“I like needles,” confesses Dr. Parvin Mousavi as she vividly recalls the one that delivered an epidural anaesthetic into her spine during childbirth. “Maybe I was distracted, but it was marvellous how well it went in.”
She subsequently learned how the use of such needles calls for a skill set that verges on an art form, namely the ability to find a sweet spot between vertebrae where the penetration can be most effective with the fewest complications. Unless there is some internal injury or deformity, the individuals who conduct this procedure generally succeed with no more information than what they can glean from outside a patient’s body. As for those problematic cases, Mousavi is improving their prospects through her own work as a professor in the School of Computing.
She credits that work to an innate sense of the fundamental contribution that medicine makes to all our lives. Even as Mousavi’s career rode the information technology boom of the 1990s, she remained fascinated by how hardware and software could capture the intricacies of complex processes, including those rooted in our own bodies. This interest has led her to look at ways of turning the vast amounts of data available from medical imaging and analysis into clinical progress with procedures such as needle insertion.
The need for such progress may be comparatively limited in the case of spinal insertions, but it is urgently required in another, even more intimate venue – checking the prostate for cancer. In this case clinicians find themselves poking this sensitive region as many as 12 to 20 times in order to obtain tissue samples. Should a substantial number of these samples contain cancer, a difficult decision-making process precedes any further tests or treatment. Doctors must weigh the genuine possibility of causing permanent damage – including erectile dysfunction – against the severity of the disease, which is all too often present only to a very limited extent.
“You’re not necessarily interested in diagnosing every single prostate cancer because most of them will not be the reason somebody passes away,” says Mousavi. “According to our Urology collaborators, for indolent cases of prostate cancer, over-diagnosis is not benefiting the patient because the intervention choices you’re giving them sometimes result in worse outcomes in terms of quality of life and overall health than the disease itself.”
With the goal of making those choices easier for physicians, she began working with specialists in urology, radiology and pathology to obtain ultrasound images of the prostate in a new manner. She refers to this novel imaging modality as “temporal ultrasound imaging”, which takes continuous images of a position in the prostate for a short amount of time.
“Traditionally, ultrasound manufacturers have used multiple images of the same site in an organ to improve the noisy ultrasound data,” she explains. “No one had ever thought of taking prolonged images and looking at the information content of this time course data.”
Among the key collaborators on clinical aspects of this system was Dr. Sandy Boag, an associate professor in Pathology and Molecular Medicine, who obtained entire prostate glands that were donated by patients for this research.
“First Parvin did the ultrasound imaging on the intact gland,” he says. “We had a special device to keep the gland in the same position while we sectioned it for microscopic examination, to preserve the same orientation so we could correlate the microscopic findings with the ultrasound results.”
The combined temporal ultrasound data and detailed pathology findings built a database to inform an intelligent computer system, which could then direct ultrasound imaging orientation in real time – while a patient is undergoing the test – and offer an accurate and personalized set of targets for biopsy needles.
Mousavi points out that Magnetic Resonance Imaging can also provide this kind of guidance, but it must be performed prior to biopsy and possible targets have to be aligned to the patient during the procedure. “Alignment itself has been a difficult problem in medical imaging for the past 20 years,” she argues. “It’s challenging.”
With temporal ultrasound imaging, this challenge can be overcome, as the image reveals in real-time those parts of the prostate that most likely contain cancer. These sites will be apparent on what she dubs a “heat map” that computer software can generate in a matter of seconds, so that it provides a vital surgical reference before any needle breaks the skin. Clinical feasibility studies have been performed at multiple centres in Canada and have also begun at the National Institutes of Health Clinical Center in Bethesda, Maryland.
“Our initial findings are exciting but you have to have enough examples from people with various types of tissue that are confirmed in pathology,” says Mousavi. “Then you teach your model reliably and it will be competent to create heat maps for patients that who haven’t been seen before. It provides a precise tool for the clinician suggesting ‘go and biopsy here’.”
Boag notes that their collaboration benefitted from his own engineering background, which gave them some common ground for discussion.
“We were able to speak some of the same language and that helped get it going,” he says, adding that the research has only recently begun to make the major transition from laboratory studies to application in real patients. “It shows a lot of promise but we’re not yet at the point where we can use her system to make the diagnosis.”
Nevertheless, he anticipates that this augmented targeting strategy could mean far fewer samples would have to be taken from a patient’s prostate to determine the type of cancer and the appropriate extent of treatment. Ultimately, the system might be capable of determining the nature of a patient’s prostate cancer without taking any samples at all.
This work is just one of several projects undertaken through the university’s Medical Informatics Laboratory, founded in 2003 and directed by Mousavi. Her work on ultrasound-guided interventions represent part of a wide-ranging initiative to employ predictive models and machine learning algorithms that can assist the doctors, nurses and technicians who operate on us.
“It’s important; it’s relevant,” she concludes. “Computers have the ability to help and an area that needs this help is medicine, because we are all mortal at the end of the day.”
