A Novel Ontology and Machine Learning Driven Hybrid Clinical Decision Support Framework for Cardiovascular Preventative Care

Contents

1 INTRODUCTION

1.1 Organisation of Thesis

1.2 Motivation and aims

1.3 Original Contributions

1.4 Publications

2 LITERATURE REVIEW

2.1 Clinical Decision Support Systems

2.1.1 Ontology Driven Clinical Decision Support Frameworks

2.1.2 Clinical Decision Support Systems in Cardiovascular Care

2.1.3 Cardiovascular Risk Estimation Systems for Disease Prevention

2.1.4 Machine Learning Driven Cardiovascular Decision Support Systems

2.1.5 Role of Feature Selection in Clinical Decision Support Systems

2.2 Conclusion and Discussion

3 A Novel Ontology and Machine Learning Driven Hybrid Clinical Decision Support Framework for Cardiovascular Preventative Care

3.1 Proposed Framework

3.2 ODCRARS for Cardiovascular Preventative Care

3.2.1 Ontology driven intelligent context aware information collection component

3.2.2 Patient Medical Records

3.2.3 Ontology Driven Decision Support

3.3 Machine Learning Driven Prognostic Modelling for Cardiovascular Preventative Care

3.4 Machine Learning Driven Prognostic Model

3.4.1 Data Acquisition

3.4.2 Data Pre-Processing

3.4.3 Feature Selection

3.4.4 Prognostic Model Development

3.4.5 Prognostic Model Validation and Evaluation

3.4.6 Online Clinical Prognostic Model

3.5 Conclusion and Discussion

4 Ontology Driven Clinical Risk Assessment and Recommendation System (ODCRARS) for Cardiovascular Preventative Care

4.1 Implementation of the Ontology Driven Clinical Risk Assessment and Recommendation System (ODCRARS)

4.2 Ontology driven intelligent context aware information collection: Design and Implementation

4.2.1 Ontology Driven Intelligent Context Aware Ontology Model

4.2.2 Adaptive Clinical Questionnaire: Design and Implementation

4.2.3 Proposed Novel Decision Tree based Approach

4.2.4 Dynamic Adaptation

4.3 Patient Medical Records

4.4 Patient Semantic Profile : Design and Implementation

4.4.1 Ontology Development

4.5 Ontology Driven Clinical Decision Support: Design and Implementation

4.5.1 Recommendation Ontology

4.6 Clinical Rules Engine: Design and Implementation

4.6.1 Clinical Rules Data - Patient Fact Representation

4.6.2 Jess: Java based Rules Engine

4.6.3 Partitioning the Rules

4.6.4 Cardiovascular Risk Assessment

4.7 System Implementation: Integration of ODCRARS and MLDPS

4.7.1 Patient Module

4.8 Doctor’s Module

4.8.1 Integration of the ODCRARS with the machine learning driven cardiac chest pain and heart disease prognostic models

4.9 Conclusion and Discussion

5 Machine Learning Driven Prognostic System (MLDPS) for Cardiovascular Preventative Care

5.1 Case Study 1: Rapid Access Chest Pain Clinic

5.1.1 Background

5.1.2 Aims

5.2 RACPC Clinical Dataset 1

5.2.1 Data Acquisition

5.2.2 Data Preparation

5.2.3 Missing Data Handling

5.2.4 Feature Selection

5.2.5 Prognostic Model Development: Experimental Setups and Results

5.2.6 Final Diagnosis

5.2.7 Evaluation of RACPC Results

5.2.8 Results of Comparative Machine Learning Classification

5.2.9 Analysis of Variance (ANOVA) Test for Performance Evaluation

5.3 RACPC Clinical Dataset 2: Demonstrating Effects of missing Data on Verification Results

5.3.1 Background

5.3.2 Pre-processing of Missing Data using Probability Estimation

5.3.3 Expectation Maximisation (EM) Approach

5.3.4 Experiments

5.3.5 Classification for the Incomplete Clinical Data

5.3.6 Filling the Incomplete Data

5.4 RACPC Clinical Case Study: RACPC Clinical Dataset 3

5.4.1 Study Group 1: Clinical Risk Factors

5.4.2 Evaluation

5.4.3 Performance evaluation of experimental setups

5.4.4 Study Group 2: Test Results

5.4.5 Evaluation

5.4.6 Performance evaluation of experimental setups

5.4.7 Implementation of online Clinical Prognostic Models

5.4.8 Machine Learning Driven Cardiac chest pain prognostic model’s integration with the recommendation system

5.5 Case Study 2: Heart Disease

5.5.1 Background

5.5.2 Aims

5.5.3 Data Preparation

5.5.4 Feature Selection

5.5.5 Prognostic Model Development

5.5.6 Prognostic Model Validation and Evaluation

5.5.7 Performance evaluation of experimental setups

5.5.8 Implementation of online Clinical Prognostic Models

5.6 Case Study 3: Breast Cancer Prognostic Modelling

5.6.1 Background

5.6.2 Aims

5.6.3 Candidate Clinical Variable Selection

5.6.4 Prognostic Model Development

5.6.5 Prognostic Model Validation and Evaluation

5.6.6 Performance Evaluation of Experimental Setups

5.6.7 Online Clinical Prognostic Model

5.7 Verification and Validation of the Clinical Prototypes

5.7.1 Validation of the Machine Learning Driven System (MLDPS) and Ontology Driven Clinical Risk Assessment and Recommendation System (ODCRARS)

