Automated Gleason Scoring System

This project is an implementation of an automated Gleason scoring system using convolutional neural network (CNN) models, specifically ResNet-50 and EfficientNet. The goal is to classify prostate tissue into Gleason scores, helping to improve the diagnosis of prostate cancer.

Overview

Models Used

• ResNet-50: This model achieved an impressive 97% accuracy during testing, proving to be highly reliable in predicting Gleason grades.
• EfficientNet: This model achieved a 62% accuracy, performing moderately well in comparison.

Visualizations

Here are some visualizations to better understand the results:

Original and Augmented Images

Training Metrics for EfficientNet

Training Metrics for ResNet-50

Grad-CAM Visualizations

• Benign Tissue:
• Gleason 3:
• Gleason 4:
• Gleason 5:

Classification Metrics for ResNet-50

Misclassified Example

Confusion Matrix

Dataset

The dataset contains 200 tissue microarray (TMA) images, divided into training, validation, and testing sets in a 70:15:15 ratio. The images are categorized into the following classes:

• Benign
• Gleason3
• Gleason4
• Gleason5

Methodology

1. Data Preparation

   • The dataset was divided into training, validation, and test sets.
   • Images were resized, normalized, and augmented to improve model performance.

2. Model Training

   • Models were trained using the PyTorch framework.
   • Techniques like dropout and learning rate scheduling were used to prevent overfitting and optimize performance.

3. Visualization

   • Grad-CAM was used to visualize the areas of tissue influencing the model’s predictions.

Models

The repository includes the following saved PyTorch models:

1. EfficientNet:

   • Location: models/efficientnet_best.pth
   • Description: Trained EfficientNet model with a test accuracy of 62%.

2. ResNet-50:

   • Location: models/resnet_best.pth
   • Description: Trained ResNet-50 model with a test accuracy of 97%.

You can load these models for inference or further fine-tuning using the code below:

import torch
from efficientnet_pytorch import EfficientNet
from torchvision.models import resnet50

# Load ResNet-50
resnet_model = resnet50(pretrained=False)
resnet_model.load_state_dict(torch.load("models/resnet_best.pth"))
resnet_model.eval()

# Load EfficientNet
efficientnet_model = EfficientNet.from_name("efficientnet-b0")
efficientnet_model.load_state_dict(torch.load("models/efficientnet_best.pth"))
efficientnet_model.eval()

Results

• ResNet-50 achieved a test accuracy of 97%, making it a reliable choice for classification.
• EfficientNet achieved a test accuracy of 62%, with potential for improvement.
• Grad-CAM visualizations effectively highlighted tumor regions in tissue samples.

Future Work

• Expand the dataset to improve model generalization.
• Experiment with other CNN architectures like DenseNet.
• Develop ensemble models to boost overall performance.

How to Use

1. Clone this repository:

   git clone https://github.com/yourusername/gleason-score-classifier.git

2. Install the required dependencies:

   pip install -r requirements.txt

3. Open and run the Jupyter Notebook for training and testing:

   jupyter notebook src/Gleason_Score_Classifier.ipynb

License

This project is licensed under the MIT License. Feel free to use and modify the code as needed.

Acknowledgments

• This implementation leverages the PyTorch framework.
• Grad-CAM visualizations were inspired by research on model interpretability.