Skip to main content
Donate
Using AI to predict tumour response
​​​​​(L to R), Caryn Geady is first author of the study, and Dr. David Schultz and Dr. Benjamin Haibe-Kains are co-senior authors. Photo: UHN Research Communications
For patients with metastatic cancer, individual tumours have different sensitivities to cancer therapies. A group of scientists from UHN's Princess Margaret Cancer Centre has introduced a new computational method for predicting tumour-specific responses to treatments in patients experiencing metastasis.

"As cancer develops, subpopulations of cells arise with differences in their molecular characteristics and tumour microenvironment," says Dr. Benjamin Haibe-Kains, Senior Scientist at the cancer centre and senior author of the study. "This can lead to a situation where there is a large amount of heterogeneity in cancer cells within an individual patient.

"Cancer heterogeneity is associated with poorer treatment outcomes, and must be addressed to improve precision oncology," says Dr. Haibe-Kains, who is also a Tier 2 Canada Research Chair in Computational Pharmacogenomics and professor in the Department of Medical Biophysics at the University of Toronto (U of T), and the Scientific Director at Cancer Digital Intelligence.

The differences in characteristics between metastatic sites in a patient create a situation where tumours have a varied response to treatment.

Recently, radiomics — a field of medical research that involves extracting and analyzing quantitative features from medical images such as CT scans — has emerged as a potential way to predict treatment outcomes.

"We investigated the use of radiomic biomarkers to predict tumour-specific treatment resistance in patients with leiomyosarcoma — a cancer that arises from smooth muscle cells — that has spread to multiple sites," says Caryn Geady, doctorall student in Dr. Haibe-Kains' lab and first author of the study.

"We looked at 202 lung metastases from 80 patients and examined both pre-treatment and treatment follow-up CT scan features to use advanced machine learning techniques to develop a model to predict the progression of each metastasis."

For each lesion, or tumour area, that was analyzed, the relative change in lesion volume from baseline was evaluated as a treatment response metric. Researchers then tested their models for their ability to accurately predict tumour response.

The team found that their model using radiomic biomarkers provided a 4.5-fold increase in predictive capability compared to a no-skill classifier — a model used as a baseline to compare the performance of more advanced models.

"This research shows that predicting individual tumour responses offers a novel strategy to manage metastasis," says Dr. David Shultz, a clinician investigator at the Princess Margaret, co-senior author of the study and an associate professor of radiation oncology at U of T.

"It has the potential to guide selective targeting of treatment-resistant cells alongside systemic therapy."

This work was supported by the National Cancer Institute of the National Institutes of Health and The Princess Margaret Cancer Foundation.

Read more about the study.


 This story first appeared on UHN News
 
Build-Hotfix130_165-v2