Image generation using GenAI for enhancing visual quality of radiology scans

Situation

In the medical field, imaging methods such as magnetic resonance imaging (MRI) are often used to obtain information about the internal state of a patient. MRI scans provide a 3D view of the body, but the resolution of the image can vary depending on the direction of the scan. This can be limiting for some clinical applications, such as the detection of malignant tumors.

Solution

The aim of this project was to develop a system for generating isotropic images from equivalent anisotropic MRI images. The isotropic image has the same resolution in all three directions, and a higher overall quality of image.
The solution of transforming an anisotropic image into an isotropic image consists of the following steps:
Registration: Three anisotropic scans are registered in plane and in 3D. This is done using a key-point-based deformation method.
Super-resolution: The registered scans are enlarged to isotropic size. This is done using an interpolation method.
Filling in missing voxels: In some cases, it may be necessary to fill in missing voxels in the isotropic image. This is done using an interpolation method.
The deployment was done using Docker technology, which allows for easy distribution and management of software applications.

Result

The development of the solution to create isotropic images from equivalent anisotropic MRI images was successful. The solution was developed in accordance with the client’s requirements and was tested and deployed successfully. The solution will allow medical professionals to visualize patient data more accurately and effectively.
The Customer is now evaluating the Solution in clinical practice (as an experimental tool) and raises funds for its further development, certification and commercialization.