“Pneumonia Detection With Transfer Learning: A Deep Cnn- Based Comparison Of Detection And Segmentation

Abstract

In pneumonia, the lung tissue undergoes swelling, a condition caused by viral or bacterial infections. This swelling, accompanied by increased lung moisture, results in difficulty in deep breathing. Recognizing these symptoms is crucial, as pneumonia can lead to a substantial number of fatalities. A fundamental diagnostic method for pneumonia involves chest X-rays. This research publication presents a comprehensive overview of recent advancements in pneumonia diagnosis, introducing the authors' unique approach. The integrated models underwent rigorous testing for image classification, demonstrating exceptional performance by models such as VGG19, MobileNetV2, ResNet152V2, SeresNet152, and ResNext101. Utilizing X-ray images from both patients and healthy subjects, the study incorporated visual enhancements and augmented data through compressed archive downloads. Various models with different precisions were employed, revealing that ResNeXt101 and SeresNet152 exhibited the lowest accuracy among the models at 90%. Conversely, VGG19, MobileNetV2, and ResNet152V2 secured the top positions with a commendable accuracy of 92%. All models demonstrated satisfactory outcomes. A dedicated session on transfer learning followed, indicating that the accuracy of results from transfer learning models was slightly lower than that of the original models. Transfer learning was employed to enhance the precision of models, resulting in accuracies of 91.98% for ResNet152V2, 88.11% for SeresNet152, 91.47% for MobileNetV2, 88.19% for VGG19, and 91.60% for ResNeXt101. Despite this, the accuracy remained notably lower compared to the original models. This research holds significance for data scientists as it provides essential insights into pneumonia diagnosis, emphasizing the importance of continuous advancements in medical research.