Recently, Radiology, a top international journal in the field of radiology, published the latest research results of the joint team of Huawei Cloud EI Innovation Incubation Lab, School of Telecommunications of Huazhong University of Science and Technology, and Department of Radiology of Union Hospital of Tongji Medical College of Huazhong University of Science and Technology: AI algorithm detects aneurysm with a sensitivity of 97.5%, helping doctors to improve the sensitivity of clinical diagnosis by about 10 percentage points and reduce the leakage rate by 5 percentage points. The rate of missed diagnoses was reduced by 5 percentage points.
Radiology releases joint medical research results from Huawei Cloud, Huazhong University of Science and Technology
Cerebral aneurysms are weakened areas of blood vessels in the brain and rank in the top 3 causes of cerebrovascular disease, with the risk of leakage or rupture, which can sometimes be fatal. The risk of aneurysm rupture depends on the size, shape and location of the aneurysm, so detection and characterization of cerebral aneurysms are key to guiding treatment.
CT angiography imaging (CTA) is currently the main imaging tool for evaluating intracranial aneurysms, but due to the small size of cerebral aneurysms and the complexity of intracranial vessels, it takes a long time for even professional radiologists to make a diagnosis, and some small aneurysms may be missed.
Huawei Cloud EI Innovation Incubation Lab, together with the School of Telecommunications of Huazhong University of Science and Technology and the Department of Radiology of Union Hospital of Tongji Medical College of Huazhong University of Science and Technology (Wuhan Union Hospital), has developed a fully automated and highly sensitive algorithm for detecting cerebral aneurysms based on CTA images using ModelArts, a one-stop AI development platform. The algorithm output will give information on the probability of aneurysm existence, aneurysm location and diameter size, and outline suspicious aneurysms on the original CTA images.
The joint Huawei Cloud team trained the model based on ModelArts, a one-stop AI development platform, and tested it with a dataset of 534 CT angiograms, which included 649 aneurysms.
From the data, the algorithm detected 633 of the 649 brain aneurysms, with a sensitivity of 97.5%. The study also identified eight new aneurysms that were overlooked in the initial clinical evaluation. Six of these eight aneurysms were less than 3 mm in diameter and two were between 3 and 5 mm, indicating that the algorithm also has very good performance for tiny aneurysms.
Dr. Xi Long, a radiologist at Wuhan Union Hospital who participated in the joint project, said, “The deep learning algorithm has shown excellent performance in detecting aneurysms. We found that very few aneurysms that were overlooked in the initial clinical diagnosis reports were successfully identified by the deep learning algorithm.”
Meanwhile, validation results on an additional 400-case external dataset showed that radiologists’ performance with algorithmic assistance improved in both diagnostic efficiency and diagnostic accuracy, with the most significant improvement especially for those with less experience. With AI assistance, radiologists’ clinical diagnostic sensitivity for brain aneurysms improved by about 10 percentage points and the rate of missed diagnoses decreased by 5 percentage points.
In recent years, Huawei Cloud EI Innovation Incubation Lab has focused its investment on solving major technical challenges in the medical field, and relevant papers have been included in the top journal of the Medical Society, and have won industry-leading levels in several authoritative challenge events such as LUNA-2016, HC-2018, ISLES-2018 , MICCAI2019, MICCAI2020, etc. The research results involve cervical cancer screening, stroke segmentation, ventricular segmentation, automatic generation of plain film diagnostic reports, new coronary pneumonia screening, aneurysm detection, etc.