a national resource for computational science education

HOME UPEP Shodor Blue Waters

View Position
Position TitleParallel Adaptive Medical Image/Video Processing
SummaryOptimize the accuracy, speed, storage, and adaptation of medical image and video processing for mobile, telemedicine, real-time scan, diagnosis, biopsy, surgery, or treatment. Include select focused image compression, reduction, skeleton extraction, texture analysis, segmentation, object tracking using AI, HCI, and HPC techniques. Spatial and temporal parallel algorithms are needed for multi image regions training for parameter tuning, and real-time video targets segmentation/tracking.
Job DescriptionThe Intern will 1) first learn our current imaging/video processing tools, 2) test them with medical image/video data to assess the capability and limitations on sequential and parallel platforms, and 3) design/implement parallel real-time improvements for our tools.

Accurate image segmentation relies greatly on proper threshold selection between the target image objects (cell, organ, artery, tumor, etc.) and their surrounding background. Our previous automatic pathology screening was based on a fixed grayscale, e.g. 200, used for a batch of biopsy images from the same microscope. By comparing cell concentration and cell nucleus sizes between cancerous and normal image groups, our tool successfully distinguishes normal brain cells from questionable ones. To apply our image tool to broader types of images, automatic grayscale thresholding is necessary for developing intelligent, adaptive image segmentation. However, auto grayscale thresholding is challenging, because that not all images are captured under the same conditions. In this project, a user can select several sample RoIs in a given image, and tune the threshold set-ting until the objects that the user is interested stand out the best in each RoI. Our Java-based tool then combines the individual local RoI sample grayscale thresholds into an overall threshold to be used for the whole image, as well as similar batch of images. In the cases that a single global threshold is not sufficient, variable/adaptive local threshold can be applied. The expert-guided grayscale thresholding tool can be further extended to tune edge detection, texture features, multi RoI, and other image segmentation techniques. For precise treatment, our tool can help in accurately locating target objects and protect normal tissues during surgery, or radiation, proton, thermal and nanoshell treatments.
Conditions/QualificationsStrong self-motivated, innovative problem solver a must. Versatile in both hardware and software platforms needed. Familiar with machine intelligence technique a plus.
Start Date06/01/2011
End Date05/31/2012
LocationComputer Engineering, U of Houston - Clear Lake, Houston, Texas (Vicinity of NASA JSC and Texas Medical Center)