Recent talks by members of MIDAG

Statistics on Anatomic Geometry (MICCAI 2005 tutorial)


Current Research Areas

Anatomic Geometry & Deformations and Their Population Statistics (Steve Pizer, 2004)

Multi-object representations for classes of 3D anatomical structures
(Tutorial given at SPIE Medical Imaging conference, February 2003)

Presentations from Statistics of Geometry Tutorials (October 2002)

Object Representation and Segmentation

Statistical Morphology in Psychiatric Illness

The in vivo measurement of the size, shape and temporal shape change of biological structure is becoming increasingly important to study neurodevelopmental and degenerative disorders. We study and develop general methedology for statistical morphology. Recent studies hypothesize that the application of accurate and reliable methods of shape analysis will provide a stronger discrimination measure of brain pathomorphology than current quantitative volumetric methodologies
Shape Representation via Spherical Harmonics: hippocampus (schizophrenia project), lateral ventricle (mono/di-zygotic twins).
Combined Boundary-Medial Shape Representation: methods, shape analysis results.

Computer Aided Diagnosis and Display in Radiology

Vessels and Tumors in Neurosurgery

Radiation Oncology Computing Research

Methodology for Validation of Segmentation


Additional Projects
Image Analysis and Display Technologies 

3D/2D Registration of Curves: Application to Endovascular Intervention

The treatment of vascular disease involving the central nervous system uses state-of-the-art minimally invasive technologies. Using tiny flexible catheters used from within blood vessels, most vessels in the brain and the spinal cord can be accessed to treat disabling diseases without performing open surgery. This project develops a methodology to register a 3D model of the brain vascular tree, obtained with pre-operative image analysis, with 2D angiographic images acquired during the intervention. 

Correction of intensity bias field in Magnetic Resonance Imaging

Image intensity variations induced by radiofrequency coil inhomogeneities but also by the patient severely impede image segmentation by statistical pattern recognition. The new method calculates a parametric estimate of the bias field using prior information about tissue probability density functions.

Interactive Segmentation of 3D Volume Data(IRIS)

Object Based Segmentation, Registration and Shape Measurment

The research in MIDAG focuses on the development of methods of image analysis visualization, and image processing of medical images and the production of approaches to solving clinical problems that benefit from these methods. In image analysis our methods solve problems of segmentation, registration, shape measurement, and computer aided diagnosis and involve both deformable structural models approaches and statistical pattern recognition approaches. 

MIND, DICOM server image searching and transfer 
NLM Insight Image Processing Library (parts included in the CADDLab project) 

Clinical Objectives 
3D Ultrasound Imagine of the Neonatal Brain 

This project aims at developing improved image processing methodology for 3D Ultrasound image data to study lateral cerebral ventricles in the neonatal brain. Mild enlargement of the fetal lateral ventricles is hypothesized as a marker of increased risk for poor neurodevelopmental outcome. 


Many of the imaging projects linked from here include Biostatistics collaborators. 
Breast Cancer Detection and Diagnosis 
  • Diffraction Enhanced Images using Monoenergetic Xrays 
  • Digital Mammography 
  • Softcopy Display 
  • Contrast Enhancement Research 

Structural Segmentation and Shape Characterization of Brain Lesions 

This project develops a new structural segmentation technique for blobby 3D structures. Segmentation is driven by object shape rather than image intensity and results in a full shape characterization of the individual objects which helps answering clinical hypothesis about lesion distribution in regard to anatomical and functional brain regions. 


TIPS Procedure 

  • Tubes 
  • 2D/3D Segmentation via Tubes 
Tumor/AVM Segmentation 
Methods for the Validation of Segmentation
 Valmet: A new validation tool for assessing and improving 3D object segmentation

Last updated 05/22/2002
Please e-mail the webmaster for any issues concerning this page.