Помогите подкорректировать английский для Research Statement

[pippo]

Здравствуйте, подаю документы на postdoc где требуют Research Statement. Английский изучал в неанглоязычной стране, так что ошибки при написаниее делаю часто, особенно с артиклями. Пожалуйста, не мог бы кто нибудь проверить правильность моего англиского? Заранее огромное спасибо


My primary area of research is image processing and real-time 3D visualization. I'm particularly interested in applied research and collaboration with specialists from various fields. During the past several years my work has been concentrated on development and application of image processing and 3D visualization methods for Virtual Colonoscopy.
Virtual Colonoscopy is a technique used to detect polyps (the pre-cancerous growths), cancer, and other diseases of the large intestine. Images are taken using computerized tomography (CT) and later, using a specific software, three-dimensional view of the inside of the large intestine is reconstructed for virtual inspection. First versions of Virtual Colonoscopy (VC) systems required a lengthy virtual navigation inside the reconstructed intestine to detect abnormalities. The next generation of VC systems tend to incorporate a Computer Aided Detection (CAD) modules for automatic detection of suspicions areas. This reduces the inspection time (cost effectiveness) and increases the specificity (detection effectiveness) of the doctor's findings since the CAD provides a 'second opinion' by highlighting suspicion regions.
In future I'm interested in working on new problems related to computer graphics and image processing and analysis found in Virtual and Augmented Reality fields. In what follows, I provide a brief description of the problems I've worked on, and explain plans for future work.
Research summary
Image processing
The CAD system developed by our group includes several image processing and analysis methods to extract and analyze the colon surface regions for presence of abnormalities.
As in many image processing pipelines one of the first methods used is a segmentation – the method to extract an object of interest from the data. The amount of data generated by a CT scanner in CT Colonography usually exceeds 200 MB and depends on the slice resolution. From VC point of view only a small fraction of these data are significant for diagnosis purpose. Reducing the original dataset to a minimally sufficient subset allows high optimization of the processing pipeline. The extracted colon surface is used by the CAD system for abnormalities' detection.
During my Master thesis I've developed (jointly with colleague) a fully automatic and accurate algorithm for colon segmentation. The algorithm uses automatic method for colon detection and locally adaptive region growing method to extract its boundaries. The overall process takes around 5-10 seconds on a standard one-CPU PC.
The next step in the processing pipeline of our CAD system is the removal of high-intensity tagged residuals - result of the patient's preparation scheme. To minimize the inconveniences for patients in CT Colonography and become an acceptable as a cancer-prevention screening method for population a Digital Cleansing method has been developed. This method digitally 'cleanse' tagged residuals from the dataset thus allows to reduce or even totally eliminate the need for conventional cleansing methods. The Digital Cleansing method also allows to inspect (visually or by the CAD system) otherwise hidden areas of the colon surface.
One of the results of my PhD research is a fully automatic multi-stage Digital Cleansing algorithm to detect and remove tagged residuals from the CT dataset. The algorithm is able to automatically detect the tagged residuals and adaptively remap values to reconstruct the colon surface without aliasing artifacts which otherwise present in case of a simple thresholding.

Virtual navigation
The use of a CAD system may dramatically reduce the diagnosis time in CT Colonography. If CAD system succeeds the radiologist is presented with a list of suspected regions which have to be approved or rejected by the doctor. In cases where the CAD system fails (insufficient preparation, reconstruction artifacts, etc.) a manual mode of the Virtual Colonoscopy system is used. The manual navigation inside the reconstructed colon is a complicated and a tedious task. In case of a positive finding (polyp or other abnormality) a distance from the rectum have to be calculated to be used as a reference in the conventional colonoscopy exam.
A Flying Path is a convenient method for automatic virtual navigation and distance estimation. The first version of the Flying Path algorithm I've developed is based on a well known in graph theory shortest-path algorithm. The calculated 3D line with this method could not be used directly for navigation purpose due to its shape and vicinity to the colon surface. In order to center the calculated 3D line inside the colon lumen I've developed iterative method which uses a local relocation of the path points and splines to smooth the final results.
The main problem of the shortest-path based method is its failure to extract a complete path in presence of holes inside the colon surface (virtual one, generated by reconstruction, segmentation or digital cleansing algorithms). This leads to wrong distance estimation and potential missing of the polyps. A second version of the flying path algorithm is based on 3D skeletons. The 3D skeleton calculation usually is an expensive operation, but I've implemented a highly optimized version which takes only around 5-10 seconds for a common CT dataset (~300 MB). The calculated skeleton have many extra branches and loops due to a very complicated shape of the colon and holes in its surface. I've developed a special heuristic method which uses a distance from the boundary and a distance from the source information to calculate a complete flying path of the colon and NURBS to smooth the final results of the algorithm. Later the algorithm has been extended with the manual editing functionality. It allows doctors to manually correct the results of automatically calculated flying path using series of operations in real-time.

Real-time 3D and volume visualization
The 3D real-time visualization techniques are extensively used by the VC systems for data representation. The first versions of VC systems used a so called Surface Rendering method, a polygonal approximation of the colon surface.
The reconstructed colon surface consists of more than 2-3 millions of polygons which was not possible to render in real-time on standard PC hardware. In order to increase the rendering performance and responsiveness of the VC system several offline and online acceleration techniques were employed, like decimation, view-frustum, portal and occlusion culling, etc.
During my Master thesis I've developed the complete 3D processing and visualization pipeline which allows a real-time navigation inside the polygonal approximation of the colon with a speed of over 40 frames per second (FPS) on standard PC hardware. I've used VTK library for mesh extraction and processing and C++ with OpenGL for acceleration techniques and visualization.
The surface rendering method has a wide hardware support (all graphics hardware is optimized to render triangles), but suffers from several limitations, one of them the approximation errors introduced during the processing. For example, during the mesh extraction or mesh processing (smoothing, decimation, etc.) small surface features, like small polyps could be eliminated. Also the presence of data noise could produce a wrong surface. These errors could lead to skipping of diagnostically significant surface abnormalities which is unacceptable in medical applications and thus other more precise and robust methods have to be used.
Until recently a Volume Rendering technique has been accessible only on a specialized and very expensive hardware. With the advent of graphics hardware it become possible to realize real-time Volume Rendering algorithms on standard PC hardware.
During my PhD research I've developed a real-time Ray Casting method for Virtual Colonoscopy. The first version of the algorithm uses a hybrid of Surface and Volume Rendering techniques and various acceleration methods to achieve a performance of 30-50 frames per second on a standard PC hardware. The second version uses the data extraction and partition methods to achieve the same performance without utilizing the helper mesh.
All the above algorithms have been integrated into commercial software and extensively tested on hundreds of CT datasets where they have shown good performance and results.
Future plans
There remain many interesting and significant problems to be addressed and studied in the area of image processing and 3D real-time visualization, in both theory and computation. I'd like to continue my research related on image processing, path planning and real-time 3D and volume visualization, but applied in Virtual and Augmenter Reality fields.