This project aims to detect pneumonia in chest X-ray images using convolutional neural networks (CNNs). We explore the effectiveness of transfer learning from a pre-trained ResNet model and compare it with a simple CNN architecture.
This project employs two models for pneumonia detection in chest X-ray images:
models.py: Implementation for our own simple CNN and the Lightning Module that we use for both models.simpleCNN.ipynb: First Implementation of our own simple CNN model.simpleCNN_tuning.ipynb: Implementation of hyperparameter tuning, using a complexer CNN model that is defined in the models.py.Project-1-ResNet50.ipynb: Implementation of a model based on ResNet with transfer learning.
The models are evaluated using accuracy, F1-score, sensitivity, and specificity metrics:
- Accuracy: Provides an overall estimate of model correctness, but may be influenced by dataset imbalance. We expect an accuracy above 80%.
- F1-score: A trade-off between precision and sensitivity/recall.
- Sensitivity: Indicates the proportion of positive cases correctly diagnosed.
- Specificity: Indicates the proportion of negative cases correctly diagnosed.
Through this project, we aim to develop an accurate and reliable model for pneumonia detection in chest X-ray images.