University of Patras
Computer assisted diagnosis of brain tumors based on statistical methods and pattern recognition techniques
Abstract
dc:descriptionThe process of brain tumor characterization requires a rather intricate assessment of the various Magnetic Resonance (MR) image and spectroscopic features and is typically performed by experienced radiologists. Despite the inherently subjective nature of many of the decisions associated with this process, an expert radiologist is able to perform this task with a significant degree of precision and accuracy. However, in the effort to deliver more effective treatment, clinicians are continuously seeking for greater accuracy in the pathological characterization of brain tissues. The aim of the present thesis was to design, implement, and evaluate a software classification system for discriminating between different brain tumor types on Magnetic Resonance Imaging (MRI), employing textural and spectroscopic features. The clinical material consisted of sixty seven T1-weighted post-contrast MR brain images (21 metastases, 19 meningiomas, and 27 gliomas), obtained from patients with verified and untreated intracranial tumors. Thirty-six 2-dimensional textural features (2D), from the image histogram and the co-occurrence and run-length matrices, were extracted from each one of 67 MR-images. Similarly, an equal number of 3-dimensional textural features (3D) were also calculated in the attempt to maximize classification performances. Finally, MR-spectroscopy features were also incorporated for improving classification accuracies. Classification methods employed included i/ a modified Probabilistic Neural Network (PNN) and Support Vector Machines (SVM) algorithms, incorporating a non-linear Least Squares Features Transformation (LSFT) into the classifiers and ii/ an ensemble classification scheme employing the LSFT-SVM classifier. The LSFT improved classifiers’ performances, increased class separability, and resulted in dimensionality reduction. For evaluating the performance of the designed classification schemes, evaluations were performed by means of the external cross validation process, which is considered indicative of the generalization performance of the designed classification system to ‘unseen’ cases. It was found that the LSFT features transformation enhanced the performance of the PNN and SVM algorithms, achieving classification accuracies of 73.48 % in distinguishing metastatic from primary tumors and 88.67% in discriminating gliomas from meningiomas. When volumetric 3-dimensional features were employed, these results improved to 88.18% for discriminating between metastatic and primary tumors and 97.33% for distinguishing gliomas from meningiomas. The textural features employed in the design of the optimum classification scheme were associated primarily with image texture homogeneity. Finally, when MR-spectroscopy features were also incorporated, classification accuracy was boosted up from 95% in discriminating meningiomas from metastasis to 100%. The MR-image features that participated in the optimum feature vector were related to the degree of homogeneity, the amount of randomness and the dispersion of the gray-tone intensity values within the texture of the tumor. These textural characteristics are related to textural parameters that physicians employ in diagnosis and they were proportional to the textural imprint of brain tumors, i.e. gliomas have heterogeneous texture while meningiomas appear to be homogeneous in MR imaging. The MR-spectroscopy feature that participated in the optimum feature vector was the Choline (Cho)/N-Acetyl Aspartate (NAA) metabolite integral ratio. It was found that both meningiomas and metastases are characterized from low concentrations of NAA while meningiomas exhibit higher concentrations of Cho than metastases, which could be attributed to increased synthesis of tumor cell membranes. Finally, the proposed system might be of value as an assisting tool for brain tumor characterization on volumetric MRI series.
Degree
thesis:*- Grantor dc:publisher
- University of Patras
- Year dc:date
- 2010
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Georgiadis, Pantelis
- Γεωργιάδης, Παντελής
Subjects
dc:subject × 14- Όγκοι εγκεφάλου
- Πυρηνικός μαγνητικός συντονισμός
- Αναγνώριση πρωτύπων
- Brain tumors
- Nuclear magnetic resonance (NMR)
- Pattern recognition
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Ιατρική και Επιστήμες Υγείας
- Βασική Ιατρική
- Natural Sciences
- Computer and Information Sciences
- Medical and Health Sciences
- Basic Medicine
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/27954