Colonic polyps appear like elliptical protrusions on the inner wall of the colon. Curvature based features for colonic polyp detection have proved to be successful in several computer-aided diagnostic CT colonography (CTC) systems. Some simple thresholds are set for those features for creating initial polyp candidates, sophisticated classification scheme are then applied on these polyp candidates to reduce false positives. There are two objective functions, the number of missed polyps and false positive rate, that need to be minimized when setting those thresholds. These two objectives conflict and it is usually difficult to optimize them both by a gradient search. In this paper, we utilized a multiobjective evolutionary method, the Strength Pareto Evolutionary Algorithm (SPEA2), to optimize those thresholds. SPEA2 incorporates the concept of Pareto dominance and applies genetic techniques to evolve individual solutions to the Pareto front. The SPEA2 algorithm was applied to colon CT images from 27 patients each having a prone and a supine scan. There are 40 colonoscopically confirmed polyps resulting in 72 positive detections in CTC reading. The results obtained by SPEA2 were compared with those obtained by our old system, where an appropriate value was set for each of those thresholds by a histogram examination method. If we keep the sensitivity the same as that of our old system, the SPEA2 algorithm reduced false positive rate by 76.4% from average false positive 55.6 to 13.3 per data set. If the false positive rate is kept the same for both systems, SPEA2 increased the sensitivity by 13.1% from 53 to 61 among 72 ground truth detections.
Delineation of objects within medical images is often difficult to perform reproducibly when one relies upon hand-segmentation. To avoid inter- and intra-user variability, a semi-automatic segmentation method can more accurately and consistently determine the object boundaries. This paper presents a semi-automatic process for determining the length and volume of the spinal cord between adjacent pairs of intervertebral discs and the total length and volume of the spinal cord. A level set segmentation was performed on MRI data with user selected landmarks in order to obtain a segmentation of the spinal cord. The length and volume measurements were performed on 20 segments from C1 to L1 with five sets of user selected landmarks. Our results show that the average spinal cord segment length was 21.55 mm with a standard deviation of 25.11% and the average spinal cord segment volume was 2,217.16 mm3 with a standard deviation of 80.51%. The measurement variability of a single anatomical length across multiple trials of different sets of seed points was three orders of magnitude lower (0.06%) than the variability across different anatomical lengths (25.23%), while the measurement variability of a single anatomical volume across multiple trials of different sets of seed points was two orders of magnitude lower (0.37%) than the variability across different anatomical volumes (79.24%). Our method has been demonstrated to be potentially insensitive to intra- and inter-user variability.
We introduce an intuitive measure of computer aided detection (CAD) system performance that can handle simultaneous variation of multiple parameters. On the example of CAD of colon polyps we demonstrate how this measure was used to find the optimal parameters and make improvements to the "water-plane" algorithm that finds initial polyp candidates on the colon wall. In particular, we improved the merging of overlapping clusters to only create fused clusters if they shared at least 50% of their vertices and adjusted size and thickness filter criteria to retain more true positive detections. The system, containing all optimizations, improved significantly over the original system that found initial detections based only on colon surface curvature. This improvement was measured by both free response operating curve (FROC) analysis and our new performance measure.
Virtual colonoscopy is becoming a more prevalent way to diagnose colon cancer. One of the critical elements in detecting cancerous polyps using virtual colonoscopy, especially in conjunction with computer-aided detection of polyps, is that the colon be sufficiently distended. We have developed an automatic method to determine from a CT scan what percentage of the colon is distended by 1cm or larger and compared our method with a radiologist's assessment of quality of the scan with respect to successful colon polyp detection. A radiologist grouped 41 CT virtual colonoscopy scans into three groups according to the degree of colonic distention, "well", "medium", and "poor". We also employed a subvoxel accurate centerline algorithm and a subvoxel accurate distance transform to each dataset to measure the colon distention along the centerline. To summarize the colonic distention with a single value relevant for polyp detection, the distention score, we recorded the percentage of centerline positions in which the colon distention was 1cm or larger. We then compared the radiologist's assessment and the computed results. The sorting of all datasets according to the distention score agreed with the radiologist's assessment. The "poor" cases had a mean and standard deviation score of 78.4% ± 5.2%, the "medium" cases measured 88.7% ± 1.9%, and the "well" cases 98.8% ± 1.5%. All categories were shown to be significantly different from each other using unpaired two sample t-tests. The presented colonic distention score is an accurate method for assessing the quality of colonic distention for CT colonography.
