Unlocking AI-powered approaches to cancer treatment and detection

October 2, 2023

Health Insights Research

By Natalie Richard

New research from Vector Faculty Member Parvin Mousavi leverages machine learning techniques to help enhance cancer care. Her work could lead to more accurate breast cancer surgeries and quicker detection and diagnosis of prostate cancer. The work speaks to the potential of AI models to increase the efficiency of care for cancer, the leading cause of death in Canada.

Improving breast cancer removal with AI research

“Everyone knows someone or have themselves gone through [cancer],” says Mousavi. Currently about 25 to 30 percent of breast cancer surgeries have positive margins which means patients will need follow-up surgeries to remove the remaining cancerous cells.

Blob

“No surgeon wants to leave cancer behind, so if they don’t think they have [all the cancerous tissue], they may take more healthy tissue than they really need to. There needs to be something better than this and that is where my research comes in.”

Parvin Mousavi

Vector Faculty Member

When treating breast cancer, surgeons create and cauterize incisions. Using an existing cutting-edge technique called rapid evaporative ionization mass spectrometry (REIMS), the resulting plume of smoke from cauterized incisions is suctioned through a device called a mass spectrometer which immediately analyzes the biochemical profiles of the burnt tissue samples. Within a few seconds, clinicians can potentially get a clear idea of the makeup of the tissue they are cutting, learning in near real-time when they’ve reached or inadvertently crossed the edges of the cancerous tissue they’re trying to remove.

“As you burn, you can imagine that you are burning through all sorts of tissue, so that is very noisy data. You don’t have labels — this is surgery, not pathology — because you are burning the tissue as you move. The problem is infinitely difficult.” 

Leveraging AI provides important insights in understanding these complex datasets. Mousavi and her team are building powerful deep learning models capable of accurately categorizing the type of tissue burned, helping surgeons to conserve healthy tissue while removing cancerous cells. To make it work, they trained the program using data collected with a surgical tool called the iKnife, powered by REIMS technology, allowing the model to tell the difference between types of tissue based on their chemical signatures.

In applying machine learning techniques to surgical navigation, Mousavi’s research aims to completely remove all cancerous cells with minimal healthy tissue loss, reducing the need for additional surgeries and improving patient outcomes. 

Given the promise of their initial results, Mousavi’s technology is currently in a feasibility trial at the Kingston Health Sciences Centre. A number of hospitals in Europe are also exploring the technology. Mousavi also recently received a $700,000 grant for the team’s work by the Canadian Institutes of Health Research. The money will allow Mousavi and her team to explore NaviKnife, a next-generation imaging and surgical navigation system that builds off the iKnife device and uses novel real-time metabolomic tissue typing, machine learning, and image-guided tracking to assist a surgeon in identifying and tracing a tumour boundary during surgery. 

Blob

“This is the most exciting part for me. My passion [for AI and cancer research] is that I want to see it in clinics and have something that I developed in use.”

Parvin Mousavi

Vector Faculty Member

Enhancing prostate cancer detection with AI-enhanced ultrasound

Another area of research Mousavi and another team of researchersare exploring is prostate cancer detection by accelerating AI adoption for advanced ultrasound technology. 

Prostate cancer accounts for one-fifth of all new cancer cases in men. However, the chances of a successful recovery significantly increase when the cancer is treated early. To detect prostate cancer, patients receive an MRI and then an ultrasound which helps clinicians figure out where to biopsy for definitive diagnosis. But MRI wait times in Canada can be long. To get around this problem, Mousavi and her team partnered with Exact Imaging, a Toronto-based medical imaging company that specializes in high-resolution micro-ultrasound systems, to devise a system that can provide real-time imaging and biopsy guidance. 

Unlike typical ultrasound machines, Exact Imaging’s high-frequency ultrasound is ideal for imaging the prostate. “Conventional ultrasound imaging is only helpful to navigate biopsies but on its own is not sufficient for cancer diagnosis, while high-frequency imaging allows for observing changes almost at the cellular level,” says Mousavi. 

To further enhance the image, Exact Imaging is exploring an AI-enabled approach to identify the most likely areas of cancer from ultrasound images during biopsy. Mousavi and her team have created models specifically for guiding prostate cancer biopsies using multi-centre trial data. “When applying our AI framework, we’ve had much better luck with the association of the ultrasound images with cancer in the prostate.”

The specificity and accuracy of the AI model offers promising results in detecting cancer. This high-resolution imaging technology enhanced by AI, once validated, could mean bypassing the need for MRIs for some patients, leading to faster diagnosis, and critically, enhancing patient outcomes. 

Researching AI potential today for better patient outcomes tomorrow

Mousavi and her team have already received several academic accolades for their work, including the Best Presentation award at this year’s Information Processing in Computer-Assisted Interventions (IPCAI) conference. Their work, and that of other AI-researchers and practitioners from the Vector community, could lead to more personalized and better care, improving efficiency in our healthcare system, decreasing wait times, and improving outcomes for many Canadians. But she emphasizes that when deploying AI in health, the pipeline from theory to practice requires everyone involved to be on the same page and with appropriate ethical guidelines. 

“As wonderful as AI is, it only has an impact in health care with a multidisciplinary team that is willing to work together — clinicians, engineers, students, everyone. If one of us falls, the rest of us can’t stand.”

Click here to learn more about the Vector Institute’s work in the health AI ecosystem.

Related:

Headshot of Vector Faculty Member Xi He.
Insights
Trustworthy AI

How Vector Researcher Xi He uses differential privacy to help keep data private

Solar panels sit on green grass. There are trees and skyscrapers in the background
Insights

Harnessing AI For Sustainability

Three people stare at a laptop with a Vector logo on it
Insights
Research

Vector Research Blog: Structured Neural Networks for Density Estimation and Causal Inference