User login
SAN DIEGO – Artificial intelligence is improving the accuracy of optical biopsies, a development that may ultimately avoid the need for tissue biopsies of many low-risk colonic polyps, Michael Byrne, MD, said at the annual Digestive Disease Week.
Dr. Byrne, chief executive officer of Satisfai Health, founder of ai4gi, and gastroenterologist at Vancouver General Hospital; Nicolas Guizard, medical imaging researcher at Imagia; and their colleagues at ai4gi developed a “full clinical workflow” for detecting colonic polyps and performing optical biopsies of the polyps.”
Using narrow band imaging (NBI) enhanced with artificial intelligence, the system was used to review 21,804 colonoscopy frames and it achieved a “near-perfect” diagnostic accuracy of 99.9%. In an assessment of colonoscopy videos that included 125 polyps, the system had 95.9% sensitivity, with a specificity of 91.6% and a negative predictive value of 93.6%, Dr. Byrne said.
The speed of the system’s decision-making is rapid, with a typical reaction time of 360 milliseconds. The system was able to make diagnostic inferences at a rate of 26 milliseconds per frame.
With exposure to more learning experiences, the artificial intelligence system improved and committed to a prediction for 97.6% of the polyps it visualized. Dr. Byrne said this result represented a 12.8% improvement from previously published data on the model’s performance.
Dr. Byrne and his colleagues found the system had a tracking accuracy of 92.8%, meaning that this percentage of polyps was both correctly detected and assigned to a unique identifier for follow-up of the site of each excised polyp over time. The interface worked even when multiple polyps were seen on the same screen.
In a video interview, Dr. Byrne discussed the implications for gastroenterology and plans for a clinical trial for rigorous testing of the model.
ai4gi is developing the AI colonoscopy technology. Dr. Byrne is founder of the ai4gi joint venture, which holds a technology codevelopment agreement with Olympus US.
SAN DIEGO – Artificial intelligence is improving the accuracy of optical biopsies, a development that may ultimately avoid the need for tissue biopsies of many low-risk colonic polyps, Michael Byrne, MD, said at the annual Digestive Disease Week.
Dr. Byrne, chief executive officer of Satisfai Health, founder of ai4gi, and gastroenterologist at Vancouver General Hospital; Nicolas Guizard, medical imaging researcher at Imagia; and their colleagues at ai4gi developed a “full clinical workflow” for detecting colonic polyps and performing optical biopsies of the polyps.”
Using narrow band imaging (NBI) enhanced with artificial intelligence, the system was used to review 21,804 colonoscopy frames and it achieved a “near-perfect” diagnostic accuracy of 99.9%. In an assessment of colonoscopy videos that included 125 polyps, the system had 95.9% sensitivity, with a specificity of 91.6% and a negative predictive value of 93.6%, Dr. Byrne said.
The speed of the system’s decision-making is rapid, with a typical reaction time of 360 milliseconds. The system was able to make diagnostic inferences at a rate of 26 milliseconds per frame.
With exposure to more learning experiences, the artificial intelligence system improved and committed to a prediction for 97.6% of the polyps it visualized. Dr. Byrne said this result represented a 12.8% improvement from previously published data on the model’s performance.
Dr. Byrne and his colleagues found the system had a tracking accuracy of 92.8%, meaning that this percentage of polyps was both correctly detected and assigned to a unique identifier for follow-up of the site of each excised polyp over time. The interface worked even when multiple polyps were seen on the same screen.
In a video interview, Dr. Byrne discussed the implications for gastroenterology and plans for a clinical trial for rigorous testing of the model.
ai4gi is developing the AI colonoscopy technology. Dr. Byrne is founder of the ai4gi joint venture, which holds a technology codevelopment agreement with Olympus US.
SAN DIEGO – Artificial intelligence is improving the accuracy of optical biopsies, a development that may ultimately avoid the need for tissue biopsies of many low-risk colonic polyps, Michael Byrne, MD, said at the annual Digestive Disease Week.
Dr. Byrne, chief executive officer of Satisfai Health, founder of ai4gi, and gastroenterologist at Vancouver General Hospital; Nicolas Guizard, medical imaging researcher at Imagia; and their colleagues at ai4gi developed a “full clinical workflow” for detecting colonic polyps and performing optical biopsies of the polyps.”
Using narrow band imaging (NBI) enhanced with artificial intelligence, the system was used to review 21,804 colonoscopy frames and it achieved a “near-perfect” diagnostic accuracy of 99.9%. In an assessment of colonoscopy videos that included 125 polyps, the system had 95.9% sensitivity, with a specificity of 91.6% and a negative predictive value of 93.6%, Dr. Byrne said.
The speed of the system’s decision-making is rapid, with a typical reaction time of 360 milliseconds. The system was able to make diagnostic inferences at a rate of 26 milliseconds per frame.
With exposure to more learning experiences, the artificial intelligence system improved and committed to a prediction for 97.6% of the polyps it visualized. Dr. Byrne said this result represented a 12.8% improvement from previously published data on the model’s performance.
Dr. Byrne and his colleagues found the system had a tracking accuracy of 92.8%, meaning that this percentage of polyps was both correctly detected and assigned to a unique identifier for follow-up of the site of each excised polyp over time. The interface worked even when multiple polyps were seen on the same screen.
In a video interview, Dr. Byrne discussed the implications for gastroenterology and plans for a clinical trial for rigorous testing of the model.
ai4gi is developing the AI colonoscopy technology. Dr. Byrne is founder of the ai4gi joint venture, which holds a technology codevelopment agreement with Olympus US.
REPORTING FROM DDW 2019
