Abstract:

We present RERBEE (robust efficient registration via bifurcations and elongated elements), a novel feature-based registration algorithm able to correct local deformations in high-resolution ultra-wide field-of-view (UWFV) fluorescein angiogram (FA) sequences of the retina. The algorithm is able to cope with peripheral blurring, severe occlusions, presence of retinal pathologies and the change of image content due to the perfusion of the fluorescein dye in time. We have used the computational power of a graphics processor to increase the performance of the most computationally expensive parts of the algorithm by a factor of over × 1300, enabling the algorithm to register a pair of 3900 × 3072 UWFV FA images in 5-10 min instead of the 5-7 h required using only the CPU. We demonstrate accurate results on real data with 267 image pairs from a total of 277 (96.4%) graded as correctly registered by a clinician and 10 (3.6%) graded as correctly registered with minor errors but usable for clinical purposes. Quantitative comparison with state-of-the-art intensity-based and feature-based registration methods using synthetic data is also reported. We also show some potential usage of a correctly aligned sequence for vein/artery discrimination and automatic lesion detection.

Paper available at IEEE.

Abstract

Accelerating CT reconstruction algorithms with general purpose GPU has attracted plenty of attention in recent years. Many researchers have studied the techniques of implement CT reconstruction algorithms on different GPUs and different code development environment to explore their capability and performance of acceleration. This work is to investigate the performance of stream computing of filtered backprojection (FBP) using an ATI Radeon HD 4870 graphics card. CUDA and ATI stream computing are two main platforms in the field of high performance computing with general purpose GPU. ATI has released a set of development toolkits including a high-level brook+ based on C/C++ and a low-level Compute Abstraction Layer (CAL). We have investigated the performance of Brook+ in our former work. Here, we study the performance of ATI-CAL in the acceleration of CT reconstruction algorithm, and compare it with other GPGPU techniques.

Paper available at IEEE.

Abstract:

Automated image analysis of histopathology specimens could potentially provide support for early detection and improved characterization of breast cancer. Automated segmentation of the cells comprising imaged tissue microarrays (TMAs) is a prerequisite for any subsequent quantitative analysis. Unfortunately, crowding and overlapping of cells present significant challenges for most traditional segmentation algorithms. In this paper, we propose a novel algorithm that can reliably separate touching cells in hematoxylin-stained breast TMA specimens that have been acquired using a standard RGB camera. The algorithm is composed of two steps. It begins with a fast, reliable object center localization approach that utilizes single-path voting followed by mean-shift clustering. Next, the contour of each cell is obtained using a level set algorithm based on an interactive model. We compared the experimental results with those reported in the most current literature. Finally, performance was evaluated by comparing the pixel-wise accuracy provided by human experts with that produced by the new automated segmentation algorithm. The method was systematically tested on 234 image patches exhibiting dense overlap and containing more than 2200 cells. It was also tested on whole slide images including blood smears and TMAs containing thousands of cells. Since the voting step of the seed detection algorithm is well suited for parallelization, a parallel version of the algorithm was implemented using graphic processing units (GPU) that resulted in significant speedup over the C/C++ implementation.

Paper available at IEEE.

Abstract

This paper presented a new method for scene images classification via Partially Connected Neural Network. The neural network has a mesh structure in which each neuron maintain a fixed number of connections with other neurons. In training, the evolutionary computation method was used to optimize the connection target neurons and its connection weights. The model is able to receive a large number of input neurons and make it possible that classification of scene images needed neither any image preprocessing nor any feature extraction. Thus, the new method overcome the bug that loss and uncertainty of image information brought by man-made feature selection in the past. A large-scale GPU parallel computing method was used to accelerate neural network training. Though experiments of the method, we report a satisfactory classification performance especially for the scene images which contain artificial objects.

Paper available at IEEE.

Abstract:

Medical simulations of lung dynamics promise to be effective tools for teaching and training clinical and surgical procedures related to lungs. Their effectivenessmay be greatly enhanced when visualized in an augmented reality (AR) environment. However, the computational requirements of AR environments limit the availability of the central processing unit (CPU) for the lung dynamics simulation for different breathing conditions. In this paper, we present a method for computing lung deformations in real time by taking advantage of the programmable graphics processing unit (GPU). This will save the CPU time for other AR-associated tasks such as tracking, communication, and interactionmanagement.An approach for the simulations of the three-dimensional (3-D) lung dynamics using Green’s formulation in the case of upright position is taken into consideration.We extend this approach to other orientations as well as the subsequent changes in breathing. Specifically, the proposed extension presents a computational optimization and its implementation in a GPU. Results show that the computational requirements for simulating the deformation of a 3-D lung model are significantly reduced for point-based rendering.