MRIgRT: Inside AI Tumor Tracking

Hello World!

There is always something interesting happening in medical imaging, but every now and then I come across something that makes me stop and think, “Okay, this is actually pretty cool.”

This time, it was AI powered tumour tracking. We often hear about AI detecting abnormalities or helping analyze medical images, but what if AI could actually help us follow a tumour while it is moving? Now, you might be wondering, “Why would a tumour move?” Well, it depends on where it is. Tumours in the chest and abdomen can move as a patient breathes. A lung tumour, for example, doesn’t stay perfectly still during radiotherapy. And that creates a pretty important problem. How do you accurately deliver radiation to something that is constantly moving?

This is where AI comes in.

Radiotherapy is all about precision. We want to deliver the prescribed dose to the tumour while limiting the dose received by surrounding healthy tissue. AI is being investigated for several parts of this process, including tumour localisation, segmentation, motion tracking and motion prediction. The last one is particularly interesting. Instead of simply asking:

“Where is the tumour right now?”

we can ask:

“Where is the tumour going to be a fraction of a second from now?”

Because there is always some delay between acquiring an image, processing it and responding to the movement. If the tumour moves during that delay, knowing where it was may not be enough. AI could potentially learn the pattern of that movement and predict where the tumour will be next. So basically: The tumour moves. Imaging sees it . AI predicts the movement. Treatment responds. Sounds a little like science fiction, right? Well, researchers are actually working on it.

What does the research say?

A recent 2026 systematic review looked at AI for real time motion tracking in MRI guided radiotherapy(MRIgRT), reviewing 28 studies. Many reported localisation or prediction errors of less than 2 mm in their respective study settings. That’s pretty impressive, isn’t it? But here’s where we need to pump the brakes a little. The review also found that 23 of the 28 studies focused on the liver, only five reported external validation, and prospective clinical and dosimetric validation remains limited. So while the results are promising, we aren’t at the point where we can simply tell an AI system:

“Find the tumour and follow it.”

Not yet, anyway.

And then Philips enters the conversation

This is what made me connect the research to a recent CNBC interview with Philips CEO Roy Jakobs. In the Philips’ Q2 2026 results, Philips highlighted its developments in AI and MRI including a 4D MR solution for radiation therapy planning designed to visualise moving tumours. And this is where MRI becomes particularly interesting. Instead of looking at anatomy as one frozen 3D image, 4D imaging adds time to the equation.

3D + time = 4D

For a tumour that moves with respiration, that extra dimension can provide valuable information about how the tumour changes position during the breathing cycle. Combine that with AI capable of analysing and potentially predicting motion, and you start to see where this technology could be heading.

But can AI really follow the tumour?

Well…

Not perfectly. Not yet.

There are still questions about how these systems perform with different patients, tumour locations, breathing patterns and image qualities. And if AI is going to influence radiation delivery in real time, the level of validation and quality assurance required is understandably very high. As a radiographer, I don’t see this as AI replacing us. Someone still needs to understand the patient, the imaging, the treatment and perhaps most importantly, when something doesn’t look right. Technology can provide information. Clinical judgement gives that information meaning. I think we’re moving towards a future where medical imaging isn’t simply about taking an image and looking at it afterwards. We’re moving towards real time imaging, AI analysis and adaptive treatment. The tumour moves. AI tracks. The treatment responds. We’re not quite there yet. But honestly, I’m very curious to see where this goes.

Stay curious.

Grace…


Comments

Popular posts from this blog

Hello World!

MRI in 60 Seconds?: Inside Ultrasonic CT

AI and MRI: Inside Multiple Sclerosis