Image Guided Decision Support (IGDS) System
[ Abstract] [Paper (.ps.gz)] [Demo Paper (.ps.gz)][Introduction] [Micro-Controller] [Decision Support]
[Architecture] [Future work]
|
|
Image Guided Decision Support (IGDS) System[ Abstract] [Paper (.ps.gz)] [Demo Paper (.ps.gz)][Introduction] [Micro-Controller] [Decision Support] [Architecture] [Future work] |
Demos
Complete demo (mpg, 28M, 3:20min). First part only (mpg, 12M, 1:29min).
Presentations
|
|
Introduction
| Recent advances in networking, visualization, robotics, and computer technology allow today real-time diagnosis, consultation, and education by using images obtained through remote microscopy. We present a new approach in telepathology, the Image Guided Decision Support (IGDS) system, which integrates components for both remote microscope control and decision support. Using the Micro-Controller component the physician can command a robotic microscope from a distance and obtain high-quality images to be used in diagnosis. The image understanding-based Decision Support component of the system locates, retrieves and displays cases which exhibit morphological profiles consistent with the case in question. The IGDS system has a natural man-machine interface containing engines for speech recognition and voice feedback. |
Multiuser Micro-Controller
The primary user can adjust the light path or focus of the microscope,
change the objective lens, move the specimen on the robotic stage, or
copy the current image to the Decision Support part for further
analysis. These actions are possible by graphical input or by
speech. A fusion agent capable of multimodal inputs interprets the
commands, calls the appropriate method, and gives voice
feedback. Currently the system employs a speech recognizer engine with
finite-state grammar and a restricted task-specific vocabulary. The
recognition is speaker-independent.
Examples of voice commands are:
Set Light ##, Set Focus ##, Change #, Transfer, Move Right (Left,
Up, Down), Update the System.
Decision Support|
A typical retrieval session is started by loading the query image from
the microscope and selecting a rectangular region which contains the
cell of interest. The region is then automatically color segmented
using an algorithm based on nonparametric cluster analysis (see nucleus delineation examples). The
area of the selected nucleus is taken proportional to
the number of pixels inside the nucleus region (the images in the
database have all the same scale).
|
|
|
The nucleus shape is characterized through similarity
invariant Fourier descriptors. The number of harmonics
which reliably represent the shape was obtained by analyzing the
stability of the segmentation algorithm. The results showed that the
segmentation is sufficiently stable for the use of the first 10
harmonics (40 Fourier coefficients) for the computation of the
Euclidean distance between two nucleus shapes.
|
|
The texture analysis module employs the multiresolution
simultaneous autoregressive model (MRSAR). A 15-dimensional feature
vector of the nucleus texture is obtained in accordance to this model.
The covariance matrix of the local feature vectors within each cell
nucleus is also computed, and the distance between two nuclei in terms
of their texture is given by the Mahalanobis distance of their MRSAR
feature vectors.
By default, the system retrieves the closest eight matches by
computing a similarity metric between the query image
and each of the images in the database. The suggested classification
of the query image is based on majority voting among the
retrievals. Note that in addition to the four original classes this
strategy may also produce a NO DECISION class. At present, three
query attributes are used: the area, shape and texture of the selected
nucleus. The similarity metric is defined as a linear combination of
the normalized distances corresponding to each attribute. The downhill
simplex procedure was employed to obtain the weights of the linear
combination. The objective function
was chosen the sum of probabilities of correct decision.
System performance was assessed by implementing ten-fold
cross-validated classification. The cross-validated
probabilities of correct decision were 0.8389, 0.9, 0.8333, and 0.73
for CLL, FCC, MCL, and NORMAL, respectively. These results are very
promising if we take into account the difficult task of
differentiating among lymphoproliferative disorders based solely on
morphologic criteria.
Typical voice commands are: Show Microscope, Open Image
##, Segment the Image, Search the Database, Show Video, Clinical
Data #. Examples of voice feedback are: Image ## Opened,
Segmentation Completed, Analyzing Texture, Database Search Completed,
Suggested Class: CLL (FCC, MCL, NORMAL).
Architecture
| Both the Micro-Controller and Decision Support components of the IGDS system were built following a client-server architecture. Click here for a detailed description of the Decision Support architecture. The Client part is intended to be used in small hospitals and laboratories to access through the Internet the database at the Server site. |
Future work
