Student Spotlight: Personalised Dosimetry for Radionuclide Therapy

Medical physics and radiobiology underpin much of modern cancer care, shaping how cancers are detected, monitored, and treated across nearly every clinical pathway in oncology. One research group working at the forefront of the radiobiology field is the Experimental Radiotherapeutics lab, led by Prof. Kate Vallis, which focuses on understanding and exploiting the biological effects of external beam ionising radiation and therapeutic radionuclides in solid tumours. On the medical physics side, the Clinical Academic Group for Medical Physics, led by Asoc. Prof. Daniel McGowan, conducts a range of medical physics and clinical engineering research within the hospital and university, with particular expertise in molecular imaging, including radiomics and image reconstruction methods.

Mark Macsuka is a Clarendon DPhil Scholar in Oncology in the Experimental Radiotherapeutics lab and the Clinical Academic Group for Medical Physics, where he works on personalised lutetium-177 (177Lu) dosimetry for radionuclide therapies. Using medical imaging data, Mark develops computational models that predict tumour response to radionuclide therapy, with the aim of informing individualised dosimetry strategies that maximise treatment effectiveness while minimising side effects. Below, Mark shares more about his project, offering insights into the role of AI in cancer treatment and the potential for improved outcomes through personalised treatment approaches.

What is radionuclide therapy?

Radionuclide therapy is a form of precision therapy that uses radioactivity. We have a drug that specifically binds to molecules associated with cancer cells, and we label this drug with a radionuclide. As the radionuclides decay, they emit high-energy particles that damage the cancer cells while largely sparing healthy tissue. In the context of neuroendocrine cancer, this particle is usually an electron from the decay of Lutetium-177 (177Lu).

Neuroendocrine tumours (NETs) are a good candidate for this therapy, as they often overexpress the somatostatin receptor, for which we have very specific drugs. These drugs accumulate in tumours but otherwise wash out from the system fairly quickly, usually with minimal off-target damage. Patients with NETs routinely receive four treatment cycles, each involving the injection of an activity of 7.4 GBq (Giga Becquerel). That means 7.4 billion decays per second at the time of injection. Some of this radiation is detected during scans to generate medical images that visualise how the drug is distributed throughout the body and how that changes over time.

A one size fits all approach

One problem with this treatment regime is that the injected amount is not personalised, and my research aims to work towards optimising this treatment based on dosimetry.

An important distinction when it comes to "dosing" in the context of radionuclide therapy is that "dose" has become an overloaded term. Usually, it refers to the mass amount of the drug injected, for which activity (the GBq) is a good analogue. However, in the context of radiotherapy, "dose" and "dosimetry" mean the amount of energy absorbed by cancer cells. This absorbed energy determines how many secondary particles are generated in the path of the original decay particle, which in turn relates to the frequency of DNA damage, and ultimately cell death. What I am interested in is measuring the second meaning, the absorbed energy, and using that information to guide how much activity to inject in the first place. These are often conflated, even in the field.

One-size-fits-all dosing (in the sense of injected activity) is suboptimal, as how much we inject is actually not very predictive of how much of the drug will go to the tumour or normal organs to deposit its energy. It is this energy deposition that causes treatment response and side effects, so optimising this is important.

At the end of the day, radionuclide therapy is a form of radiotherapy, so I believe it should be treated as such. In external beam radiotherapy, patients are prescribed doses (in the sense of energy absorbed), and there is a century of radiotherapy knowledge out there showing how much energy we need to give to tumours to expect a response or how much we can give to normal organs before expecting toxicities. This concept, at the moment, is underutilised in radionuclide therapy. My work focuses on calculating absorbed dose and estimating the expected response to radionuclide therapy, taking into account all of the physics and biology knowledge we already have from external beam radiotherapy. If successful, this approach would elevate radionuclide therapy by replacing one-size-fits-all prescribing with treatments that are tailored for both efficacy and safety.

Testing machine learning models in simulated clinical trials

My project uses data from a Canadian clinical trial (OZM-067) that aimed to assess the safety and efficacy of 50% dose escalation of Lu-DOTATATE in Cycles 2-4. My analysis includes 73 patients and 137 tumours, a fairly large dataset for neuroendocrine cancer. Each patient had four cycles of therapy and was imaged three times after each cycle using a SPECT scanner. I am using these medical images to identify NETs, calculate the absorbed energy to tumours and normal organs, and to see how these factors correlate with treatment response.

The computational aspects of my work include modelling how the drug behaves in tissues over time, how much energy deposition this corresponds to, and how these absorbed doses compare to those routinely used in external beam radiotherapy. Among simpler approaches, I am using machine learning to identify image-derived metrics that can predict if a patient's tumours will shrink.

Most recently, I found that the higher absorbed dose a tumour gets, the higher the probability of response (no surprise there), and that if a tumour gets a lot less in cycle 4 compared to cycle 1, then that correlates to response as well (some surprise here). My prediction models take these two findings into account to predict the probability that a patient is going to respond.

This is only half the story, though. Another model I am using from the literature estimates the probability of kidney toxicity, the main organ at risk, based on absorbed dose. I am combining the response and toxicity models into one big master model that estimates the "complication-free tumour control probability" for different administrations. These investigations act like "simulated clinical trials". By adjusting the various parameters, I can explore how different treatment strategies impact the complication-free tumour control probability. For example, if we give more in the first two cycles, does that increases this number, signalling that patients are more likely to have a complication-free response under this "front-loading" strategy? So far, what I have found is that yes, front-loading matters a lot. My data also suggests that we need to explore administering more treatment to patients per cycle than what we currently do, because it is safe and beneficial to do so for most patients. Dosimetry can select which patients to give these escalated injected doses to, and crucially, exactly how much.

Studying a DPhil in Medical Physics

Doing a DPhil in medical physics sets you up with a unique and transferable skill set. There is obviously clinical medical physics at hospitals, where you support the day-to-day running of radiotherapy, nuclear medicine, imaging, or radiation protection departments. It can be computational, like in my case, and take you into areas of deep learning, simulations, and prediction modelling. It is also wide-ranging in research scope, and can be very clinical, as it is for me. I sometimes do lab experiments with cells, or calibrate the scanners with radiation-filled cylinders, or do animal experiments to test a hypothesis generated from the results of my image analysis. Radionuclide therapy itself is still fairly niche, but it is growing exponentially. There are countless drugs in clinical trials and a massive potential for this modality to become routine in the future. Working in medical physics is also very multidisciplinary, and there is always something to learn. For a project like mine, you need to have a working knowledge of biology, physics, pharmacology, maths, and data science, and this multidisciplinarity naturally leads to many collaborations with specialists in these areas.

Written by Mark Macsuka

Video 1: SPECT/CT imaging after 177Lu radionuclide therapy, showing the distribution of 177Lu activity throughout the body.

Video 2: SPECT /CT reconstruction