Publications

Abstract : Pediatric Pneumonia is an acute respiratory infection caused by bacteria, viruses, or fungi and reports nearly 800,000 deaths worldwide. The prime reason for this high toll is the delayed diagnosis and overpriced tests. Chest X-rays, being non-invasive, painless, and cost-efficient, are preferred as the standardized test for pediatric pneumonia diagnosis. Experts study these X-rays to detect pneumonia. However, the use of lower levels of radiation for pediatric screening makes it a challenging task. A computer-aided diagnosis model is thus required for accurate prediction. Our work proposes a novel architecture using separable convolutions, residual connection, and residual attention learning with cost-sensitive learning for robust classification of the commonly available pediatric pneumonia dataset. The model was tested on the publicly available pediatric pneumonia dataset, achieving an accuracy of 95.83%, precision of 96.19%, recall of 97.18%, F1 score of 96.68%, and an AUC score of 95.38%. The proposed architecture results in competing performance with recent works. The performance shows its reliability for real-time deployment in early pediatric pneumonia diagnosis.