Medical Image Analysis SBIA UPenn Christos Davatzikos Director
Medical Image Analysis @ SBIA. UPenn Christos Davatzikos Director, Section of Biomedical Image Analysis Professor, Radiology, ESE, BE University of Pennsylvania http: //www. rad. upenn. edu/sbia
http: //www. rad. upenn. edu/sbia
Human brain imaging: • AD/aging, • Schizophrenia • Treatment of Diabetes • Effects of hormonal and Techniquesenvironmental exposures to the brain • Effects of alcohol Clinical studies Segmentation and labeling Spatial normalization/ registration Morphological analysis Hi-D Pattern classification as diagnostic tool Image statistics and datamining Biomechanical Biophysical modeling Animal imaging: Treatment planning: • Brain tumors • Optimized prostate biopsy • DTI of the developing mouse brain • Hormonal effects on Monkey brain development Breast MRI Cardiac image analysis, FMRI postprocessing Facial expression analysis for affective disorders
Segmentation and Labeling Problem definition Automated segmentation of anatomical structures from the images of different modalities, for quantitative comparison and shape analysis. Motivation • Volumetric measurement of anatomical structures. • Longitudinal comparison, revealing disease-related changes. • Group comparison of healthy controls and patients. Sulcal labeling Examples Hippocampal segmentation Prostate segmentation Spine segmentation
Spatial normalization and registration Problem definition Image data co-registration and integration, from different individuals, modalities, times, and conditions, for comparative image analysis such as characterization of structural, functional variability and abnormality. Motivation • Group comparison, by computing inter-individual variability. • Longitudinal comparison, for measuring diseases, i. e. AD. • Longitudinal follow-up, with/without treatment. • Different modalities integration, for conveying complementary information. Warping of subject to model
Measuring volumes of anatomical structures : An atlas with anatomical definitions is registered to the patient’s images Model HAMMER Subject
Distribution of HAMMER (6/21/2008)
4 -Dimensional Image Analysis for Measurement of Temporal Changes Original Images 3 -D Segmentation CLASSIC 4 -D Segmentation Year 1 Year 2 Year 3 Year 4
Multi-parametric Pattern Classification Pattern Abnormality Classification score
AD vs CN classifier applied to MCI: most MCI’s have AD-like MRI profiles Subsequent cognitive decline Data from ADNI
Segmentation of Brain Lesions FLAIR threshold = -1 Manual definition threshold = 0 threshold = 2
Diffusion Tensor Imaging: Manifold-based Statistical Analysis B 0 image and at least 6 gradient directions/slice Reconstructed DTI Geodesic-based and Kernelbased Statistics Developing mouse brain : old Vs young Multi-parametric Information Schizophrenia patients Vs Controls
Thank you
- Slides: 14