Label Efficient Active Learning for Coronary Artery Disease Detection and False Negatives Elimination
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Coronary artery disease (CAD) remains a leading cause of global mortality. Stacked‐ensemble machine learning (ML) models trained on structured clinical, laboratory, and echocardiographic measurements can achieve baseline accuracies of 93.26 % on the Cleveland dataset, 90.12 % on Statlog, 90.15 % on Kaggle Framingham, 95.56 % on Z Alizadeh Sani, and 89.06 % on the in house JIPMER cohort. Pre model enhancements like feature evaluation (significant‐feature selection, differential analysis, normalization) and error‐reduction methods (cost‐sensitive learning, manual weight adjustment, precision–recall thresholding, engineered features) raise performance to at most 92.19 % on JIPMER dataset, yet false negatives persist. Active learning (AL) addresses both label scarcity and misclassification risk by iteratively querying the most informative unlabeled cases for expert annotation. We present an AL framework built on a five-model stacked ensemble (Linear SVM, RBF SVM, KNN, Random Forest, AdaBoost), applied to in house JIPMER and four different public datasets. Our model boosted accuracy to 97.66 % on the JIPMER dataset and eliminated all false negatives using an AL-enhanced stacked ensemble. These results demonstrate that AL can substantially boost diagnostic reliability and cost‐effectiveness of ML‐based CAD prediction tools, supporting rapid clinical deployment in resource constrained settings.
Fields of Study & Subjects
Artificial Intelligence in Healthcare
Machine Learning and Algorithms
COVID-19 diagnosis using AI
Publication details
- Journal / source
- CRC Press eBooks
- Pages
- 405–419
- Publication date
- 26 March 2026
- Language
- EN
- Document type
- Book chapter