Identification of human cortical structures in Magnetic Resonance Imaging by encapsulating expert knowledge in fuzzy logic /
Magnetic Resonance Imaging (MRI) is an important non-invasive technique to examine the human brain. The identification of different brain components and structures is a necessary precursor to conduct both qualitative and quantitative measurements. Because of the extreme complexity and variability...
| Main Author: | |
|---|---|
| Format: | Thesis Book |
| Language: | English |
| Published: |
[Place of publication not identified] :
[publisher not identified] ;
1996.
|
| Subjects: | |
| Online Access: | http://proxy.library.tamu.edu/login?url=http://proquest.umi.com/pqdweb?did=739668381&sid=1&Fmt=2&clientId=2945&RQT=309&VName=PQD |
| Summary: | Magnetic Resonance Imaging (MRI) is an important non-invasive technique to examine the human brain. The identification of different brain components and structures is a necessary precursor to conduct both qualitative and quantitative measurements. Because of the extreme complexity and variability of human cortical structures from person to person, knowledge of anatomical and neurological structure is needed to analyze the acquired brain images. To handle the uncertainty due to the complexity and variability of human cortical structures, fuzzy logic and rules are applied to encapsulate the expert anatomical knowledge into a computer system. Three major techniques have been developed in this research. The first technique is a fuzzy knowledge-based system that encapsulates expert knowledge to identify the human cortical structures. Because the major cortical structures, such as sulci and gyri, are located on the brain surface, a two-dimensional brain surface map has been constructed. In this research, this system successfully identified the central sulci and lateral sulci. Then, according to the position of these sulci, the system can label the location of the frontal lobe. The second technique is a hybrid fuzzy image segmentation system that is used to conduct both qualitative and quantitative measurements on the brain MRI. This system combines the advantages of both supervised learning system, which is a fuzzy knowledge-based system, and unsupervised learning method, which is a fuzzy classification algorithm. A self-adaptive membership justification method has been invented to automatically identify the parameters of the fuzzy rules. Both normal brain image classification and lesion detection have been conducted and confirmed by medical experts. The last technique is a new multi-prototype fuzzy c-means (MFCM) algorithm that extends the fuzzy c-means (FCM) algorithm by allowing each class to have more than one prototype. The original FCM algorithm has been widely used in medical image segmentation. We have evaluated the MFCM algorithm using Fisher's IRIS data set. The new algorithm achieves a better performance than FCM in terms of both accuracy and speed. In these new techniques, fuzzy logic has been widely used. The major a vantage of using fuzzy logic is that the fuzzy rules encapsulate the expert knowledge in linguistic form. This research not only demonstrates a promising alternative approach to the human cortical structure identification, but also introduces several new techniques for both medical image segmentation and fuzzy classification. Most of these new techniques are sufficiently general to be applied to other problem areas. |
|---|---|
| Item Description: | Vita. "Major Subject: Computer Science". |
| Physical Description: | xiii, 124 leaves : illustrations ; 28 cm. Issued also on microfiche from University Microfilms Inc. |
| Bibliography: | Includes bibliographical references: pages 110-123. |