Proceedings IEEE International Symposium on Biomedical Imaging

Describes a new technique for estimating the out-of-plane distances between neighbouring 2D B-scan images. Probabilistic analysis of the speckle signal intensity leads to a distance estimation algorithm based on linear regression. The quality of fit to the regression line indicates the reliability of the distance estimate. The new technique can be used to determine the three out-of-plane degrees-of-freedom of scan plane motion. The in-plane translations and rotation can be deduced by correlating neighbouring B-scan images. As a result, the relative 3D positions of all the 2D B-scans can be calculated without the need for a position sensor attached to the ultrasound probe. The approach is useful for recording 3D data when a position sensor is not available, or for improving the local consistency of a conventional sensor-based recording.

We present a bottom-up approach for automatic cancer cell detection in multispectral microscopic thin Pap smear images. Around 4,000 multispectral texture features are explored for cancer cell detection. Using two feature screening measures, the initial feature set is effectively reduced to a computationally manageable size. Based on pixel-level screening results, cancerous regions can thus be detected through a relatively simple procedure. Our experiments have demonstrated the potential of both multispectral and texture information to serve as valuable complementary cues to traditional detection methods.

We wish to investigate the complex temporal dynamics of spatially distributed, highly interconnected brain regions, but available methods do not measure all relevant aspects of the system. We must take a convergent multidisciplinary approach, employing multiple methods and defining the relationships between disparate methods. Having measured multiple properties of the system, we seek to understand the functional relevance of observed spatial and temporal patterns of activity through analytical methods, descriptive modeling and ultimately, theoretical modeling. Functional MRI studies can identify active regions - the nodes of the distributed networks that comprise functional brain systems. Coupled with new techniques for investigating and modeling human anatomical connectivity in vivo, these methods allow us to define the functional architecture of human cognition. Human electrophysiological techniques (e.g. MEG and EEG) allow us to investigate activity on millisecond time scales not routinely accessible with fMRI. However, temporal coding can take many forms and should not be considered solely in terms of discrete activation at specific anatomical nodes in the system. The challenges for modeling these macroscopic data are formidable.

Strain encoded magnetic resonance elastography (SENC MRE) has been proposed for imaging elastic variation by obtaining images whose intensity directly represents through-plane strain. It is useful for detecting tumors because tumors exhibit different elastic properties than the surrounding tissue. SENC MRE, however, assumes homogeneous tissue elasticity in the through-plane direction, an assumption that requires the acquisition of relatively thin image slices. This causes a reduction in the signal-to-noise-ratio (SNR) of the images and lengthens the scanning time when acquiring large volumes. In this paper, we evaluated imaging with thick slices (>10 mm) where the homogeneity assumption would fail in cases of small tumors that are thinner than the image slice. We propose enhancing the SENC MRE technique to detect thin tumors by modifying the image slice profile. Phantom experiments were performed to demonstrate the ability of the improved SENC MRE technique to detect the presence of thin tumors, as small as one quarter of the slice thickness.

We give a method for estimating the optimal rotation between anatomical curves corrupted by segmentation noise. We formulate the energy function probabilistically and use an annealing algorithm with explicit outlier rejection to stabilize our solution. We also provide methods for a nearly symmetric estimation of the rigid transformation between three-dimensional curve pairs. We evaluate the performance of the algorithm numerically with respect to the energy function and computation time.

In order to assess the effect of the use of a high-resolution micro-angiographic detector on the accuracy of performing interventional localization tasks, we set up a simple task to localize a guide wire tip relative to one edge of a thin marker wire. Trained observers were asked to evaluate the micro-angiographic images as well as the comparison standard x-ray image intensifier (XRII) images, both of which were taken with the guidewire in the same positions. The observers were able to get more accurate estimation results using images obtained with the micro-angiographic detector than the XRII; the average distance error and performance variability both were reduced by factors of 2-3.

Some of the methods used to reduce estimation errors make use of simplified models of the Doppler sample volume (SV). However, these models do not accurately take into account the SV's shape and pressure distribution. The SV can be accurately simulated through the inclusion of a Doppler time gate to a B-mode simulation based on the underlying physics of ultrasound propagation and reception from a linear phased array transducer, including the effects of focusing and attenuation. A commercial diagnostic B-mode/Doppler ultrasound system was simulated to produce four SVs at different axial depths and lengths. It was found that every SV's axial length was larger than the theoretical length used in its design. The further the SV was away from a focal depth, the greater its size in that dimension. The pressure distribution was not symmetrical about the center of the SV, but rather skewed towards the transducer.

Segmenting, or outlining, the prostate boundary is a very important task in diagnosing and treating cancer. In this paper, an algorithm is described for semiautomatic segmentation of the prostate from 3D ultrasound images. The algorithm uses model-based initialization and mesh refinement using an efficient deformable model. Initialization requires the user to select only six points from which the outline of the prostate is estimated using shape information. The estimated outline is then automatically deformed to better fit the prostate boundary. The performance of the algorithm with a single user was compared to manual outlining by the same user. The average distance between semiautomatically and manually outlined boundaries was found to be 1.19 mm, and the average difference in volumes was 7.2%.

Trong khuôn khổ chẩn đoán hỗ trợ máy tính, bài báo này đề xuất một chức năng mới cho nghiên cứu phế quản phổi dựa trên chụp cắt lớp vi tính (CT). Nó cung cấp một phương thức hình ảnh cụ thể - phế quản CT ảo - cùng với việc điều tra định lượng thông qua một mô tả dựa trên trục trung tâm. Chức năng này dựa trên việc tái cấu trúc 3D dựa trên năng lượng của cây phế quản cho đến các phân nhánh thứ 6. Sự độc đáo của phương pháp tái cấu trúc 3D nằm ở việc kết hợp các tiềm năng lan tỏa trục và bán kính trong một quy trình tăng trưởng, được áp dụng cho một tập con các phế quản có bậc thấp được trích xuất từ khối lượng CT bằng một toán tử hình thái học toán học mạnh. Trục trung tâm của cây, cung cấp khả năng điều hướng và định lượng, được tính toán thông qua một quy trình đệ quy áp dụng cho từng đoạn phế quản, dựa trên sự lan tỏa của đường biên địa lý. Được xác thực bởi các bác sĩ chẩn đoán hình ảnh chuyên nghiệp, các ví dụ về phế quản CT ảo được trình bày cho cả phế quản bình thường và bệnh lý, và mối quan tâm về phế quản CT định lượng trong chẩn đoán lâm sàng và theo dõi bệnh được thảo luận.

Processing Affymetrix/spl reg/ GeneChip/spl reg/ array images requires accurate and robust methods. Algorithms are designed to handle a wide variety of input image data. Execution time and interface to database products are considered during design and development. Evaluating the image processing algorithms include analysis of application results, such as, gene expression and genotyping.