New imaging software adapts to radiologists
Researchers at the University of Waterloo in Ontario, Canada, have utilized artificial intelligence research on medical imaging to create an application that gradually "learns a doctor's biases and preferences until it can think like that doctor when analyzing medical images."
The prototype's results have been promising enough to garner a $750,000 venture capital investment for Omisa, a spinoff company created to commercialize the technology. Omisa is an acronym for Omni-Modality Intelligent Segmentation Assistant.
Engineering Professor Hamid Tizhoosh, PhD, told the Iran Times that he came up with the idea for Segasist while observing that radiologists and other clinicians, including surgeons, spend a lot of time on manual marking of lesions and tumors either to make a diagnosis by characterizing the tumor or to plan the treatment.
"I also realized that there is a huge gap between the objective understanding of engineers on one side and the subjective perception of medical experts on the other side. I started working on methods that could bridge that gap," he said.
In 2006, a $25,000 grant from the Ontario Centers of Excellence helped the team develop a prototype based on their technology. The University of Waterloo helped commercialize the technology; for that, the university will receive 25 percent of Omisa's revenues.
The prototype is set to be turned into commercial software, and Omisa expects radiologists will be able to try out the program this summer. Omisa's goal is to have the software ready for the 2009 RSNA conference in Chicago, reported Iran Times.
"Presently, all available software solutions in medical image analysis deliver results that the medical expert--radiologist, oncologist, surgeon--usually does not find accurate with respect to the medical knowledge. Hence, the medical experts frequently modify the results--mainly the outline of the lesion or tumor. However, the existing software solutions do not learn anything from these user modifications and continue to deliver the same result over and over again if the same image or a similar one is processed," Tizhoosh said. "Our software learns over time how the radiologist changes the results in order to provide more accurate tumor marking, which results in the most striking feature of this software."
"In addition, since Segasist stores the knowledge of individual radiologists, it can provide a second opinion to any radiologist if the software is being used within network of radiologists-e.g. a hospital or, even better, a network of hospitals," he added. "This second opinion is automatically created based on the stored knowledge of all radiologists and can be provided instantly. This will reduce errors and save costs."
The prototype's results have been promising enough to garner a $750,000 venture capital investment for Omisa, a spinoff company created to commercialize the technology. Omisa is an acronym for Omni-Modality Intelligent Segmentation Assistant.
Engineering Professor Hamid Tizhoosh, PhD, told the Iran Times that he came up with the idea for Segasist while observing that radiologists and other clinicians, including surgeons, spend a lot of time on manual marking of lesions and tumors either to make a diagnosis by characterizing the tumor or to plan the treatment.
"I also realized that there is a huge gap between the objective understanding of engineers on one side and the subjective perception of medical experts on the other side. I started working on methods that could bridge that gap," he said.
In 2006, a $25,000 grant from the Ontario Centers of Excellence helped the team develop a prototype based on their technology. The University of Waterloo helped commercialize the technology; for that, the university will receive 25 percent of Omisa's revenues.
The prototype is set to be turned into commercial software, and Omisa expects radiologists will be able to try out the program this summer. Omisa's goal is to have the software ready for the 2009 RSNA conference in Chicago, reported Iran Times.
"Presently, all available software solutions in medical image analysis deliver results that the medical expert--radiologist, oncologist, surgeon--usually does not find accurate with respect to the medical knowledge. Hence, the medical experts frequently modify the results--mainly the outline of the lesion or tumor. However, the existing software solutions do not learn anything from these user modifications and continue to deliver the same result over and over again if the same image or a similar one is processed," Tizhoosh said. "Our software learns over time how the radiologist changes the results in order to provide more accurate tumor marking, which results in the most striking feature of this software."
"In addition, since Segasist stores the knowledge of individual radiologists, it can provide a second opinion to any radiologist if the software is being used within network of radiologists-e.g. a hospital or, even better, a network of hospitals," he added. "This second opinion is automatically created based on the stored knowledge of all radiologists and can be provided instantly. This will reduce errors and save costs."