utilizes a variety of machine learning models. These include a meta-classifier ensemble for enhanced prediction robustness, a CNN-BiLSTM model with attention mechanisms to analyze ECG data, and a U-Net model for MRI-based image analysis. Hyperparameter tuning methods, including grid search and random search, are applied to ensure optimal model performance. Once trained, the system selects the best-performing model based on diagnostic accuracy. A Deployment and integration layer allows the system to seamlessly interface with hospital information systems (HIS) and Electronic Medical Records (EMR), facilitating real-time application in clinical environments while ensuring compliance with data privacy regulations. This integration enables
a
healthcare providers to access predictive insights within their existing workflow, enhancing the system’s practical utility. The Prediction and Risk Scoring Layer processes patient data through the optimized model to generate a quantitative risk score. Patients are categorized into risk levels— such as low-risk, medium-risk, and high-risk—enabling healthcare professionals to identify high- priority cases quickly and accurately. Finally, an Output and Decision Support Layer translates the risk ^scores into actionable insights and diagnostic recommendations. This layer provides healthcare providers with specific guidance on preventive measures, follow-up diagnostics, or tailored treatment options, supporting proactive care and personalized interventions.

Overall, this invention addresses the need for an accurate, scalable, and interpretable diagnostic tool for heart disease. By integrating data from multiple sources, optimizing machine learning algorithms, and ensuring seamless deployment in clinical settings, the system empowers healthcare providers with data-driven insights to facilitate early detection, preventive care, and improved patient outcomes. The invention thus represents a transformative approach to cardiovascular healthcare, shifting the focus from reactive treatment to proactive prevention.

SHORT DESCRIPTION OF DIAGRAMS:
Figure 200: Describes the System Architecture of the Machine Learning-Driven System and Method for Predictive Diagnosis of Heart Disease and Risk Assessment.

DETAILED DESCRIPTION OF DIAGRAMS:
FIG. 200 is a system architecture diagram illustrating a Machine Learning-Driven System and Method for Predictive Diagnosis of Heart Disease and Risk Assessment.The Heart Disease Prediction System 200 includes a Data Collection Layer 140, which contains various inputs for the system, including Electronic Health Records (EHR) 141, Wearable Device Data, MRI Scan Report of patients 142 and Public Health Datasets 143 (c.g., UC1 repository) which has information

about patient demographics, physiological metrics, medical history, lifestyle factors. 144 is a Data Preprocessing Layer that has a series of steps mentioned below:
• Data Cleaning: Handies missing values and removes noisy data.
• Normalization: Ensures consistent data scale.
• Data Balancing: Techniques like SMOTE or RUSBoost for handling class imbalance.
• Output: Cleaned, normalized dataset ready for analysis.

Feature Extraction & Selection Layer 145 is a technique to identify key risk factors. Methods like PCA to optimize performance have been used. Model Training and Optimization Layer 146 with Al Module, which performs the following:
• Algorithms Used: Meta-classifier ensemble, CNN-BiLSTM with attention for ECG analysis, U-Net for imaging data.
• Optimization Techniques: Hyperparameter tuning (e.g.. Grid Search, Random Search)
• Model Selection: Chooses the best-performing model based on accuracy and precision.

Deployment and Integration Layer 147 integrates with Hospital Information Systems (HIS) and Electronic Medical Records (EMR). Prediction and Risk Scoring Layer 148 Generates a risk score or classification for heart disease. Output and Decision Support 149 is used for Risk prediction and diagnostic suggestions and enables early intervention and personalized treatment recommendations. FIG. 200 illustrates the comprehensive structure of the Heart Disease Prediction System 200, consisting of multiple layers that facilitate data intake, processing, feature extraction, model training, prediction and deployment in clinical environments.

Data Collection Layer (140):
The Data Collection Layer is the foundational component of the Heart Disease Prediction System, designed to aggregate diverse and comprehensive datasets required for accurate and reliable predictions. This layer integrates multiple sources of patient-related data to create a holistic profile, enabling the system to identify and assess risk factors for heart disease effectively. Key data sources within this layer include Electronic Health Records (EHR) (141). Wearable Device Data and MRI Scan Reports (142) and Public Health Datasets (143).

