To reach a breakthrough in computer analysis of medical images, University of Waterloo engineering Prof. Hamid Tizhoosh and his colleagues had to acknowledge a shortcoming that is hard for engineers to admit.
Engineers like consistency and objectivity, Tizhoosh says. There have to be rules for everything that are the same for all users and subjects.
But humans don’t work that way. Their judgments are inherently subjective.
That’s why previous attempts to electronically analyze medical images to doctors’ satisfaction had failed. No matter how finely tuned its algorithms, there was simply no way to design a computer program that could identify objects the way a given person saw them.
Tizhoosh says his team has overcome this difficulty by merging years of research on artificial intelligence with the field of medical imaging. The result is a computer program, Segasist, that gradually learns a doctor’s biases and preferences until it can think just like that doctor when analyzing an image.
The results have been promising enough to earn a $750,000 venture capital investment for Omisa Inc., the spinoff company created to commercialize the technology.
Incorporated last spring, the company already has five employees and will open its first office in Toronto next week.
Omisa stands for Omni-Modality Intelligent Segmentation Assistant. Segmentation refers to the identification of objects, such as organs, lesions and tumours, from an MRI, CT scan, ultrasound or similar image.
Because existing computer programs do such a poor job at segmentation, doctors often have to pick out objects themselves using programs like Adobe Photoshop. This can be time-consuming when dealing with many images.
“You see highly qualified surgeons sitting in their offices using the mouse and manually clicking point by point where is the tumour for those images,” Tizhoosh says.
The professor first became aware of the problem when he was working on his PhD in Germany in the 1990s. Part of a project to improve the quality of images for radiation therapy, he saw how difficult it was to meet physicians’ demands.
“The way doctors do this cannot be put into equations,” he says. “The information was ambiguous or vague.”
The Iranian-born Tizhoosh came to Canada in 2000 and joined UW a year later. He developed the Omisa technology over the next several years with the help of graduate students.
In 2006, a $25,000 Ontario Centres of Excellence grant helped the team develop a prototype.
Tizhoosh turned to the university for help commercializing the technology. The university will receive 25 per cent of Omisa’s revenues in exchange for its role.
Last year, intellectual property consultant Jacqueline Csonka-Peeren, a veteran of electronics manufacturer Celestica Inc., was hired as Omisa’s president.
In December, venture capital firm First Leaside Visions LP of Uxbridge invested $750,000 in Omisa.
“We felt they had a great product that could demonstrate both cost savings and better outcomes in the health-care system,” says Douglas Hyatt, a consultant with First Leaside. “We felt that with our money and other funds they were able to raise, they would be able to do their development work throughout this time and will have a market open to them when we pull out of this recession.”
Omisa has hired a programmer to turn the prototype into a commercially viable piece of software. The company expects radiologists to try out the software this summer.
The goal is to have the software ready for the Radiological Society of North America convention in Chicago next fall.
Clinics and hospitals would be the target market for the software.
The technology could have applications in fields outside of medicine, such as astronomy and mining, Csonka-Peeren says.
BinaryHealthCare (BNHC) 