Low radiation dose requirements create relatively noisy images that contribute to high numbers of false positive detections in CAD for CT colonography. Presumably image denoising techniques such as non-linear, edge-preserving smoothing filters can improve automatic colonic polyp detection in CT colonography by reducing overall per patient false positive rates. Here, we have evaluated multiple edge-preserving smoothing filters to determine whether this is so. Prone and supine scans from 81 asymptomatic, average-risk adults with adenomatous polyps were studied with and without smoothing. FROC curves were generated to analyze CAD results. A single, clinically relevant operating point was compared between the best smoothing filter results and the unsmoothed data. Improvement in performance was observed, but the differences were not found to be statistically significant for average dose CT colonography.
Given a segmented CT scan data of the colon represented as a triangle mesh, our water-plane algorithm will detect polyp candidates. The water-plane method comprises of pouring water into a polyp protrusion from the outside of the colon and in raising the “water-plane” until it cannot be incremented any further without causing water leakage. The method starts at a vertex and uses average normal of all triangles adjacent to the starting vertex to generate the initial water-plane, which will make the starting vertex “wet” but leave its neighboring vertices “dry”. The method will continue to wet neighboring vertices one by one and then their neighbors and so on until the water-plane cannot move any further without causing water leakage. The water-plane movement alternates between just raising the water level in completely convex regions and tilting about one or two anchor vertices that have neighbors that would get wet if the water level was raised any more. The final set of wet vertices is a cluster that is an initial polyp candidate. The water-plane method was compared against the current polyp candidate detection method in our Computer Aided Detection of Colon Polyps software pipeline, called the surface curvature method. It finds clusters of connected vertices that all exhibit elliptical curvature. The water-plane method showed multiple improvements in polyp candidate detection. It detected polyp candidates missed by the surface curvature method. It exhibited continuous polyp candidate regions instead of non-uniform or incomplete regions detected by the surface curvature method. And finally, it avoided some false positive detections reported by surface curvature method.
The presented method significantly reduces the time necessary to validate a computed tomographic colonography (CTC) computer aided detection (CAD) algorithm of colonic polyps applied to a large patient database. As the algorithm is being developed on Windows PCs and our target, a Beowulf cluster, is running on Linux PCs, we made the application dual platform compatible using a single source code tree. To maintain, share, and deploy source code, we used CVS (concurrent versions system) software. We built the libraries from their sources for each operating system. Next, we made the CTC CAD algorithm dual-platform compatible and validate that both Windows and Linux produced the same results. Eliminating system dependencies was mostly achieved using the Qt programming library, which encapsulates most of the system dependent functionality in order to present the same interface on either platform. Finally, we wrote scripts to execute the CTC CAD algorithm in parallel. Running hundreds of simultaneous copies of the CTC CAD algorithm on a Beowulf cluster computing network enables execution in less than four hours on our entire collection of over 2400 CT scans, as compared to a month a single PC. As a consequence, our complete patient database can be processed daily, boosting research productivity. Large scale validation of a computer aided polyp detection algorithm for CT colonography using cluster computing significantly improves the round trip time of algorithm improvement and revalidation.
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 bowel preparation, oral contrast agents, and image segmentation were developed to extract the colon lumen from computed tomography (CT) images of the colon. 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 within the tagged stool/fluid region. Prior to obtaining CT images, the patient undergoes a bowel preparation. A statistical method for maximum a posterior probability (MAP) was developed to identify the enhanced regions of residual stool/fluid. The method utilizes a hidden MRF Gibbs model to integrate the spatial information into the Expectation Maximization (EM) model-fitting MAP algorithm. The algorithm 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. Experimental results are promising.
In our previous work, we developed a virtual colonoscopy system on a high-end 16-processor SGI Challenge with an expensive hardware graphics accelerator. The goal of this work is to port the system to a low cost PC in order to increase its availability for mass screening. Recently, Mitsubishi Electric has developed a volume-rendering PC board, called VolumePro, which includes 128 MB of RAM and vg500 rendering chip. The vg500 chip, based on Cube-4 technology, can render a 2563 volume at 30 frames per second. High image quality of volume rendering inside the colon is guaranteed by the full lighting model and 3D interpolation supported by the vg500 chip. However, the VolumePro board is lacking some features required by our interactive colon navigation. First, VolumePro currently does not support perspective projection which is paramount for interior colon navigation. Second, the patient colon data is usually much larger than 2563 and cannot be rendered in real-time. In this paper, we present our solutions to these problems, including simulated perspective projection and axis aligned boxing techniques, and demonstrate the high performance of our virtual colonoscopy system on low cost PCs.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.