Machine Learning-Based Classification of ECG Biomarkers for Brugada Syndrome Diagnosis

Aug 5, 2026

Upload by Oliver

Brugada Syndrome is a genetic disorder marked by abnormal electrocardiogram patterns and an elevated risk of sudden cardiac death, with prognosis that remains difficult to predict. The thesis develops machine-learning classifiers to categorize biomarkers extracted from BrS patients’ ECGs, after biomarker preprocessing and labeling. Models compared include random forest, decision tree, support vector machine, and XGBoost. To reduce overfitting and handle class imbalance, principal component analysis and linear discriminant analysis are applied, alongside clinical-feature removal; the best results come from LDA with clinical data using random forest. Further refinement requires hyperparameter tuning, revised criteria, and a larger patient database.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages135

Reading Time5 h 40 min

AccessFree to download

Abstract

Methods and Models

Classifiers

Dimensionality Reduction and Feature Handling

Results and Next Steps

FINAL DEGREE THESIS Bachelor’s degree in biomedical engineering MACHINE LEARNING-BASED CLASSIFICATION OF ECG BIOMARKERS FOR BRUGADA SYNDROME DIAGNOSIS Report and Annex Author: Souhaila Dari Berraha Supervisor: Jordi Solà Soler Department ESAII Co-supervisor: Flavio Palmieri (Universitat de Barcelona) Elena Arbelo Lainez (Hospital Clínic i Universitat de Barcelona) Call: 2024, June Abstract Brugada Syndrome is a recently discovered genetic disorder characterized by abnormal electrocardiogram patterns and an increased risk of sudden cardiac death. The unpredictable prognosis of BrS requires further study, which is the focus of the AI4BrS project—a collaboration between CREB and IDIBAPS. This thesis focuses on classifying biomarkers extracted from BrS patients'ECGs using machine learning techniques. These biomarkers were pre-processed, labelled, and used to train classifiers such as random forest, decision tree, support vector machine, and XGBoost. To address challenges such as overfitting and class imbalance, dimensionality reduction techniques like Principal Component Analysis and Linear Discriminant Analysis were employed, along with the removal of clinical features. The results indicated that applying LDA on biomarkers combined with clinical data using the RF classifier yielded the best performance. However, further work is necessary, including adjusting hyperparameters, revising classification criteria, and expanding the patient database to improve the model's accuracy and generalization. This study underscores the potential of machine learning in enhancing the diagnosis and risk stratification of Brugada Syndrome, providing valuable tools for clinical decision-making. The interdisciplinary approach, combining medicine with advanced computational methods, highlights the significance of this research in advancing medical diagnostics and improving patient outcomes. Resum El Síndrome de Brugada és un trastorn genètic descobert recentment, caracteritzat per patrons anormals en l'electrocardiograma i un augment del risc de mort sobtada cardíaca. El pronòstic imprevisible del BrS requereix un estudi més profund, que és l'enfocament del projecte AI4BrS, una col·laboració entre CREB i IDIBAPS. Aquesta tesi se centra en la classificació de biomarcadors extretsdels ECG de pacients amb BrS utilitzant tècniques de machine learning. Aquests biomarcadors van ser preprocessats, etiquetats i utilitzats per entrenar classificadors com boscos aleatoris, arbres de decisió, màquines de vectors de suport i XGBoost. Per abordar desafiaments com el sobreajustament i el desequilibri de classes, es van emprar tècniques de reducció de dimensionalitat com l'Anàlisi de components principals i l'Anàlisi discriminant lineal, juntament amb l'eliminació de característiques clíniques. Els resultats van indicar que aplicar l'Anàlisi discriminant lineal als biomarcadors, combinats amb dades clíniques utilitzant el classificador deboscos aleatoris, va produir el millor rendiment. No obstant això, és necessari realitzar més treball, incloent l'ajust d'hiperparàmetres, la revisió de criteris de classificació i l'expansió de la base de dades de pacients per millorar la precisió i la generalització del model. Aquest estudi subratlla el potencial del machine learning per millorar el diagnòstic i l'estratificació deriscos de la Síndrome de Brugada, proporcionant eines per a la presa de decisions clíniques. L'enfocament interdisciplinari, que combina medicina amb mètodes computacionals avançats, ressaltala importància d'aquesta investigació en l'avanç dels diagnòstics mèdics i la millora dels resultats per als pacients. Resumen El Síndrome de Brugada es un trastorno genético descubierto recientemente, caracterizado por patrones anormales en el electrocardiograma y un aumento del riesgo de muerte súbita cardíaca. El pronóstico impredecible del BrS requiere un estudio más profundo, que es el enfoque del proyecto AI4BrS, una colaborac

FAQs

What problem does the thesis address in Brugada Syndrome diagnosis?

It targets the unpredictable prognosis of Brugada Syndrome by using machine learning to classify biomarkers extracted from patients’ ECGs to improve diagnostic and risk-stratification support.

Which machine learning classifiers are evaluated?

The study trains and compares random forest, decision tree, support vector machine, and XGBoost classifiers.

How are overfitting and class imbalance handled?

It applies dimensionality reduction methods such as principal component analysis and linear discriminant analysis, and also removes clinical features to address learning challenges.

Aug 5, 2026

Upload by Oliver