1. Electronic Health Records (EHR) (141): EHRs provide an extensive, digital history of a patient’s medical information, encapsulating prior diagnoses, treatment records, prescribed medications, allergies. family hiedicaFliistor)' and previous laboratory testresults. EHRs play a critical role in chronic disease management as they enable the predictive system to understand long-term patterns and recurring health issues, essential in accurately predicting heart disease risks based on historical medical data. This continuous documentation of a patient’s health over time allows the system to analyze trends and detect underlying health patterns that could indicate cardiovascular risk factors.

2. Wearable Device Data and MRI Scan Reports (142): In modern healthcare, wearable devices like smartwatches and fitness trackers have emerged as valuable tools for capturing real-time physiological metrics such as heart rate, blood pressure, physical activity, sleep patterns and oxygen saturation. These devices provide the system with continuous, real-time data that reflects the patient's current physiological state, offering insights into daily fluctuations in heart health parameters. Additionally, MR! scan reports of patients contribute critical anatomical and functional information, particularly regarding the structure and functionality of cardiac tissues. MRI data can help the system detect structural abnormalities in the heart, such as variations in chamber size or wall thickness, that may correlate with a higher risk of heart disease. The integration of wearable data and MRI reports provides the system with both real-time and in-depth physiological insights, enhancing the prediction model’s accuracy.

3. Public Health Datasets. (143): This component laps into large-scale public datasets, such as those available from repositories like the UCI Machine Learning Repositor). These datasets offer aggregated information on a broad population level, including demographic data (age, gender, socioeconomic status), physiological attributes (e.g., average cholesterol levels, BMI distribution) and lifestyle factors (e.g., smoking, alcohol consumption, exercise habits). Public health datasets enrich the system with contextual data that may not be captured in individual health records or wearable data. This data helps in building a more comprehensive understanding of the environmental and social determinants of heart disease, allowing the system to compare individual patient data against broader population trends and identify risk factors that may be more prevalent in certain demographic or lifestyle groups.

By combining these three sources, the Data Collection Layer (I40) enables the Heart Disease Prediction System to build a robust and multi-faceted dataset, integrating both individual and population-level insights. This comprehensive data foundation allows the system to make more precise predictions and support better-informed clinical decisions.

Data Preprocessing Layer (144):
The Data Preprocessing Layer is a crucial component in the Heart Disease Prediction System, responsible for refining and preparing raw data for analysis, ensuring it is suitable for accurate model training and prediction. This layer focuses on enhancing data quality by addressing common issues that could impair the predictive accuracy of the machine learning models. Key steps within this layer include Data Cleaning, Normalization and Data Balancing. These processes transform diverse inputs from the Data Collection Layer into a cohesive, reliable dataset that lays the foundation for robust predictive modeling.

I.
Data Cleaning: This step is essential to ensure that the dataset is free from errors, inconsistencies and inaccuracies. Data cleaning involves addressing missing values, such as filling in gaps or removing records with incomplete information, depending on the nature and extent of the missing data. Additionally, noise reduction techniques, are applied to filter out irrelevant or erroneous data points, which could otherwise distort the predictive accuracy of the model. For instance, outliers or artifacts in wearable device data or errors in EHR entries are identified and corrected. By performing data cleaning, the system enhances the reliability of its dataset, minimizing the risk of misleading predictions caused by data inaccuracies.

2. Normalization: Given that the system handles data from varied sources—ranging from physiological metrics to lifestyle factors—data normalization is performed to maintain consistent data scales across all features. Normalization adjusts the values of numeric data fields to a common scale, typically between 0 and 1 or -1 and I, without distorting differences in the data’s ranges. For example, heart rate data from wearable devices and cholesterol levels from medical records are standardized to prevent any single feature from disproportionately influencing the prediction model due to its scale. This step ensures that the system processes data uniformly, enhancing the model's stability and convergence during training.

