การตรวจจับและระบุตำแหน่งของจุดเลือดออกในสมองจากภาพถ่ายคลื่นแม่เหล็กไฟฟ้าโดยใช้เครือข่ายแบบสังวัฒนาการ

Abstract

Cerebral micro-bleeding (CMB) is the condition where small vessels in the brain suffer small chronic hemorrhages. The main cause of CMB is hypertension. CMB does not have any symptoms in the short term but could cause the Alzheimer's disease in the long term. CMB patients generally have difficulties with daily activities such as speaking, walking, and remembering. Currently, CMB is manually diagnosed by the doctor, and there are many limitations (using time and not accurate, up to the doctor experience). The performance and reliability are importance. This thesis presents an automatic micro-bleeding detection and location from susceptibility weighted imaging (SWI) brain scan framework. The data used with respect to this research was collected from 26 patients that obtained from the local hospital in Phuket. A set of images from 16 patients was used as a training set and a set of images from 10 patients was used as test set. The process started with the training set images were segmented thus the region of interests (ROIs) were identified. Once a collection of ROIs has been obtained. The classifier generation process was then commenced. With respect to the work presented in this thesis, there were 4 architectures of Convolutional neural network were carried out: (i) Alexnet (ii) Googlenet (iii) Resnet50 and (iv) Squeezenet. The experiments also evaluated the classification performance using the different number of samples. There were five experiments had been conducted: 1,000 2,000 3,000 4,000 and 5000 samples. There were 4 areas of the brain were extracted using shape matching technique. The areas include: (i) cortical (ii) subcortical (iii) cerebellum and (iv) brainstem. There was 95.76% accuracy was recorded for CMB detection while 100% accurate was recorded for area (location) identification on the test set.