JSAI2020

Presentation information

General Session

General Session » J-13 AI application

[2E1-GS-13] AI application: Medical application (2)

Wed. Jun 10, 2020 9:00 AM - 10:40 AM Room E (jsai2020online-5)

座長:水本智也(フューチャー株式会社)

10:20 AM - 10:40 AM

[2E1-GS-13-05] False Positive Reduction Using Vascular Structure and Location Information for Cerebral Aneurysm Detection Model on MR Angiography

〇Kohei Tashiro1, Yuki Terasaki2, Hajime Yokota3, Joji Ota4, Takuro Horikoshi5, Yasukuni Mori6, Hiroki Suyari6 (1. Chiba University Faculty of Engineering, 2. Chiba University Graduate School of Science and Engineering, 3. Chiba University Graduate School of Medicine, 4. Chiba University Hospital Department of Radiology, 5. Chiba University Hospital Department of Radiology , 6. Chiba University Graduate School of Engineering)

Keywords:Cerebral aneurysms, MRA, False positive reduction, Convolutional Neural Network

Rupture of a cerebral aneurysm is a major cause of subarachnoid hemorrhage. Therefore, early detection and medical treatments of aneurysm are crucial. MRA images are widely used to diagnose cerebral aneurysms, and several methods using CNN have been proposed to detect cerebral aneurysms on MRA images automatically. In particular, the method using Multi-Modal CNN, which combines 3DMRA and 2D images through MIP (Maximum Intensity Projection) method, can achieve high sensitivity. However, this method does not utilize the location and geometrical information of lesion candidates in the brain, resulting in many false detections as a drawback. In this study, after reducing the number of false positives by using the location and vascular structure of lesion candidates in preprocessing steps, we put them into Multi-Modal CNN as inputs. As experiments result, we achieved a lower false detection rate with the same detection sensitivity when compared to the previous research.

Authentication for paper PDF access

A password is required to view paper PDFs. If you are a registered participant, please log on the site from Participant Log In.
You could view the PDF with entering the PDF viewing password bellow.

Password