3. Data Balancing: Since heart disease is often less prevalent than other health conditions, datasets may exhibit class imbalances, with fewer positive cases (patients diagnosed with heart disease) than negative cases. To address this, data balancing techniques like
; SMOTE (Synthetic Minority Over-sampling Technique) or RUSBoost (Random
. Under-Sampling with Boosting) are applied to equalize the distribution of classes. These techniques either create synthetic samples for underrepresented classes or down-sample the majority class, respectively. By balancing the dataset, the system mitigates the risk of b bias toward the majority class, enabling the prediction model to effectively recognize and i1 classify instances of heart disease, regardless of their relative scarcity in the dataset. pF Output: The Data Preprocessing Layer outputs a dataset that is cleaned, normalized and balanced, ensuring it is well-prepared for the subsequent stages of feature extraction, model training and predictive analysis. This high-quality, preprocessed dataset forms the backbone of the predictive system, contributing significantly to the accuracy and robustness of heart disease risk assessment.

The Data Preprocessing Layer (144) thus plays an indispensable role in transforming raw, heterogeneous data into a structured, reliable format, allowing the Heart Disease Prediction System to operate with greater precision and effectiveness in clinical settings.

Feature Extraction & Selection Layer (145):
The Feature Extraction & Selection Layer is a pivotal component in the Heart Disease Prediction System, tasked with isolating the most relevant information from the vast datasets collected and preprocessed in previous layers. This layer identifies critical risk factors that contribute to heart disease prediction, effectively enhancing model performance while minimizing the computational load: By focusing on the most informative features, the system can achieve high predictive accuracy without unnecessary complexity. Two primary processes in this layer are Feature Extraction and Feature Selection.

1. Feature Extraction: This process involves transforming raw data into meaningful and compact representations that are more interpretable and computationally efficient for the model. Feature extraction techniques, such as Principal Component Analysis (PCA). decompose complex, high-dimensional data into a set of uncorrelated components that capture the essential variance in the dataset. By doing so. the system reduces redundancy
_ __ and consolidates related variables into principal components that summarize the information with minimal loss. For example, physiological metrics from EHRs or wearable devices, which may include multiple overlapping indicators of cardiovascular health (e.g.. heart rate, blood pressure, cholesterol levels), are condensed into fewer, highly informative features. This not only enhances model efficiency but also aids in the interpretability of results by providing a distilled view of the most significant health indicators.

2. Feature Selection: In addition to feature extraction, this layer employs feature selection techniques to identify and retain only the most impactful risk factors while discarding irrelevant or redundant features. Feature selection algorithms, often leveraging statistical methods or machine learning models (e.g.. mutual information scores, recursive feature elimination), assess each feature's contribution to the target prediction—heart disease risk in this case. By focusing solely on features that provide substantial predictive power, the system avoids overfitting and enhances model generalizability. For instance, factors like age, BMI, lifestyle habits and specific biomarkers may emerge as key predictors, while less impactful features are removed to streamline the model.

The combined use of Principal Component Analysis (PCA) for feature extraction and advanced feature selection techniques significantly improves the system’s performance by reducing dimensionality and focusing on essential information. This process not only accelerates model training and inference but also enhances prediction accuracy by removing noise and irrelevant data, allowing the Heart Disease Prediction System to concentrate on the risk factors that matter most.

In summary, the Feature Extraction & Selection Layer (145) effectively distills the dataset into a set of critical, non-redundant features, facilitating an optimized, high-performing prediction model that is both computationally efficient and clinically interpretable. This refined feature set is then passed to the Model Training and Optimization Layer for further processing.

Model Training and Optimization Layer (146):
The Model Training and Optimization Layer serves as the core of the Heart Disease Prediction
System, where advanced machine learning algorithms are employed, optimized and evaluated to
deliver high-accuracy predictions. This layer is responsible for transforming the refined features
from the Feature Extraction & Selection Layer into predictive models capable of identifying
heart disease risk with precision. The main components within this layer include Algorithm
Selection. Optimization Techniques and Model Selection, each of which plays a vital role in
achieving a robust and reliable diagnostic system.

