A multi-scale CNN with atrous spatial pyramid pooling for enhanced chest-based disease detection.

Abstract

We introduce a sophisticated deep-learning model designed for the early detection of COVID-19 and pneumonia. The model employs a convolutional neural network-integrated with atrous spatial pyramid pooling. The atrous spatial pyramid pooling mechanism enhances the convolutional neural network model's ability to capture fine and large-scale features, optimizing detection accuracy in chest X-ray images. This improvement, along with transfer learning, significantly enhances the overall performance. By utilizing data augmentation to address the scarcity of available X-ray images, our atrous spatial pyramid pooling-enhanced convolutional neural network achieved a validation accuracy of 98.66% for COVID-19 and 83.75% for pneumonia, which beats the validation results of the other state of the art approaches (the metrics used for evaluation were accuracy, precision, F1-score, recall, specificity, and area under the curve). The model's multi-branch architecture facilitates more accurate and adaptable disease prediction, thereby increasing diagnostic precision and robustness. This approach offers the potential for faster and more reliable diagnoses of chest-related conditions.