We narrowed Medical Application to two possibilities
|
1. Telemedicine / Interventional Informatics /CareWeb (SU College of Nursing, Suny HSC, NPAC, Syracuse City Scool District)2. 3D Visualization of Medical Data - Visible Human (Data available, 14 GB from National Library of Medicine)
|
We have demonstrations of CareWeb which is quite similar in structure to C2 application (A3) SanD Diego, Health Info/Bridge, January Washington DC, HealthInfo, April Syracuse, Telemedicine Conference, May
|
We have focused on visualization issues in 2) - 3D interactive visualization
|
of human internal organs in their true form and shape
|
Funded by the National Library of Medicine implemented at the University of Coloradio HSC.
|
39 - year old white male who donated his body to science after being convicted of murder and sentenced to death.
|
Magnetic Resonance Images (MRI - 256x256x12) - head scanned axially, the other sections scanned coronally.
|
Computed Tomography (CT - 512x512x12) - soft tissue and bones.
|
Anatomical color photographs - 1878 transverse slices, each 1 milimeter wide; Each slice of the original data is in a 2048x1216 pixel 24-bit color image.
|
NPAC has obtained a copy of the Visible Human data set 14GB and
|
license to use it.
|
Female VH not downloaded : 5,000 cross sections (40 GB). Good for 3D reconstructions (cubic voxels, 0.33mm size).
|
We have cropped the original images and removed the gelatin backgroud
|
We have constructed slices in two orthogonal panels (sagittal and coronal views); aligning required - linear interpolation between slices; best fit of features as a function of translations and rotations
|
The resulting images were converted to JPEG format using a 75%
|
For 3D reconstruction we focused on the human head
|
Image Clipper Application program (demo, written in Java) allows to cut the selected portion of the image and apply the procedure to all images in the stack (700 images XxY)
|
Data stored in voxel represenation in Illustra object-relational database (x, y, z, r, g, b)
|
The original dataset reduced to 70 images XxY
|
Image volume must be segmented into its anatomical constituents.
|
Automatic segmentation is not possible.
|
Interactive, semiautomatic segmentation is time consuming (one man-year to segment 300 anatomical objects).
|
Algoritms ( histogramming, thresholding, edge enhancements, non-rational uniform B-spline).
|
Map Seperation methon in AVS (demo) (mussle easy, bone difficult).
|
Map separation implementation in Java (demo).
|
Bone tissue cannot be separated from anatomical images.
|
We must come back to CT images and do aligning CT-anathomy.
|