/Transfer_Learning_on_breast_dataset

In this repository we will find codes on using tranfer learning to identify breast cancers from mammograph images. The models used were ResNet18 and ResNet34.

Primary LanguageJupyter Notebook

Breast Cancer detection With Deep Transfer Learning

Topic: Investigating The Detection of Breast Cancer With Deep Transfer Learning Using ResNet18 and ResNet34.

Introduction

This project investigates the detection of breast cancer using deep transfer learning techniques, specifically employing ResNet18 and ResNet34 architectures. The research utilizes mammogram images from the RSNA (Radiological Society of North America) dataset.

Framework and Libraries

  • FastAI
  • PyTorch(In FastAI)
  • Pandas
  • Matplotlib
  • Numpy
  • Torchvision(In FastAI)

Dataset

The dataset used consists of mammogram images obtained from the RSNA dataset. This dataset contains a large collection of images labeled with breast cancer-related information, making it suitable for training and evaluating deep learning models for breast cancer detection. You can download the RSNA dataset from RSNA dataset or use your own dataset.

Methodology

  1. Data Preparation: Pre-process and augment the mammogram images.
  2. Model Selection: We have two files one is named ResNet18 and the other ResNet34. The file you select will be the model I use with my dataset training.
  3. Model Training: Train the selected model using the RSNA dataset.
  4. Evaluation: Assess the model's performance using appropriate evaluation metrics.
  5. Comparison: Compare the performance of ResNet18 and ResNet34 models.

Files

In the repository, all the two models trained with has a Jupyter Notebook (.ipynb) and a Python(.py) file extension each.

Usage

To replicate the experiments:

  1. Clone this GitHub repository and run it in Google Colab.
  2. Download the RSNA dataset from RSNA dataset or use your own dataset.

In the codes

  1. Pre-process the dataset.
  2. Train the desired model.
  3. Evaluate the trained model.
  4. Analyze and interpret the results.

Results

The project repository contains model performance metrics and visualizations.

Demo Web APP

Conclusion

This project contributes to breast cancer detection research by investigating deep transfer learning techniques using ResNet18 and ResNet34 architectures. Findings can inform future developments in medical imaging and healthcare technology.

For questions or collaborations, contact the following research authors: