Binary Classification of Polyps (Cancer Precursors) through CNNs

This project involves training a Convolutional Neural Network (CNN) for binary classification of polyp images. Polyps are precursors for cancer found durign colonoscopies. The model is trained to classify images into two categories: Positive and Negative.

• Model can be used through the polypdetector.py script
• Script to train model can be found under polyp_ai.py
• Trained model can be found under Polyp_AI_Model_V0-2.pth
• Performace / Results can be found under RESULTS.md
• Citations can be found under CITATIONS.md

Disclaimer

• To train the model, there is no dataset on Github. Dataset is extracted from Synapse once you run the script locally
• Use your own API token on line 40 to extract the data from synapse client
• Lines 13 - 64 and 470 - end of file contain all code used to train the model
• The commented section between 64 - 470 of the script was my attempt at implementing an Object detection model with the use of CNNs that outputs masks.

Project Structure

• Model: BinaryClassificationCNN
• Hyperparameters:
  • Learning rate: 0.0005
  • Number of workers: 2
  • Input channels: 3
  • Hidden units: 32
  • Number of batches: 16
  • Number of epochs: 3
  • Image size: 128
  • Device: cuda if available, else cpu

Setup

Prerequisites

• Python 3.x
• PyTorch
• torchvision
• matplotlib
• numpy
• tqdm
• Pillow
• synapseclient

Installation

Install the required packages using pip/pip3:

bash pip3 install torch torchvision matplotlib numpy tqdm pillow synapseclient

Training and Evaluation Loop

The model uses binary classification with logits loss as the loss function. The optimizer used was Adam.

This code was device agnostic, meaning both your computer's CPU or a Google Colab GPU can be used to run the script.

```python

Initialize the model, loss function, and optimizer

model = BinaryClassificationCNN(inputchannels, hiddenunits).to(device) loss_fn = nn.BCEWithLogitsLoss() optimizer = optim.Adam(model.parameters(), lr=lr)

Train the model

trainandevaluate(model, trainloader, testloader, lossfn, optimizer, accuracyfn, device, epochs) ```

Visualizing Predictions

The project includes functionality to visualize random predictions from the test set. The visualization displays 9 images in a 3x3 grid with true labels, predicted labels, and prediction probabilities.

Visualization

```python

Visualize predictions

visualizepredictions(model, testloader, device, num_images=9) ```

Confusion Matrix

To evaluate the model's performance, a confusion matrix can be generated to display True Positives (TP), True Negatives (TN), False Positives (FP), and False Negatives (FN).

Confusion Matrix

```python

Generate confusion matrix

confmatrix = generateconfusionmatrix(model, testloader, device) ```

Conclusion

This project demonstrates the use of a CNN for binary classification of polyp images. The model's performance is evaluated using accuracy, loss, and a confusion matrix to analyze the classification results.

Feel free to modify the hyperparameters and experiment with different configurations to improve the model's performance.