Optimized pathological and visual content-based neuroimaging retrieval

Neuroimaging provides important insights for understanding neurobiology and is essential for accurate neurological and neurosurgical diagnosis and patient care. The volume and complexity of the neuroimaging datasets have greatly increased due to advances in scanning instrumentation. These large datasets now pose challenges for images retrieval / management and more effective approaches are needed. Content-based image retrieval (CBIR) takes advantage of the rich visual/physiological information in the images and can provide the opportunity for more efficient and reliable image retrieval. Although a number of investigators have used CBIR systems in neuroimaging, few of these approaches have explored all the potential features in these images. We suggest that such image retrieval could be optimized by using pathological and domain-specific visual features rather than texture features alone. We used the cerebral metabolic rate of glucose (CMRGlc) as the physiological parameter from static brain [ F ] 2-fluorodeoxy-glucose (FDG) positron emission tomography (PET) images and a customized disorder-oriented mask (DOM) specific for a particular neurodegenerative disorder, with the regions of interest (ROIs) specific for each disease sub-type. We designed 8 Gabor filter banks with different parameter settings and identified the optimum Gabor function parameter setting for the visual feature extraction. Our experimental data indicate that optimization of the Gabor filter parameters, targeted to disease specific regions enhances retrieval precision.