This project is the implementation of "CutPaste: Self-Supervised Learning for Anomaly Detection and Localization" for the detection of brain anomalies within the scope of the Master-Seminar: Unsupervised Anomaly Detection in Medical Imaging (IN2107, IN45010) seminar course.

First, you have to initialize the environment

Note Using a virtual environment would be better

pip install -r requirements.txt

Download and extract the data

Note Ensure that the data folder is located in the same directory as the project and is configured appropriately according to the model.yaml configuration file.

wget <link of the data>
unzip data.zip

This project utilizes two distinct datasets for training.

1. fastMRI Dataset

   • Source: fastMRI
   • Description: The fastMRI dataset comprises MRI images that have been collected using accelerated MRI techniques, allowing for faster acquisition times.
   • Number of Images: 130

2. IXI Dataset

   • Source: IXI Dataset
   • Description: The IXI dataset consists of brain MRI images collected from the IXI project, providing a diverse set of images for analysis and processing.
   • Number of Images: 581

The tests were conducted based on the parameters outlined in the table below.

Comprehensive ROC Curve figures and detailed results for each pathology are available in the results folder.