Medical Community: Abstracts

Abstracts:

AN IMAGE SEGMENTATION APPROACH TO EXTRACT COLON LUMEN THROUGH COLONIC MATERIAL TAGGING AND HIDDEN MARKOV RANDOM FIELD MODEL FOR VIRTUAL COLONOSCOPY

 

SPIE:
Lihong Li, Dongqing Chen, Sarang Lakare, Kevin Kreeger, Ingmar Bitter, Arie Kaufman, Mark Wax, Petar Djuric, and Zhengrong Liang

Submit to: MI03
Chairs: Dr. Anne V. Clough and Dr. Chin-Tu Chen
Title of Conference: Physiology and Function from Multidimensional Images

Summary:
Purpose: Virtual colonoscopy provides a safe, minimal-invasive approach to detect colonic polyps using medical imaging and computer graphics technologies. Residual stool and fluid are problematic for optimal viewing of the colonic mucosa. Electronic cleansing techniques combining both bowel preparation and image segmentation were developed to extract colon lumen from the abdominal computed tomographic (CT) images. In this paper, we present a new electronic colon cleansing technology, which employs a hidden Markov random filed (MRF) model to integrate the neighborhood information for overcoming the non-uniformity problems in the tagged stool/fluid region.

Methods: Prior to obtaining CT images for virtual colonoscopy, the patient undergoes bowel preparation of mild laxatives and a low residue diet. This also includes four 250cc doses of 2.1% w/v barium sulfate suspension taken with meals and two doses of 60 cc of gastroview (diatrizoate meglumine and diatrizoate sodium solution) taken the night prior to and the morning of the procedure. After the patient's colon was inflated with CO2 (1 -2 L) given through a small bore rectal tube, CT images are obtained in a less than 40 second breath-hold sequence using standard virtual colonoscopy parameters of 120kVp, 100 - 120 mA and 34 -40 field of view. The acquired data was reconstructed at 1 mm intervals with a 512 x 512 array size resulting in 300 - 450 images which were reconstructed into a 3D data set.

A statistical method for maximum a posterior probability (MAP) was developed to identify the enhanced regions of residual stool/fluid from the CT images. Unlike other non-MRF approaches, which have an intrinsic limitation on taking into account the spatial information, our method utilizes a hidden MRF Gibbs model to integrate the spatial information into the well-established Expectation Maximization (EM) model-fitting algorithm. Given the image data set, our method estimates the model parameters and segments the voxels iteratively in an interleaved manner, converging to a solution where the model parameters and voxel labels are stabilized within a specified criterion.

Results:
The new electronic cleansing method has been evaluated by volunteer and patient datasets. Figure 1 demonstrates the segmentation results on the tagged region with the standard EM algorithm (left) and the integration of hidden MRF-EM algorithm (right). Clearly, the results of the hidden MRF-EM algorithm are much better than those of the standard EM algorithm. Figure 2 demonstrates an overview of the entire extracted colon lumen. Figure 3 is the internal colon view rendered by our virtual colonoscopy system. In addition, the computational performance of the algorithm was also tested. It took approximately 5 minutes for a 400 image of 512X512 array size to classify the tagged region on an AMD 900 MHz-based PC system

New or breakthrough work:
In this new approach, the algorithm models the tagged materials and colon object by an isotropic Markov random field. Unlike the non-MRF approaches, our MRF-based spatial information has been integrated to the EM algorithm to estimate model parameters and segmenting voxels simultaneously. The use of a hidden MRF model overcomes the non-uniformity problems in the tagged fluid/stool regions, which are the major obstacle for virtual colonoscopy without the conventional physical colon cleansing technique. By eliminating or minimizing this bowel preparation with stool/fluid tagging and image segmentation, virtual colonoscopy will be more accepted by patients.

Conclusion:
We present an electronic colon cleansing technology based on a hidden MRF model and MAP-EM algorithm for extracting colon lumen from the abdominal CT images. We have tested this algorithm on a user-friendly virtual colonoscopy system which utilizes navigator through the colon. The results indicate that the algorithm is feasible for virtual colonoscopy to detect colonic polyps.

Keywords: virtual colonoscopy, virtual endoscopy, volumetric segmentation, Markov random field, electronic colon cleansing, massive polyp screening, early cancer detection, interactive visualization.

Back to Abstracts