Chest CT Scan Image Classification

Project Objectives

The main objective of this project is to develop a deep learning model to classify chest CT scan images to detect the presence of lung diseases. This project has several sub-objectives:

1. Data Preprocessing: Implement preprocessing techniques to prepare the CT images for model training.
2. Model Construction: Use a convolutional neural network (CNN) architecture, specifically ResNet-50, to train the classification model.
3. Optimization and Evaluation: Optimize the model using advanced techniques and evaluate its performance on a test dataset.
4. Implementation and Usage: Develop an inference pipeline to use the model for classifying new CT images.

Technologies Used

This project uses several technologies and Python libraries for data processing, model training, and result visualization. Some of the main technologies used are:

• Programming Language: Python
• Environments and Tools: Jupyter Notebook
• Deep Learning Libraries:
  • PyTorch
  • Torchvision
• Image Processing Libraries:
  • OpenCV
  • Pillow
• Data Manipulation and Analysis Libraries:
  • NumPy
  • Pandas
• Visualization Libraries:
  • Matplotlib
  • Seaborn

Dataset

The dataset used in this project is the Chest CT Scan Images available on Kaggle. This dataset contains chest CT scan images categorized into different classes based on the detected lung condition.

Dataset Description

• Number of Images: 1000+ CT images
• Categories: Various categories representing different lung conditions
• Format: Images are in JPEG/PNG format
• Resolution: Variable

Techniques Used

Data Preprocessing

Before training the model, the images undergo a preprocessing pipeline that includes:

• Resizing: Adjusting images to a uniform size suitable for model input.
• Normalization: Scaling pixel values to improve training efficiency.
• Data Augmentation: Applying random transformations such as rotations, translations, and brightness changes to increase the variety of the training dataset and improve model generalization.

Model Construction

The main model used in this project is an optimized ResNet-50. Key steps include:

1. Loading Pretrained Model: Using a ResNet-50 model pretrained on ImageNet as a starting point.
2. Modifying the Output Layer: Adjusting the output layer to match the number of classes in our dataset.
3. Defining Loss Function and Optimizer: Using cross-entropy as the loss function and Adam as the optimizer.

Optimization and Evaluation

To optimize and evaluate the model, the following techniques were used:

• Training and Validation: Splitting the dataset into training and validation sets to monitor model performance during training.
• Evaluation Metrics: Using accuracy, recall, F1-score, and confusion matrix to assess model performance.
• Regularization Techniques: Implementing dropout and early stopping to prevent overfitting.

Implementation and Usage

The developed inference pipeline allows loading new CT images and classifying them using the trained model. The pipeline includes:

• Model Loading: Functions to load the trained model from storage.
• Preprocessing New Images: Applying the same preprocessing techniques to new images.
• Prediction: Generating and visualizing model predictions.

Installation

To install all the necessary dependencies, use the provided requirements.txt file:

pip install -r requirements.txt