5.8 Summary and Conclusion

6 CONCLUSIONS AND FUTURE WORK

6.1 Conclusions

6.2 Discussion and Summary of Contributions

6.3 Future Work

6.3.1 Utilisation of Fuzzy Cognitive Maps for Collaborative Care

6.3.2 Active Manifold Learning Strategy in Machine Learning Driven Prognsotic Modelling based on Big Data

6.4 Limitations

Objectives and Topics

This thesis aims to develop a novel hybrid clinical decision support framework for cardiovascular preventative care by integrating ontology-driven risk assessment with machine learning-based prognostic modeling. The primary research question addresses how to effectively leverage legacy clinical patient data and standardized clinical guidelines to improve patient-centric, evidence-based care in primary and secondary healthcare settings.

• Development of an Ontology-Driven Clinical Risk Assessment and Recommendation System (ODCRARS) for patient interviews and clinical rules implementation.
• Creation of a Machine Learning Driven Prognostic System (MLDPS) for predicting cardiovascular risks and other chronic conditions like breast cancer.
• Implementation of context-aware, adaptive information collection techniques using decision tree-based questionnaires.
• Evaluation of various machine learning and feature selection techniques for prognostic model performance in clinical datasets.

Book Excerpt

Summary of Chapters

1 INTRODUCTION: Outlines the importance of clinical data for health learning and introduces the hybrid framework proposed to address limitations in conventional systems.

2 LITERATURE REVIEW: Reviews existing clinical decision support systems, ontology-driven frameworks, and machine learning techniques applied in cardiovascular care.

3 A Novel Ontology and Machine Learning Driven Hybrid Clinical Decision Support Framework for Cardiovascular Preventative Care: Presents the overall architecture of the integrated system and its two key components: ODCRARS and MLDPS.

4 Ontology Driven Clinical Risk Assessment and Recommendation System (ODCRARS) for Cardiovascular Preventative Care: Details the design and implementation of context-aware information collection, semantic profiling, and the clinical rules engine.

5 Machine Learning Driven Prognostic System (MLDPS) for Cardiovascular Preventative Care: Describes the methodology for developing prognostic models, including data acquisition, preprocessing, feature selection, and validation using clinical datasets.

6 CONCLUSIONS AND FUTURE WORK: Summarizes the contributions of the research and suggests future research directions, such as incorporating Fuzzy Cognitive Maps and active learning strategies.

Keywords

Clinical Decision Support Systems, CDSS, Ontology, Machine Learning, Cardiovascular Care, Prognostic Modeling, Risk Assessment, Semantic Profile, Feature Selection, Evidence-Based Medicine, Big Data, Healthcare Analytics, Diagnostic Support, Patient-Centric Care

Frequently Asked Questions

What is the core purpose of this research?

The research aims to create a hybrid clinical decision support framework that combines expert-driven ontological knowledge with data-driven machine learning to improve cardiovascular preventative care.

What are the two primary components of the proposed system?

The two main components are the Ontology-Driven Clinical Risk Assessment and Recommendation System (ODCRARS) and the Machine Learning Driven Prognostic System (MLDPS).

How is patient information handled in the system?

Information is collected through an ontology-driven, context-aware adaptive questionnaire, which is then transformed into a Patient Semantic Profile to ensure data interoperability and intrinsic meaning.

Which scientific methods are utilized for prognostic modeling?

The system uses various machine learning classifiers, including Logistic Regression, Decision Trees, and Support Vector Machines, alongside feature selection techniques like SFFS and mRMR.

What is the role of the Clinical Rules Engine?

The rules engine acts as a core mechanism to provide cardiac risk assessment scores and control the patient flow within the integrated healthcare solution based on clinical guidelines.

What are the key keywords associated with this thesis?

Key terms include Clinical Decision Support Systems (CDSS), Ontology, Machine Learning, Cardiovascular Care, Prognostic Modeling, and Risk Assessment.

How were the prognostic models validated?

The models were validated using real clinical datasets, including the Rapid Access Chest Pain Clinic (RACPC) data from Raigmore Hospital, as well as heart disease and breast cancer datasets, with input from clinical domain experts.

What is the significance of the "adaptive questionnaire" approach?

The adaptive questionnaire uses a decision tree-based logic to mimic clinical investigative behavior, ensuring that only relevant questions are asked to the patient, thereby improving usability and data quality.