COMP 432 - Final Project Group K

Image Classification using Convolutional Neural Networks (CNNs) and Transfer Learning

This project explores using Convolutional Neural Networks (CNNs) to address image classification tasks from real-world applications in computational pathology and computer vision. The project is divided into two main tasks:

1. Training and Analyzing CNN Models on Pathology Data:

   • Train CNNs on a Colorectal Cancer Tissue Dataset.
   • Analyze and visualize feature representations using t-SNE.
   • Quantitative analysis of the model's performance

2. Knowledge Transfer and Cross-Dataset Analysis:

   • Transfer learned feature representations to other datasets (Prostate Cancer and Animal Faces).
   • Compare feature extraction against a pre-trained CNN encoder trained on ImageNet.
   • Use a classical machine learning algorithm (Random Forest) to classify the extracted features on the Prostate Cancer and Animal Faces datasets.

The main objectives are:

• To study how CNNs generalize across datasets.
• To conduct a detailed analysis of model performance using t-SNE visualizations.
• To classify extracted features using classical machine learning methods.

Requirements

To set up the environment, use:

pip install -r requirements.txt

Dataset

The datasets used in this project are:

1. Colorectal Cancer Tissue Dataset (used for training CNNs).
2. Prostate Cancer Dataset (used for cross-dataset analysis).
3. Animal Faces Dataset (used for cross-dataset analysis).

They are already Downloaded if you clone the Github Repository, If you want to download them:

Download Links

The datasets can be downloaded from the following sources:

• Colorectal Cancer Tissue Dataset: Download Link
• Prostate Cancer Dataset: Download Link
• Animal Faces Dataset: Download Link

Directory Structure

Once downloaded, organize the datasets into the following directory structure:

• Data/Colorectal Cancer
• Data/Prostate Cancer
• Data/Animal Faces
• Data/Sample Test Dataset (Prostate Cancer)/Prostate Cancer

Instructions

1. Training and Validating Your Model

To train and validate your CNN model on the Colorectal Cancer Dataset, follow these steps:

1. Open the task_1.ipynb notebook in Jupyter Notebook or JupyterLab.
2. Ensure the Colorectal Cancer Dataset (the sample DataSet with just 100 samples) is placed in the correct directory: Data/Sample Test Dataset (Prostate Cancer)/Prostate Cancer/.
3. Execute all the cells in the notebook to:
   • Preprocess the data.
   • Train the CNN model.
   • Validate the model performance on the validation set.
4. Once training is complete, the trained model will be automatically saved as resnet50_colorectal.pth in the task_1/ directory.
5. t-SNE visualizations of the model's feature representations will be saved in the Figures/ directory.

2. Running the Pre-Trained Model

To evaluate the pre-trained model on the provided test dataset, follow these steps:

1. Extract the test dataset from the project ZIP file.
2. Open the task_1.ipynb notebook in Jupyter Notebook or JupyterLab.
3. Locate the section titled "Evaluate Pre-Trained Model" in the notebook.
4. Ensure the pre-trained model file, resnet50_colorectal.pth, is present in the task_1/ directory.
5. Modify the test_dir variable in the notebook to point to the test dataset directory (Data/Colorectal Cancer/Test/).
6. Run the evaluation cells to compute:
   • Performance metrics (e.g., accuracy, precision, recall, F1-score).
   • Visualizations of feature representations on the test dataset.

The evaluation results, including metrics and visualizations, are displayed in the notebook and saved to the Classification_Reports/ and Figures/ directories.