New diagnostic tool could hasten ovarian cancer treatment
A new type of diagnostic system using sophisticated computer software capable of analyzing and comparing cancerous tissue against a vast databank of digital images of cancer samples may speed up treatment for ovarian cancer.
The system, being designed by Aicha BenTaieb, a Simon Fraser University computing science PhD student, aims to automate the identification of ovarian carcinomas for a faster and more reliable diagnosis.
Ovarian cancer is the fifth most common cancer for women. It is estimated that this year in Canada, 2,800 women will be diagnosed, and that 1,750 women will die from it.
The outcomes haven’t significantly improved in over 50 years. Its causes are unknown and there is a limited understanding of its progression.
What is known is that there are five main subtypes. Effective treatment depends on identifying the subtype as soon as possible. But current methods are subjective, time-consuming and prone to error.
Using information collated via computers, BenTaieb believes she has found a better way to identify these subtypes.
Each subtype shows individual structural and cellular characteristics. Currently, pathologists analyze tissue samples using a microscope, digital scanner and computer software.
However, identification can easily be impaired by technical factors such as lighting and the pathologist’s experience. The analysis can also be timeconsuming.
“What happens now is a pathologist is often unsure, so he’s most likely to ask for a second opinion,” BenTaieb said.
“Depending on where you are, in Vancouver or the Yukon, that second opinion is going to take more or less time. And if you don’t have access to an expert, often the start of the treatment is delayed.”
The pathologist might also perform extra tests that are costly and not always available in every pathology centre.
With BenTaieb’s method, an artificial intelligence feature is integrated into the software that helps the pathologist analyze the tissue sample. This feature is trained, through a large data set of expert-annotated slides, to automatically identify the characteristic visual patterns for each subtype of carcinoma.
“We’re looking at the whole image, in different regions, in a more efficient way, using computers to extract image-based features,” BenTaieb said.
BenTaieb is designing the system at SFU’s Medical Image Analysis Lab under the supervision of computing science professor Ghassan Hamarneh, and in collaboration with pathologists Dr. David Huntsman and Dr. Hector Li Chang from the BC Cancer Agency.
The technology is still in its early days, and implementation in operating rooms and pathology centres may take months, if not years.
But the results have been promising. The system is already able to correctly classify the subtype in more than 92 per cent of cases.
Ideally, the system will eventually be in place in operating rooms, and take place in real time.
“A clinician or surgeon would cut into the patient,” BenTaieb said, “take some tissue, look at a sample under a microscope, and then the software would automatically say, ‘ it’s this subtype.’”