Artem Mikheev Email & Phone Number

Hanover, Nh, Us

New York, Ny, Us

New York, New York, United States

• Deep Learning solutions using LibTorch with C++• Developed an integrated imaging application, FireVoxel a 3D\4D GUI based multi-processor, high-performance volume processing under MS Windows (MFC), covering image analysis needs of the department. • Processing of PET exams, parametric map calculations, co-registration to MRI.• Developed general purpose system for the analysis of dynamic imaging experiment for MRI, PET and CT, w\wo input function. Analysis of large 4D datasets representing sequences of 3D Volumes. Implemented 20+ numerous dynamic morels such as T1-mapping, ktrans, AATH, Transit model, GKM with two site water exchange, dual input ktrans (Arterial\\Portal Vein), T2\T2* mapping, IVIM w\wo background, ADC, T2 biexponential, T1rho, Diffusion kurtosis etc.• Developed ultra-fast automatic brain segmentation algorithm with 2% precision compared to human segmentation. Received US and international patent for this technology.• Invented and implemented a novel MRI non-uniformity correction tool, BiCal, that substantially exceeds a de-facto standard N3 (2-3 times by measure of non-uniformity) on 7T MRI.• Implemented 3D/4D volume oriented DICOM dataset that allows easily browse an entire exam containing hundreds of DICOM series.• Researched and implemented algorithm and corresponding User Interface for 3D/4D volume co-registration, with several measures (cross-correlation,MI etc) and transformations upto arbitrary polynomial.• Developed a prototype for the real time, fully automatic Ultrasound to MRI registration during prostate biopsy exam.• Numerous integrated workflows for dynamic contrast evaluation of various organs.• Invented and developed novel 3D texture sensitive edge detector for 3D volume. This allows to achieve a superb segmentation results in wide variety of 3D processing tasks.• Developed a popular BrainMask tool for NIH (www.nitrc.org) • Automatic colon cleansing tool from 3D CT Virtual Colonoscopy images.

Developed the image scaling method. This method is based on detecting the edges of the images and preserving them during the scaling. It is used by CELARTEM and Extensis in “pxl SmartScale” Adobe Photoshop Plug-in. Received an US Patent for this invention2 Researched and developed a novel lossless image compression algorithm exceeding compression ratios of JPEG2000 codec by 5%. This algorithm is much lighter and simpler than JPEG2000, similar to JPEG LS..

Developed monochrome image filtering technology based on object recognition. Text areas on Land maps were scanned with electronic pen. Resulting image contains a lot of noise and non-text objects such as lines, which overlap text objects of interest. This non-context recognition is able to extract text from such cluttered image in real time (less than 1 second per image). OCR is successfully applied to such a filtered image with very high recognition rates.

Designed TSID file format. It allowed to work with the mosaic of wavelet compressed images up to 500GB. Optimized Arithmetic Decoder with 6X speed improvement over IBM’s implementation. Designed and implemented Compression Virtual Printer Driver. It allowed virtual printing from any Windows application to compressed image formats such as JPEG2000 and DjVu. Printer spool file was analyzed on the level of metafile objects, heuristics was applied to assign best compression for each particular object resulting in very high compression ratios.

Developed numerous updates and completely supported imaging technology inside the CADdy++ product line (the leading CAD system on German market at that time).Designed and implemented “CADdy++ Vektorisierung” raster-to-vector conversion system for scanned engineering drawings. This system streamlined the process of converting legacy paper documents into electronic CAD formats. The process of drawing conversion was greatly simplified by automating all stages of conversion workflow: scanned image filtering/clean-up, vectorization, automatic result verification and post-processing editing.Developed prototype of recognition of machine printed and handprint text under affine transformations for CAD systems.Researched and developed initial version of the imaging technology for the CADdy engineering system. This facilitated the display and manipulation of raster images inside the CAD system. Raster images of up to 2GB were comfortably processed on 486-33Mhz systems with 16MB of RAM.Implemented unique integer based implementation of JPEG image compression format. This permitted users to work with very large raster images without decompressing them to RAM. Decompression from original JPEG file was done on the fly for every zooming request.

Developed and integrated the API for semi-automatic drawing tracing and recognition. Delivered advanced “Object Snapping” technology for digitizing raster images. The operator would need to point only to the vicinity of the document feature (such as map grid marks) and the system would recognize the object and correct the input coordinates with high precision. As a result complete line and curve elements were converted with a single mouse click. This dramatically reduced eyestrain and the number of required mouse actions for the human operator and allowed a ten-fold productivity improvement in digitizing the technical drawings and maps.

Moscow, Russian Federation

Low level EGA\VGA graphics library written in x86 assembly. CAD raster-to-vector automatic conversion system development.

Bachelor Of Science, Engineering Physics

High School, Physics And Math