Brain aneurysms, a type of cerebrovascular disease, have an incidence rate of 1.8-3.2%. Due to their small size and lack of obvious symptoms before rupture, they are often referred to as "time bombs in the brain." Statistics show that globally, a brain aneurysm ruptures approximately every eight minutes. Additionally, 80-90% of spontaneous subarachnoid hemorrhages are caused by ruptured brain aneurysms. The mortality rate for a ruptured brain aneurysm is between 23-51%, and 30-40% of survivors suffer permanent disabilities.

AI Aids in Cerebrovascular Imaging Analysis! Shuang-Ho Hospital Develops AI-Assisted Brain Aneurysm Diagnosis Platform

Magnetic resonance angiography (MRA) is a non-invasive, radiation-free tool for the early detection of brain aneurysms without the need for contrast agents. However, due to the small size of aneurysms—sometimes only about 2 millimeters—and the complexity of the cerebrovascular structure, interpreting these images is time-consuming and requires years of specialized radiology training to ensure accurate readings. Even with such expertise, small aneurysms can still be overlooked. To assist clinicians in overcoming the challenges of early diagnosis and treatment of brain aneurysms, Dr. Yen-Ting Chen and his team at Shuang-Ho Hospital developed a brain aneurysm AI-assisted diagnostic platform. This platform is built on an MRI brain aneurysm imaging database annotated by multiple radiologists. Using deep learning technology, the platform automatically detects and segments brain aneurysms, providing crucial information on their location, size, and shape, which enhances diagnostic efficiency and accuracy.

Massive Data and High-Performance Models Lead to a Detection Sensitivity of 96.7%

The platform boasts two key technical advantages:

Extensive Image Database: The platform's brain aneurysm image database comprises 1,400 cases with lesions of varying sizes and locations, including data from different stages of the disease. This rich and diverse dataset significantly enhances the AI model's generalization capability.

High-Performance AI Model: The platform employs a two-stage deep learning model. In the first stage, it quickly screens high-risk images, and in the second stage, it accurately segments the lesions. Additionally, the platform uses a semi-supervised learning strategy to further improve model performance.

In clinical tests conducted at Shuang-Ho Hospital, the platform achieved an average detection sensitivity of 84.7%, with a sensitivity of 80% for lesions smaller than 3 millimeters. These results are comparable to the sensitivity of clinical radiologists (84.2%) in the same test set. Furthermore, the research team found that the AI model complements the radiologists' diagnoses. When using the platform, the sensitivity of doctors increased to 96.7%, reducing the misdiagnosis rate by 12.5% compared to doctors who did not use the AI platform.

Dr. Yen-Ting Chen: AI Technology Can Break Through the Challenges of Brain Aneurysm Diagnosis

Dr. Yen-Ting Chen explained that brain aneurysms are extremely dangerous, with high rates of mortality and disability upon rupture. However, because the lesions are small and difficult to detect, early diagnosis is challenging. Traditional diagnosis relies heavily on radiologists' interpretation, but human factors limit both the sensitivity and specificity of these diagnoses. AI models, on the other hand, can automatically analyze large amounts of imaging data and learn interpretation patterns, improving diagnostic accuracy. Dr. Chen's team will continue to optimize the platform and promote its application in other medical institutions, hoping to assist more patients in receiving early diagnoses and better treatment outcomes for brain aneurysms.

Resource (mandarin): AI軍師幫偵測 助拆腦中不定時炸彈！