Algorithms Used: This layer incorporates a combination of sophisticated algorithms u specifically chosen for their ability to handle diverse data types and provide enhanced
predictive power. Key algorithms include:

o Meta-classifler Ensemble: A meta-classifier ensemble approach integrates multiple base classifiers to improve model accuracy and reduce the likelihood of errors that individual models might produce. By leveraging the strengths of different algorithms, the ensemble method enhances overall diagnostic accuracy and robustness. This is particularly valuable in healthcare applications, where the combination of different models helps mitigate the risk of misclassification.

o CNN-BiLSTM with Attention for ECG Data: Convolutional Neural Networks (CNNs) paired with Bidirectional Long Short-Term Memory (BiLSTM)
networks, enhanced with an attention mechanism, are used for analyzing ECG data. CNNs extract spatial features from ECG signals, while BiLSTM captures temporal dependencies in the data, which is crucial for identifying patterns associated with cardiac abnormalities. The attention mechanism further enhances this by focusing on the most relevant portions of the ECG data, enabling the model to make more precise predictions of heart disease.

o U-Net for Image-Based Analysis: U-Net, a deep learning architecture widely used in medical imaging, is applied to analyze images, such as MRI scans, for any structural anomalies in the heart. U-Nevs ability to capture both fine and coarse
details in images makes it particularly effective for segmenting heart structures and identifying features indicative of heart disease. This approach provides an additional layer of diagnostic support, especially in cases where image-based data is available.

2. Optimization Techniques: To maximize model performance, the system employs optimization techniques like Hyperparameter Tuning. This involves adjusting model parameters, which control aspects such as learning rates, regularization strength and network architecture to find the most effective configuration for each algorithm. Grid Search and Random Search are two common methods used:
o Grid Search systematically evaluates a predefined set of hyperparameters across
a grid, ensuring thorough exploration of possible configurations.
o Random Search selects random combinations of hyperparameters within specified ranges, often yielding optimal results more efficiently for complex models. These optimization techniques ensure that each model operates at its best,
improving the system’s overall diagnostic accuracy and responsiveness.

3. Model Selection: Once training and optimization are complete, the system evaluates each model’s performance to select the most accurate and reliable one. This selection process is based on key diagnostic metrics such as accuracy, precision, sensitivity and specificity, with a particular focus on accuracy to ensure the highest probability of correct predictions. The best-performing model is then chosen as the final predictive model for heart disease diagnosis. This selection process also considers factors like computational efficiency and interpretability, aiming to balance performance with practical deployment considerations in a clinical setting.

In summary', the Model Training and Optimization Layer (146) synthesizes multiple advanced
algorithms, fine-tunes their configurations and evaluates their performance, ultimately
identifying the most effective predictive model for heart disease. This optimized model is then
passed to the next stages for integration and deployment, providing the Heart Disease Prediction
System with a high level of diagnostic precision and reliability.

Deployment and Integration Layer (147):
The Deployment and Integration Layer is designed to seamlessly embed the Heart Disease Prediction System into existing clinical infrastructures, ensuring that it becomes a practical tool for healthcare professionals. This layer focuses on integrating the predictive system with Hospital Information Systems (HIS) and Electronic Medical Records (EMR), two critical components of modern healthcare management. By connecting to these systems, the Heart Disease Prediction System can access, update and interact with patient data in real-time, enabling healthcare providers to make informed decisions quickly and accurately.

The integration with HIS and EMR systems ensures that the predictive model operates smoothly ' within 'the ”clihica! \vork"llow." ’fliis^includesi "secure "data“transfer protocols to-comply~witlT healthcare data privacy regulations (e.g.. HIPAA, GDPR) and ensure that patient data remains confidential and protected. Additionally, the system is designed for real-time operation, allowing it to generate predictions and updates without delay, which is crucial in a clinical setting where timely information can impact patient outcomes. This layer’s robust architecture ensures that the system is scalable and adaptable, allowing it to be deployed across multiple departments or facilities without compromising performance or security.

