madheshrajs/mri-reconstruction-using-densely-connected-residual-dcr-block

Convolutional Neural Network(CNN) with Densely Connected Residual(DCR) block is built. U-Net with Densely Connected Residual(DCR) block is built. High quality reconstructed images with less noise and superior visual quality is produced

MRI Reconstruction using Densely Connected Residual block

★ Convolutional Neural Network(CNN) with Densely Connected Residual(DCR) block is built.

★ U-Net with Densely Connected Residual(DCR) block is built.

★ High quality reconstructed images with less noise and superior visual quality is produced.

PROBLEM STATEMENT

Magnetic resonance imaging(MRI) is used to extract images of soft tissues of human body. It is used to analyze the human organs without the need for surgery.

Generally MRI images contain a significant amount of noise caused by operator performance, equipment and the environment, while processing if any noise is added to the image this can lead to difficulties in diagnostic characterization or object size.

Scope

The proposed project can reconstruct images from the existing image that is extracted from the MRI machine. The reconstruct image have higher quality and less noise.

Data Set

• https://www.nitrc.org/frs/?group_id=899

• 36 - 3Tesla Brain MRI dataset

• Dataset Size: 2.49 GB

Language

Python

Installation

The Code is written in Python 3.7. If you don't have Python installed go to Python website and install it. If you are using a lower version of Python you can upgrade using the pip package, ensuring you have the latest version of pip.

  pip install tensorflow
  pip install numpy
  

Working Environment

• Processor : AMD Ryzen 5 2500U with Radeon Vega Mobile Gfx, 2000 Mhz, 4 Core(s), 8 Logical Processor(s)

• Hard disk : 256 GB SSD

• RAM : 16 GB

• GPU : GTX 1050 Mobile 4GB VRA

Input 3D image

3D to 2D image

System Design

DCR architecture

CNN witH DCR architecture

U-NET with DCR architecture

Input of CNN with DCR

Output of CNN with DCR

Noisy input of CNN with DCR

Noisy output of CNN with DCR

Input of U-NET with DCR

Output of U-NET with DCR

Noisy input of U-NET with DCR

Noisy output of U-NET with DCR

Qualitative Metrics

The qualitative metrics(PSNR-Peak Signal To Noise Ratio) obtained with each model is presented in this section:

Future Enhancement

In future this work can be further improved with the help of different algorithms for improving the PSNR(Peak Signal To Noise Ratio)value. This work can be further developed by large datasets(2TB) and by using more techniques for improving the PSNR(Peak Signal To Noise Ratio)value and quality of the reconstructed image

Feedback

If you have any feedback, please reach out to us at [email protected]

Authors

• @Madhesh Raj S
• @Subamaarleshar B
• @Siva Prasath S