Artificial intelligence (AI) can automatically detect cerebral hemorrhages with a high degree of accuracy, according to new research.
Patients with intracranial cerebral hemorrhage are commonly treated in emergency departments. Detecting intracranial hemorrhage is an essential and often urgent job for radiologists, as the bleeding can be life-threatening.
Bleeding can stem from a number of causes, including trauma, aneurysms or hemorrhagic stroke.
“The challenge for the radiologist is that many CT scans coming from the emergency department will not have a significant finding,” said study co-author Jeremy J. Heit, MD, PhD, assistant professor of radiology at Stanford University School of Medicine. “But then there are cases you want to find quickly because the patient could be in danger and a radiologist might be the first one to spot the problem.”
Automated detection through AI-based software has the potential to expedite patient care, improve worklist triage and increase diagnostic confidence, research has shown.
In their 2020 research, Dr. Heit and colleagues tested one such system, RAPID ICH™, which identifies acute intracranial hemorrhage and quantifies the volume of the bleeding on non-contrast head CT.
Strong Agreement Between AI Software and Experts
In the study, Dr. Heit and colleagues evaluated RAPID ICH on a set of 308 head CT images from 158 patients with intracranial hemorrhage and 150 control patients with no hemorrhage. The intracranial hemorrhage cases included subtypes such as hemorrhage over the surface of the brain, inside the brain tissue and in the fluid spaces around the brain.
Dr. Heit and two other expert neuroradiologists analyzed the images for the presence or absence of intracranial hemorrhage and compared results with the findings from RAPID ICH.
RAPID ICH detected 151 of the 158 cerebral hemorrhages for a sensitivity of 96%. The software confirmed a negative finding in 143 of 150 (95%) of the control cases.
“The primary results of the validation study demonstrate excellent agreement between RAPID software and the experts,” Dr. Heit said. “The processing time is quite rapid, with less than three minutes per case on average.”
In cases where RAPID ICH failed to detect hemorrhage, the amount of bleeding was considerably less than a millimeter and was often located in parts of the brain that are more difficult to see.
“The good news is that the amount of hemorrhage that was missed volume-wise was very, very small,” Dr. Heit said. “In general, such findings are not as life-threatening.”
RAPID ICH, part of the RapidAI™ platform of advanced imaging products by iSchemaView, is not intended to replace radiologists but to augment their work, Dr. Heit said.
“Eventually, all head CTs could go through this platform,” Dr. Heit said. “Positive findings would either get flagged or moved to the top of the radiologist worklist so the referring physician is quickly notified and emergency intervention can happen much more quickly.”
For More Information:
View the RSNA 2020 session, Automated Cerebral Hemorrhage Detection Using RAPID — SSNR02 at RSNA2020.RSNA.org.