•.Prediction and Risk Scoring Layer (148):
*
The Prediction and Risk Scoring Layer is the component responsible for interpreting patient data and generating a diagnostic output, specifically a risk score or classification that indicates the likelihood of heart disease. This layer uses the optimized predictive model developed in the Model Training and Optimization Layer to assess each patient’s risk factors and produce a clear, interpretable risk score. Patients are categorized into risk levels, such as low-risk, medium-risk and high-risk, based on their individual health profiles and the predictive model’s analysis.

The risk scoring provides a quantitative assessment of heart disease probability, which can be used by healthcare providers to prioritize high-risk patients and guide preventive or treatment measures. By producing a clear risk categorization, this layer supports quick and effective decision-making, allowing medical professionals to focus on cases that require immediate attention. This layer’s design ensures that the risk scores are generated consistently and accurately, leveraging the latest patient data from HIS and EMR systems to keep the predictions
up-to-date.
Output and Decision Support (149):
The Output and Decision Support layer serves as the final component in the Heart Disease
Prediction System, translating risk scores into actionable insights and diagnostic recommendations that support healthcare providers in making informed clinical decisions. This layer provides a user-friendly interface that presents the risk predictions and recommended actions, such as follow-up tests, lifesty le changes, or treatment options. By delivering specific, data-driven suggestions, this layer facilitates early interventions and personalized treatment plans tailored to each patient's unique health needs.

In addition to its diagnostic function, the Output and Decision Support layer plays a crucial role
in patient communication and care planning. The system's recommendations help guide
discussions between healthcare providers and patients, providing clear, evidence-based rationales for suggested interventions. This layer may also include options for ongoing monitoring and reassessment, allowing for dynamic adjustments to treatment plans as new patient data becomes
available. By supporting clinical decision-making with reliable, actionable insights, the Output
and Decision Support layer enhances patient care quality and enables more proactive, preventive
healthcare practices. In summary, these layers (147, 148 and 149) form the operational backbone
of the Heart Disease Prediction System in clinical settings. The Deployment and Integration
Layer ensures smooth system incorporation into healthcare infrastructures, the Prediction and
Risk Scoring Layer categorizes patient risk and the Output and Decision Support Layer
translates these insights into practical recommendations, enabling healthcare providers to deliver timely and personalized care.

CLAIMS:

We claim,
Claim 1: A system for predictive diagnosis of heart disease, comprising:
• a data collection layer configured to gather patient data, including electronic health records, wearable device data, MRI scan reports and public health datasets:
• a data preprocessing layer for cleaning, normalizing and balancing the data to create a standardized dataset:
• a feature extraction and selection layer for identifying critical features to enhance model performance;
• a model training and optimization layer for training machine learning models and selecting the best-performing model for heart disease prediction.

Claim 2: The system of claim 1, wherein the data preprocessing layer further comprises:
• a data cleaning module to handle missing values and reduce noise;
• a normalization module for consistent data scaling;
• a data balancing module that uses SMOTE or RUSBoost techniques to manage class imbalance.

Claim 3: The system of claim 1, wherein the feature extraction and selection layer employs
dimensionality reduction techniques, including principal component analysis (PCA), to focus on critical risk factors.

Claim 4: The system of claim 1, wherein the model training and optimization layer includes:
• a meta-classifier ensemble to improve diagnostic accuracy;
• a CNN-BiLSTM model with an attention mechanism for ECG data analysis;
• a U-Net model for image-based analysis of MRI scans;
• hyperparameler tuning via grid search or random search.

Claim 5: The system of claim I, wherein the deployment and integration layer connects with
hospital information systems (HIS) and electronic medical records (EMR), enabling real-time access to patient data in compliance with data privacy regulations.

Claim 6: The system of claim I, wherein the prediction and risk scoring layer generates a
heart disease risk score, categorizing patients into low-risk, medium-risk and high-risk
groups.

Claim 7: A method for predicting heart disease, comprising:
• collecting patient data from electronic health records, wearable devices, MRI scans and public health sources;
• preprocessing the data by performing cleaning, normalization and balancing;
• extracting and selecting key features for predictive modeling:
• training multiple machine learning models and selecting the highest-performing model;
• generating a risk score to classify patient heart disease risk levels.

Date: 22/11/2024
Mrs. Swetha S,
Dr S H Manjula