NA-MIC Project Weeks

Back to Projects List

Multimodal Registration MR2CBCT

Key Investigators

• Leroux Gaelle (University of Michigan, USA)
• Claret Jeanne (University of Michigan, USA)
• Cevidanes Lucia (University of Michigan, USA)
• Allemang David (Kitware, USA)
• Prieto Juan Carlos (University of North Carolina, USA)

Presenter location: In-person

Project Description

This project aims to develop a novel Slicer tool that combines machine learning with image processing technique with image processing techniques to automatically register MRI to Cone Beam CT (CBCT) images, enabling enhanced visualization and analysis of the TMJ complex. By integrating MRI soft tissue information with CBCT bony details, this automated technique provides clinicians with a more comprehensive patient-specific 3D model of the TMJ to improve diagnostic accuracy and treatment planning. Temporomandibular joint (TMJ) disorders affect a significant portion of the population and can cause chronic pain and disability. Accurate diagnosis is crucial for effective treatment planning, but can be challenging due to the complex anatomy and limited visibility of soft tissue structures on Cone Beam CT (CBCT) scans. MRI provides superior soft tissue contrast including the articular disc, but requires separate acquisition and manual registration with CBCT for detailed bone degeneration assessments.

Objective

The “Multimodal Registration MR2CBCT Project” aims to develop a sequennce of image anlaysis preprocessing steps prior to accurately aligning and overlaying CBCT and MRI multimodal images, using Elastix registration tools.

Approach and Plan

1. Dataset Collection:
   • Compile a comprehensive dataset consisting of MRI and CBCT files.
   • Perform manual approximation to align MRI and CBCT images initially.
   • Perform manual segmentation of the MRI.
2. Image Registration Strategy:
   • The primary goal is to achieve precise registration between MRI and CBCT images. To accomplish this, we are exploring two main approaches:

First Approach:

• Image Transformation Model:
  • Develop and train a model to transform MRI images into CBCT-like images.
• CBCT Registration:
  • Utilize existing tools to register the transformed CBCT images with actual CBCT images.

Second Approach:

• Automatic Segmentation:
  • Conduct automated segmentation of CBCT images as an initial step.
• Automated MRI Segmentation:
  • Train a model to automate the segmentation process of MRI images.
• Elastix-Based Registration:
  • Use Elastix to do the registration between MRI and CBCT images based on the segmentation. - Invert MRI to facilitate the registration process with Elastix. - Normalized MRI and CBCT 3. Validate:
• Validate the best method accuracy through rigorous testing against established benchmarks.
• Create the Slicer module interface
• Write documentationsand examples

Progress and Next Steps

Previous work

Dataset Collection:

• We compiled a comprehensive dataset consisting of MRI and CBCT files.
• Performed manual approximation to initially align MRI and CBCT images.

Image Registration Strategy:

Second Approach:

• Automated Segmentation:
  • Conducted automated segmentation of CBCT images as an initial step.
• Image Preprocessing:
  • Invert the gray scale level of the MRI
  • Normalize the MRI and the CBCT
• Elastix-Based Registration:
  • Working to use Elastix to do the registration between MRI and CBCT images using the manual segmentation. The MRI has been inverted to facilitate the registration process with Elastix.

Progress this week

• Harmonization of Mutilmodal MRI and Cone Beam CT (CBCT) / monai normalization
• Registration evaluation metrics : loss function (SSIM,NNC,NMI)

Next Steps

Image Registration Strategies:

Alternative Approach:

• CBCT Registration:
  • Develop and train a model to transform MRI images into CBCT-like images.
  • After finalizing the transformation model, utilize existing tools to register the transformed CBCT images with actual CBCT images.

Main Approach:

• Automate MRI Segmentation:
  • Train a model to automate the MRI segmentation process.
• Automate MRI approximation to CBCT:
  • Automatic approximation of the MRI on a CBCT.

Validation of the Multimodal Registration:

• Validate the Motimodal Registration result through rigorous testing.
• Create the Slicer module interface
• Write documentation and examples

Illustrations

Overlay of the CBCT automated segmentation on a MRI

Invertion of an MRI

Approximation of an MRI on a CBCT

Background and References

No response