| Individual metabolic brain diseases (MD) may not be commonly encountered in medical practice, but as a group, metabolic brain diseases account for a considerable proportion of pediatric brain pathology. Automated diagnostic methodologies may facilitate earlier and more accurate prognosis. We present a novel Computer-assisted Medical Decision Support System ( mCAD) which extracts information available from Magnetic Resonance Spectroscopy (MRS) and Diffusion-weighted Magnetic Resonance Imaging (DWI) to assist neuroradiologists in diagnosing individual metabolic brain diseases. MRS is a technique which allows for the measurement of different metabolite concentrations MD may affect the normal metabolite concentrations and MRS can be used to identify metabolic changes. The diffusion of water molecules can also be investigated through the use of DWI technique. Certain MD affect water diffusion and are candidates for DWI study.The proposed system which utilized the combination of MRS, DWI and clinical information (the age factor) achieved a sensitivity (Se) and positive predictivity (PP) of 65% and 72%, respectively, in detecting seven categories of metabolic brain diseases.We have implemented a collection of signal and image processing routines where the Discrete Wavelet Transform (DWT) and the Fuzzy Relational Classifiers (FRC) are the major mathematical tools. Wavelet analysis has overcome some problems associated with other processing methods and is a potential aid. Due to the fractal behavior of spectroscopy signals and special characteristics of diffusion-weighted images, wavelet transform has been utilized in this study. We have used the DWT to investigate the data properties and to accommodate the feature extraction routines. When data features are extracted, the need for proper classification often exists. Many different types of classifiers have been reported, but selecting appropriate classifiers requires proper investigation. The FRC mimic human subjectivity in decision making and suits the conditions where the number of cases under examination is limited and other classical pattern recognition systems (e.g. Neural Networks) are not applicable. Limitations exist due to the nature of these diseases as a single institution is not able to acquire a large dataset or to possess sufficient sample cases within a single subcategory of MD. |
