Radiomics to predict the risks of relapse

- 4 min
Marion tonneau

At the age of five, Marion Tonneau knew that she would become a pediatrician. As a young girl, she was already taking care of her family members by prescribing Smarties, neatly stowed away in her doctor’s kit. The candies have since been replaced and the little girl became a doctor who is now passionate about artificial intelligence.

Dr. Tonneau could have completed her medical specialty in Lille, France. But her desire to break out of her comfort zone brought her to Montreal, a city that she had discovered in July 2017 during an anaesthesia clerkship at Hôpital Notre-Dame.

This time she came with a fellowship from the Nuovo-Soldati Foundation for young medical researchers carrying out cancer research. Currently a radiation oncology resident, she first set foot in the Research Centre in February 2021 at the height of the COVID-19 lockdown. But that didn’t spoil her positive outlook.

Today, she works with Dr. Houda Bahig’s team on a project involving artificial intelligence and pulmonary radiotherapy, which, in 2022, received funding over three years from the Canadian Institutes of Health Research.

A story of imaging and big data

In recent years, medical imaging has undergone an unprecedented revolution : radiomics. Its goal is to better characterize tumours through the large-scale analysis, extraction and mining of data from images taken during regular imaging tests.

The idea here is to succeed in detecting, in the images, information the naked eye is unable to see, which is therefore inaccessible to doctors.

“Thanks to radiomics, I hope to predict the risk of local relapse in patients with localized non-small-cell cancer, i.e. patients not presenting metastases and  whose lymph nodes are not involved.” 

These patients will have previously been treated by stereotactic radiotherapy, a treatment that consists of aiming, with extreme precision, a multitude of very small, high-energy radiation beams towards the area needing to be treated.

On paper, her project seems simple. The reality is another matter.

An enormous amount of preparation

To succeed in establishing a radiomic signature, you have to design and train a prediction algorithm to analyze tumour images taken by a scanner and automatically extract biological information of relevance to the medical team.

“In our case, we work with a clinical database of 1,000 patients treated at the CHUM. To begin, we extracted from medical files all the radiotherapy planning scans taken prior to treatment. These scans are used to configure the machine to deliver the right dose of radiation at the right locations while minimizing toxicity.” 

When undergoing stereotactic radiotherapy for lung cancer, patients benefit from a “4D scan,” during which they breathe normally so that the position of the tumour to treat can be determined based on the respiratory cycle.

“We’ll use these “4D scans,” already annotated by the clinicians, to extract the radiomic features of the tumour to predict risks of relapse. A “4D scan” is made up of about 10 images per patient. With 1,000 patients, that makes a large quantity of data to analyze!” 

Letting the data talk

Once the analytical phase has been completed, Dr. Tonneau and her colleagues will associate the radiomic features with the known clinical data of each patient to define their prediction algorithm.

“Because we work retrospectively, we already know what patient has relapsed or not and at what point in time. Now, the idea is to test our predictive model on an independent cohort in another hospital.” 

In a few years, the young researcher hopes to be able to determine the profile of a patient at risk of relapse, modify the therapeutic strategy accordingly and optimize treatments.

Working in radiomics, an emerging sector of artificial intelligence, is a source of daily motivation for her, someone who knew nothing about it before coming to Montreal.

 I’ve loved working here. My experience at the Research Centre confirmed my desire to pursue a career as a clinician and researcher. I find it very intellectually stimulating. 

Before pouring herself heart and soul into this adventure, Dr. Tonneau must complete her residency in Lille, which means another two years in France. She would like to continue the project as a France-Quebec collaboration and possibly return to Montreal one day.

Radiomics to predict the risks of relapse