Enhanced Pediatric Pneumonia Diagnosis with Chest X-Ray using Deep Attention Mechanism

Muhammad Zaman, Tahseen Fatima, Sana Hameed, Shahzeb Haider, Adnan Akhunzada, Muhammad Azeem

Abstract: Development of deep learning technologies has resulted in a major revolution in the field of medical diagnostics in recent years. The purpose of this research is to examine the usage of Convolutional Neural Networks (CNNs) in combination with an integrated attention mechanism known as CoATNet to improve the accuracy and responsiveness of paediatric pneumonia diagnosis using chest X-ray analysis. The major goal of this research project is to build and verify a powerful deep learning model capable of detecting mild pneumonia symptoms in children’s chest X-rays. CoATNet’s innovative attention approach allows it to provide more weight to areas deemed to be extremely important. This is in addition to its ability to recognise major patterns in images. Furthermore, the paper considers the issues posed by class imbalances in medical imaging datasets and undertakes an in-depth assessment of how effectively CoATNet works across a large dataset of paediatric chest X-rays. CoATNet’s remarkable performance was shown by its ability to achieve a Micro Average of 98.2%, a Weighted Average of 98.7%, and an Accuracy of 99.7%. These measures show that the model can distinguish between instances of pneumonia and those do not have the illness, and that it is resilient when faced with class imbalances. F1 Score of 98.9% reflects a well-balanced trade-off between Precision 98.7% and Recall 98.2%. The integration of deep learning technologies with attention processes in CoATNet has resulted in a big step forward in the development of diagnostic models that are more visible and useful.

External IDs:dblp:conf/bdc/ZamanFHHAA24