An Empirical Comparison of Deep Learning Models for ARDS Detection Using a Meta Learning Model

ARDS is a severe pulmonary condition that requires timely and accurate diagnosis in order to reduce mortality. In this work, several deep learning architectures, including ResNet-101, ResNet-152, EfficientNet-B6, and EfficientNet-B7, were empirically compared for the detection of ARDS, together with a novel meta-learning–based fusion framework. The proposed stacking-based meta learner integrates the complementary predictions from individual models to enhance diagnostic performance. Experiments conducted on a multi-class CT image dataset show that the proposed model yields an accuracy of 96%, recall of 96%, and F1-score of 92% significantly higher than standalone architectures. Comprehensive evaluation using accuracy, sensitivity, specificity, AUC, and confusion matrix analysis confirms the robustness, stability, and superior generalization capability of the proposed approach. The results have underlined the effectiveness of meta-learning–driven model fusion in the development of a reliable ARDS detector, capable of clinical decision support.