Enhancing deep learning for pneumonia detection: developing web based solution for Dr. Sumait Hospital in Mogadishu Somalia

Abstract
In the realm of medical imaging, accurate and efcient detection of pneumonia from chest X-ray images is crucial
for timely diagnosis and treatment. This study explores the performance of four deep learning models—Simple
CNN, DenseNet121, VGG16, and InceptionV3—using the Kaggle "Chest X-Ray Images (Pneumonia)" dataset, which
comprises 5,863 images categorized into normal and pneumonia classes. The methodology included data normalization,
augmentation, and training, followed by evaluation based on accuracy, precision, recall, and F1-score. The results
revealed that Simple CNN achieved the highest accuracy at 92%, with notable precision (0.95 for normal and 0.90 for
pneumonia) and recall (0.83 for normal and 0.97 for pneumonia). VGG16 also performed well with an accuracy of 91%,
while DenseNet121 and InceptionV3 had lower performance, with InceptionV3 exhibiting the lowest accuracy (84%)
and higher false positive rates. Based on these fndings, Simple CNN was chosen for deployment in a Django-based
web application hosted on AWS, aimed at improving diagnostic accuracy and supporting healthcare professionals at
Dr. Sumait Hospital. The study underscores the efcacy of Simple CNN for clinical applications and suggests future
enhancements such as dataset diversifcation, multi-class classifcation, real-time processing, and the incorporation of
additional clinical data.