Comparative Analysis of CNN and U-Net Models for Pneumonia Detection in Chest X-Ray Images

Pneumonia is a serious respiratory infection characterized by damage to lung tissue caused by several specialists. This may be a major welfare issue worldwide, especially for children. In conclusion, pneumonia remains difficult to accurately establish with restorative imaging. To perform classification in this case, two deep learning models are used: U-Net and Convolutional Neural Systems (CNN), both of which use the pre-trained model VGG16. The images handled are in a DICOM setup, which is commonly used in clinical settings to organize and share imaging-related medical image information. Due to the structured design of the VGG16 model, this model is effective in understanding important features in an image such as angles, textures, and internal shapes for recognition tasks. Our strategy includes pre-processing DICOM images, expanding information, and improving data collection for preparation and training. We assess spectacle execution using measurements such as Intersection over Union (IoU), accuracy, recall, F1-Score, and precision. These results show that both models achieve a precision of around 70%, which underlines the need for assistance changes. Improving data set quality, hyperparameter tuning, advancing pre-processing strategies, and gathering defined strategies to make progress shows execution. These considerations highlight the challenges and potential settings in leveraging deep learning to infer pneumonia, with the aim of contributing to more successful and readily available demonstrative tools.