Loew Rsch Pres Jun27 Final by b8z9wwC6

VIEWS: 0 PAGES: 7

									Representative Recent Research in
        Medical Imaging

                      Murray Loew
            Biomedical Engineering Program
    Department of Electrical and Computer Engineering
                     June 27, 2007




                                                        1
    Multispectral Infrared Imaging of the Breast
                                        Thermal signatures of cancer can be
                                        visualized in the infrared (IR). Using several
                                        wavelengths (middle pair of images) provides
                                        additional information. Measures of
                                        asymmetry aid detection of anomalies.
                                        Independent component
                                        analysis allows the
                                        identification of “sources”
                                        within the breast that explain
                                        the thermal appearance and
                                        are likely to contribute to
                                        diagnosis. Studies will be
                                        conducted at GW’s Breast
                                                                              Each band of IR
                                        Clinic.
Top: initial IR image (left) and after 10 minutes’ thermal                    provides its own image;
equilibration. Middle: mid-wave (left) and long-wave IR images.               the resulting set is
Bottom: unsupervised classification of images; blue is low                    amenable to analysis by
probability of cancer; red is high.                                           numerous methods


       With R. Brem, M.D., GWU, F. Razjouyan, B.S. student, and I.
       Kopriva, Ph.D., visiting scientist                                                      2
Thermo-elastic modeling of breast tissue with gravity-induced deformation

 We can model the shape of the                   And can then model the thermal state
 breast under realistic conditions of            of the tissue – in 3-d – under the
 position and orientation                        influence of a source (e.g., a tumor) at
                                                 an arbitrary position

                This information is useful for diagnosis, prognosis, and therapy

                Strain                                          Temperature




                            With L. Jiang, doctoral student                                 3
                Applied Stereology: Tradeoffs in CT Lung Volume
                Measurement Using Subsets of Slices
          When a patient is examined at different times or with different protocols,
          how can we know whether the observed differences in a volume estimate
          are due to the patient, the protocol, or both? Specifically, we would like
          to know what is the smallest difference in lung volume that can be
          computed reliably from two sets of CT data, acquired by varying the
          number and thicknesses of the slices.
                             Our results show that thick-slice CT images
                Object
                             can be as effective as those with thin slices
                             in the estimation of lung volume. Patients in
First Section            Tm
                             remote areas often can thus avoid a trip to a
 Randomly
   Placed
                             regional CT center by using images taken on a             A set of CT image slices
        Cavalieri method for less-capable machine.                                      taken to estimate lung
        estimating volume                                                                              volume


                  With J. Reinhardt, Ph.D., U. of Iowa, and Z. Markowitz, doctoral student
                                                                                                        4
Finding Salient Features in Mammograms
                                        We have identified measures of salience that
                                        are effective at automatically identifying regions
                                        that draw the attention of mammographers.
                                        This has implications for computer-aided
                                        diagnosis and for image compression for
                                        transmission and storage.
                                        Ellipses on two subjects’ mammograms
                                        indicate fixations of human observers; dashed
                                        lines show ground truth location. Our
                                        technique produces salience maps (lower
                                        sets) for the CC and MLO views; the maps
                                        were successful in identifying the lesions and
                                        minimizing false positives.



With H. Kundel, M.D., U. of Pennsylvania, and P. Perconti, doctoral student

                                                                                      5
    Image Analysis: Insulin Granule ID and Tracking
                             Completely automatic methods have
                             been developed for identifying insulin
                             granules in β- cell photomicrographs.
                             This replaces the tedious and
                             inaccurate manual methods and makes
                             possible an estimation of the temporal
                             response function – movement of
A beta cell from the         granules toward the plasma membrane
pancreatic islets; insulin   – and also their number, location, and
granules are the black       morphology.
cores in the circles.
How many are touching                         A sample of a β-cell after processing. Red: single granule
the boundary of the                           core and halo; blue: compound granule; yellow: halo only.
cell?


         With G. Sharp, Ph.D. and S. Straub, Ph.D., Cornell U., and T. McClanahan,
         doctoral student                                                          6
          Other current work
• Optical coherence tomography (OCT) of the
  bladder (with J. Zara, Ph.D., ECE; M. Manyak,
  M.D., Urology; A. Lingley, doctoral student)

• Task-based measures of image quality for
  image compression (with D. Li, doctoral student)

• Diffusion-tensor imaging in MRI (with P. Basser,
  Ph.D., NIH; R. Freidlin, doctoral student)

                                                     7

								
To top