In a world-first study, Terry Fox-funded researchers at Sunnybrook Research Institute are using a specialized ultrasound technique, together with machine learning, to predict tumour response early on in chemotherapy treatment; allowing for modifications for a more customized and responsive cancer treatment.
“Typically, it may take clinicians multiple treatment cycles to identify patients who are not responding to chemotherapy, and decisions regarding treatment intensification are typically deferred until after completion of chemotherapy treatments and tumour resection,” says Dr. Gregory Czarnota, senior author of the study and a radiation oncologist and senior scientist at Sunnybrook Health Sciences Centre.
“The data being generated from the serial quantitative ultrasound imaging used in this study is allowing us for the first time to use the power of machine learning to identify if and how a tumour is responding (or not) to chemotherapy, in order to make more accurate predictions and decisions. Our aim in radiation oncology is always to better target the tumour in order to minimize or reduce side-effects and increase our ability to destroy cancer cells.”
In current practice, no clinical, radiological or pathological test can reliably predict the response of an individual specific tumour before or shortly after treatment begins. However, a test capable of predicting response very early in treatment course could enable personalized adjustments, such as sparing patients from ineffective therapy and facilitating a timely switch to more effective treatment.
While the five-year study is currently ongoing to recruit a total of 240 research participants, initial data on the first 146 patients from this single-centre Phase 2 randomized controlled clinical trial was recently published in npj Precision Oncology.
The trial involves patients with locally advanced (stage II-III) breast cancer planned for standard neoadjuvant chemotherapy – a pre-operative strategy where chemotherapy is given before surgery. This approach has been shown to kill cancer cells early, shrink tumours, improve operability, allow clinicians to see how the cancer responds and improve outcomes.
The participants continue to be randomized into one of two arms: an experimental or an observational one using the current practice approach.
Patients in the experimental arm are evaluated – using ultrasound imaging to generate quantitative data about the properties and microstructure of their tumour, including the detection of early tumour changes at a microstructural level – to predict their response to treatment after an initial four weeks of a 16-to-18-week cycle of up-front chemotherapy. Machine learning algorithms then use the data to predict tumour response.
The imaging-response results and predictions are disclosed to medical oncologists in their role as treating physicians who have the discretion to adapt treatment based on the prediction results. This has been leading to adaptive modifications to chemotherapy for non-responding patients at the four-week point, changing their treatments and responses to chemotherapy.
In the preliminary data, response rates to the chemotherapy were 93.0 per cent with the standard current practice approach and 96.8 per cent with the experimental arm using the customized chemotherapy treatment approach. The imaging and computational methods used achieved an accuracy of 92 per cent and a predictive value of 99 per cent for responses to chemotherapy.
“These findings suggest that ultrasound imaging and AI can be used to provide an accurate early response prediction to better adapt and more effectively guide chemotherapy in patients with breast cancer,” adds Dr. Czarnota, also an associate professor in the Departments of Radiation Oncology and Medical Biophysics in the Temerty Faculty of Medicine at University of Toronto. “This research opens new opportunities for future clinical trials focused on adaptive treatment strategies. We are well on our way to enabling more personalized treatment adapation.”
Other imaging techniques like positron emission tomography (PET) have been shown to predict early responses to neoadjuvant chemotherapy, and highlighting the potential for personalized early treatment intensification. Compared to PET, quantitative ultrasound offers the benefits of being a lower-cost, faster and portable imaging modality with fewer technical challenges, making it a promising option for implementation in clinical practice.
This research team has been funded by a Terry Fox New Frontiers Program Project Grant since 2014.
Article originally published by Sunnybrook Research Institute.