Optimized Pneumonia Detection via CT Scans: A Comparative Analysis of Transfer Learning Models


Pneumonia continues to be one of the main causes of death among children below five and the elderly population above the age of 65 years. According to the minister of state in the Ministry of Health and Family Welfare, Dr. Bharati Pravin Pawar, at least 687 children aged 1-12 months and 301 chil-dren aged 1-5 years lost their lives due to pneumonia as part of the total number of deaths the disease caused in 2022-23. The high death rate is largely prevalent in South Asia and Sub-Saharan Africa. Pneumonia also remains among the top causes of deaths even in the most prosperous countries, such as the United States, falling within the ten leading causes. Early diagnosis does a lot to help reduce fatalities. This paper addresses this problem by showing research work that is based on the application of CNN models for detecting pneumonia from chest X-ray images.A number of CNN architectures, including VGG16, ResNet50, and DenseNet121, were trained and fine-tuned with varying parameters, hyperparameters, and counts of the convolutional layers. Transfer learning has drastically increased model accuracy while reducing the time taken to train. Results In relation to the efficient use of deep learning in medical image processing, the study underscores the effectiveness of transfer learning in CNNs with minimal label data, particularly in conditions. The algorithms were able to accurately classify the X-ray images into the classes of pneumonia and non-pneumonia. This approach further elaborates on the fact that CNNs, when utilized together with transfer learning, may be suitably applied for the early and timely detection of pneumonia, eventually minimizing infant mortality rates all over the world.
PDF