Brain lesion analysis using multi-model neural images

The mismatch between infarct core and ischemic penumbra is a crucial factor in decision making for stroke treatment, hence it is very important to have stroke penumbra detection and volume estimation for acute stroke patients. Recently, Computed Tomography Perfusion (CTP) is one of the imaging modal...

Full description

Saved in:
Bibliographic Details
Main Author: Zhou, Huan
Other Authors: Lin Zhiping
Format: Final Year Project
Language:English
Published: Nanyang Technological University 2018
Subjects:
Online Access:http://hdl.handle.net/10356/74922
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:The mismatch between infarct core and ischemic penumbra is a crucial factor in decision making for stroke treatment, hence it is very important to have stroke penumbra detection and volume estimation for acute stroke patients. Recently, Computed Tomography Perfusion (CTP) is one of the imaging modalities that emerging in acute stroke diagnosis. However, up to now no automatic segmentation method has been clinically employed in assisting radiologists to complete analysis. In this project, I aim to employ Isensee “3D U-net”, a deep-learning based 3D dense volumetric segmentation neural network, to achieve automatic penumbra detection/segmentation on CTP images and evaluate its performance. In total, 179 stroke patients’ medical CTP images were used as input dataset, it turns out that the experimental results are presentable but still has room to improve in order to reach the standard for clinical usage. For future work, a fusion of multiple deep neural networks is going to be applied in the study.