Method for Image Reconstruction using Unsupervised Deep Learning and System thereof
Categories for this Invention
Technology: Image reconstruction using unsupervised deep learning techniques;
Industry & Application: Biomedical Engineering, Healthcare Industries, Magnetic Resonance Imaging(MRI) units, Medical Device;
Market: The global 3D reconstruction technology market is projected to grow at a CAGR of 11.6% during 2024-2029.
Image Gallery
Problem Statement
- In the present era, various techniques like nuclear imaging, magnetic resonance imaging, computerized tomography scan which may be used to obtain images of internal structures of objects or patients.
- However, these techniques subject to various trade-offs between speed, efficiency & quality of reconstruction.
- Hence, there is a need to address said issues.
Technology
- Present Invention explains about a system & method for image reconstruction using fully unsupervised deep learning techniques.
- Further it explains that a system of one or more computers can be configured to perform particular operations or actions by virtue of having software, firmware, hardware, or a combination of them installed on the system that in operation causes or cause the system to perform the actions.. like a method for training a neural network for image reconstruction.
- The method includes the following steps depicted in the smart chart hereinbelow:
STEP 1
- Calculating loss function (L(X,X′)) & projection data error (X-X′) based on actual projection data (X) & reference reconstructed projection data (X′);
STEP 2
- Updating one or more parameters of the neural network based on the calculated loss function (L(X,X′));
STEP 3
- Reconstructing an image (Y′) by processing the actual projection data error (X-X′) using the one or more updated neural network parameters;
STEP 4
- Transforming the reconstructed image (Y′) into new reference reconstructed projection data (X′);
STEP 5
- Iteratively performing steps S1-S4 for a predefined number of cycle for calculating an optimum loss function (L(X,X′));
STEP 6
- Generating one or more optimum parameters of the neural network using the optimum loss function (L(X,X′));
STEP 7
- Updating the neural network with the one or more optimum parameters, & Transforming the reconstructed image (Y′) by processing the projection data (X) using a trained neural network.
Key Features / Value Proposition
Technical Perspective:
- Facilitates a software framework for image reconstruction by combing the Deep Learning (DL) & the Iterative Reconstruction (IR) techniques.
- Provide fast, fully unsupervised & robust image reconstruction technique.
- Advantageous to reconstruct tomographic images without any noise/blur artifacts & allows reconstruction from the truncated data without the need for prior truncation correction.
- The present techniques do not restrict the solution space by using regularization term in the loss function.
Industrial Perspective:
- Efficient cost-effective solution and applicable in the medical imaging system to reconstruct the image.
- Provide speedy solution.
- Facilitates high quality of reconstructed image as shown in fig 2.
- Easily installed on the system that in operation causes the system to perform the action of reconstruction of image.
Questions about this Technology?
Contact For Licensing
sm-marketing@imail.iitm.ac.in
ipoffice2@iitm.ac.in
Research Lab
Prof. Balaji Srinivasan
Department of Mechanical Engineering,
Prof. Ganapathy Krishnamurthi
Department of Engineering Design
Intellectual Property
IITM IDF Ref. 2297
IN Patent No: 485152
Technology Readiness Level
TRL-4
Proof of Concept ready, tested in lab.