Smadar Shiffman

Medical image segmentation using analysis of isolable-contour maps

A common challenge for automated segmentation techniques is differentiation between images of close objects that have similar intensities, whose boundaries are often blurred due to partial-volume effects. The authors propose a novel approach to segmentation of two-dimensional images, which addresses this challenge. Their method, which they call intrinsic shape for segmentation (ISeg), analyzes isolabel-contour maps to identify coherent regions that correspond to major objects. ISeg generates an isolabel-contour map for an image by multilevel thresholding with a fine partition of the intensity range. ISeg detects object boundaries by comparing the shape of neighboring isolabel contours from the map. ISeg requires only little effort from users; it does not require construction of shape models of target objects. In a formal validation with computed-tomography angiography data, the authors showed that ISeg was more robust than conventional thresholding, and that ISegs results were comparable to results of manual tracing.

Building a speech interface to a medical diagnostic system

A description is given of the design of an interface for QMR-DT, an evolving decision-theoretic version of Quick Medical Reference, which performs diagnostic reasoning about diseases in internal medicine. QMR-DT encompasses the part of QMRs functionality that provides differential diagnoses for a set of patient characteristics, but it uses a different algorithm to compute diagnoses. The work described includes two programs that integrate off-the-shelf speech technology with programs that manipulate medical terminology. The term identifier program uses an isolated-word, speaker-dependent speech product to provide an interface for entering medical findings into QMR-DT. Frame browser is an auxiliary program used by the developers of term identifier to examine frame structures. Frame browser was also used to experiment with a continuous-speech system.<<ETX>>

BrainImageJ: a Java-based framework for interoperability in neuroscience, with specific application to neuroimaging.

The Human Brain Project consortium continues to struggle with effective sharing of tools. To facilitate reuse of its tools, the Stanford Psychiatry Neuroimaging Laboratory (SPNL) has developed BrainImageJ, a new software framework in Java. The framework consists of two components-a set of four programming interfaces and an application front end. The four interfaces define extension pathways for new data models, file loaders and savers, algorithms, and visualization tools. Any Java class that implements one of these interfaces qualifies as a BrainImageJ plug-in-a self-contained tool. After automatically detecting and incorporating new plug-ins, the application front end transparently generates graphical user interfaces that provide access to plug-in functionality. New plug-ins interoperate with existing ones immediately through the front end. BrainImageJ is used at the Stanford Psychiatry Neuroimaging Laboratory to develop image-analysis algorithms and three-dimensional visualization tools. It is the goal of our development group that, once the framework is placed in the public domain, it will serve as an interlaboratory platform for designing, distributing, and using interoperable tools.

Semiautomated editing of computed tomography sections for visualization of vasculature

The goal of our work is to help radiologists remove obscuring structures from a large volume of computed tomography angiography (CTA) images by editing a small number of sections prior to three-dimensional (3D) reconstruction. We combine automated segmentation of the entire volume with manual editing of a small number of sections. The segmentation process uses a neural network to learn thresholds for multilevel thresholding and a constraint- satisfaction neural network to smooth the boundaries of labeled segments. Following segmentation, the user edits a small number of images by pointing and clicking, and then a connectivity procedure automatically selects corresponding segments from other sections by comparing adjacent voxels within and across sections for label identity. Our results suggest that automated segmentation followed by minimal manual editing is a promising approach to editing of CTA sequences. However, prerequisites to clinical utility are evaluation of segmentation accuracy and development of methods for resolution of label ambiguity.

Interactive specification of regions of interest on brain surfaces

We describe Surface Editor-a tool for interactive specification of regions of interest (ROIs) on brain surfaces. The tool allows users to define subsurfaces by tracing around areas within a triangle-mesh brain surface. The input to the program is a triangle-mesh representation of a brain volume and a set of user-defined input points on the mesh. The program connects each pair of successive input points with a polyline that results from the intersection of the mesh with a plane that is approximately normal to the mesh. The polyline comprises coplanar line segments. The boundary of an ROI is a connected set of polylines that intersects triangle edges to form a continuous path. To validate Surface Editor we demonstrated that the program could be used to interactively delineate gyri on brain surfaces, and we showed that paths that the program generated were comparable to paths that a user generated and to shortest paths.

Semiautomated segmentation of blood vessels using ellipse-overlap criteria: Method and comparison to manual editing

Two-dimensional intensity-based methods for the segmentation of blood vessels from computed-tomography-angiography data often result in spurious segments that originate from other objects whose intensity distributions overlap with those of the vessels. When segmented images include spurious segments, additional methods are required to select segments that belong to the target vessels. We describe a method that allows experts to select vessel segments from sequences of segmented images with little effort. Our method uses ellipse-overlap criteria to differentiate between segments that belong to different objects and are separated in plane but are connected in the through-plane direction. To validate our method, we used it to extract vessel regions from volumes that were segmented via analysis of isolabel-contour maps, and showed that the difference between the results of our method and manually-edited results was within inter-expert variability. Although the total editing duration for our method, which included user-interaction and computer processing, exceeded that of manual editing, the extent of user interaction required for our method was about a fifth of that required for manual editing.