A team from the CHUM Research Centre (CRCHUM) and Polytechnique Montréal combines Raman microspectroscopic imaging with machine learning techniques to better detect aggressive forms of prostate cancer, making diagnosis accurate almost 9 times out of 10.

This year, 4,200 Canadian men will die of prostate cancer, making the disease the third leading cause of cancer deaths in the country. In 20 per cent of these cases, an aggressive form, intraductal carcinoma of the prostate, will be detected. Trouble is, pathologists in hospitals don’t have access to biomarkers to accurately identify it, and instead have to rely on visual observation of tissue samples.

Is it possible to better identify at-risk patients and improve the diagnosis of the most aggressive forms of this cancer?

Yes, say CRCHUM researchers Dominique Trudel and Frédéric Leblond, along with postdoctoral researcher Andrée-Anne Grosset. In a study recently published in PLoS Medicine, the scientists make the case for better detection, arguing it’s essential, given that this type of carcinoma is often associated with a recurrence of prostate cancer and metastases.

Molecules laid bare

In their study, the CRCHUM team analyzed tissue samples from 483 prostate cancer patients from the CHUM, the Centre hospitalier universitaire de Québec and the University Health Network, in Toronto.

They first identified the molecular signature specific to each sample using Raman microspectroscopic imaging, a technique at which Leblond, a professor in the Department of Engineering Physics at Polytechnique Montréal, excels.

Essentially, with this technique, researchers use rays of light to induce vibrations in the molecules of a sample and collect information about the chemical bonds that make it up.

Along with Grosset and Leblond, Trudel, a pathologist at the CHUM, used the hospital’s collection of Raman spectra to train algorithms to recognize and automatically classify the specific signatures of healthy tissues, intraductal carcinoma of the prostate and other forms of prostate cancer.

A tool of the future

Based on machine learning, these predictive models were then tested on the data from the two other hospital centres. The results were highly promising: the scientists were able to detect the presence or absence of intraductal carcinoma of the prostate in nearly 9 out of 10 cases.

One other advantage, according to the researchers: this new technique is faster and less expensive than those currently used in laboratories.

Before the technique can be implemented in a hospital setting, more work has to be done in terms of validating these initial results on a larger scale and refining the models, they added.

The hope is that the combined use of Raman microspectroscopic imaging and machine learning and its improved diagnostic accuracy will eventually help men with relatively harmless prostate cancer avoid unnecessary treatment.

Additional information

• Tissue samples from 483 patients were analyzed: at the CHUM (272 patients; median age = 62), Centre hospitalier universitaire de Québec (135 patients; median age = 62) and University Health Network (76 patients; median age = 61);
• This research project received funding from the CRCHUM, IVADO, the TransMedTech Institute, Mitacs, the Institut du Cancer de Montréal, Prostate Cancer Canada, the Fonds de Recherche du Québec—Santé, the Canada First Research Excellence Fund, the Ontario Institute for Cancer Research, the National Cancer Institute (National Health Institutes) and the Natural Sciences and Research Engineering Council of Canada.