Intelligent Clinical Information Platform for Assisting Heart Disease Care Pathway using Machine Learning

An average of 3 million deaths occurs each year in high-income countries due to unsafe care, with causes including diagnostic and communication failures. These failures are related to clinical information overload, the extraction of essential unstructured data, and complex health data analytics for deriving insights. The use case of this dissertation focuses on emergency room (ER) physicians, as they are the initial point of contact for patients, and time-sensitive situations occur frequently in the ER. The goal is to develop an intelligent clinical information platform (ICIP) for ER physicians, assisting patients’ care pathways using machine learning (ML). This platform provides a new, multidimensional view to represent patients’ medical conditions, focused on heart diseases. To achieve the platform’s implementation, three technical components are developed and published within this dissertation: first, a component for data extraction from remote video consultations via WebRTC; second, a data classification component using a Faster Region-Based Convolutional Neural Network (R-CNN) model together with active learning (AL); and third, a data search component with an implemented Elasticsearch pipeline and data storage unified in the FHIR standard. The research for a newly developed clinical platform is practically and industrially based on building a future clinical product. For this product, ML models are developed to analyze data from past clinical treatments using an R-CNN model for text classification and to access verbal audio data through a speech-to-text (STT) engine employing an RNN TensorFlow model and a large language model (LLM) from NLP.js. Additionally, JSON object-based rule-based reasoning from FHIR is used. It has been demonstrated that a three-tier architecture (AngularJS, Java Spring Boot, and PostgreSQL), consisting of components involving neural networks such as R-CNN, RNN (recurrent neural network), and LLM, can be implemented as a data platform for assisting heart disease care pathways. This allows physicians to interpret patients’ vital parameters, pathways, and timelines via diagrams presented in widgets on the AngularJS frontend.