AI steered interventional MRI

Background

MRI allows minimally invasive treatment of cancer and other diseases. MRI accurately reveals a tumor within an organ or at difficult to reach locations. Using real-time MRI visualization, special needles are inserted such that the tip is within the tumor. Several types of treatment exist. One such treatment is to heat the tumor tip. The tissue surrounding the tip is exposed to the treatment and tissue cells in that region are killed and subsequently broken down by the body. The size of the region depends on the duration of the treatment. This type of therapy has several advantages. MRI guided interventions prevent having to search for the tumor reducing disruption to anatomical structures surgically. The real-time feedback on the effect of the intervention ensures excellent coverage of the tumor region. The minimally invasive procedure allows faster recovery of the patient.

The accuracy of interventional MRI is unfortunately limited because during needle intervention, the body deforms and structures disappear from the imaging display. The MRI needs to be manually adjusted to acquire images in a new imaging plane that shows the structure in relation to the needle. This repeated MRI adjustment takes considerable time and leads to suboptimal needle placement.

The goal of this project is to develop Artificial Intelligence (AI) to improve MRI guided interventions by automatically tracking tumors and automatically steering MRI acquisitions. We have an MRI scanner that can be digitally steered. AI can be developed such that it can track an object from a series of MRI images. The developed AI output can be used to steer the MRI. With the improved visualization, interventions can become more accurate in less time.

Tasks

In this pilot project, the focus is on simulated interventional MRI. The tasks are as follows:

  • Develop an MRI simulation module. Synthetic MRI image series should show moving tumor targets with an adjustable motion to simulate varying levels of difficulty.
  • Develop an AI tracking module that predicts the tumor target location from a series of MRI images.
  • Develop a demonstrator that compares conventional versus AI-driven interventional MRI

People

Tristan de Boer

Tristan de Boer

Master student

Data Science, Radboud University

Henkjan Huisman

Henkjan Huisman

Associate professor

Diagnostic Image Analysis Group

Patrick Brand

Patrick Brand

PhD student

Diagnostic Image Analysis Group