Background: This study presents a deep learning project conducted for Pneumonia Detection using the PyTorch framework. The project focuses on leveraging the ResNet18 architecture and the RSNA Pneumonia Detection challenge dataset to achieve accurate disease detection. The primary objective of this research is to investigate the effectiveness of the applied deep learning techniques in detecting pneumonia cases. Method: The ResNet18 architecture was chosen due to its proven performance in various computer vision tasks. The model was trained and evaluated on the RSNA Pneumonia Detection dataset, which is widely recognized for its complexity and diversity. The dataset consists of a large number of annotated chest X-ray images, providing valuable training samples for deep learning models. To assess the model's performance and gain insights into its inner workings, artificial intelligence interpretability techniques were employed. By applying these techniques, the researchers were able to explore how the model made decisions and identify the features it considered important for pneumonia detection. This approach enhanced the interpretability and transparency of the deep learning model, addressing the black-box issue often associated with deep learning algorithms. Results: The results of this project demonstrated promising performance in accurately detecting pneumonia cases. The deep learning model achieved notable accuracy rates, showcasing its potential as a reliable tool for pneumonia detection. The interpretability techniques further provided valuable insights into the decision-making process of the model, assisting medical professionals in understanding and trusting the model's predictions. The full implementation (based on PyTorch) and the trained networks are available at github.com/Pooryamn/Pneumonia-Detection.
- Pneumonia Detection
- Medical Image Analysis
- Deep Convolutional Neural Networks (DCNN)
- Explainable artificial intelligence (XAI)