Cardiomegaly Detection from Chest X-Rays Using EfficientNetB7: A Deep Learning Approach

Cardiomegaly, or an enlarged heart, is one of the main factor of different cardiovascular diseases and requires accurate and early diagnosis to improve patient outcomes. Traditional methods for detecting cardiomegaly from chest X-rays, such as visual assessment and manual cardiothoracic ratio (CTR) measurement, are generally time-consuming, subjective, and prone to inter-observer variability. These limitations hinder consistent and rapid diagnosis, especially in resource-limited settings. This study proposes a EfficientNetB7 architecture bases on deep learning approach for automated cardiomegaly detection using the NIH ChestX-ray14 dataset to address these challenges. A total of 5,552 pre-labelled X-ray images were used, categorized into normal and cardiomegaly cases. The transfer learning technique is used to fine-tuned the model, and extensive preprocessing and data augmentation methods were applied to enhance performance and generalizability. The proposed model outperformed the traditional techniques and several baseline deep learning models with an F1-score of 99.63% and accuracy of 99.60%. Furthermore, a confusion matrix and explainability metrics were employed to validate classification reliability. This work demonstrates the potential of advanced CNN-based architectures in providing faster, more reliable, and scalable diagnostic tools for clinical deployment, reducing the reliance on manual interpretation and paving the way for AI-assisted healthcare solutions.