Say "Goodbye" to Microsoft in January 2009.
Under Computational anatomy framework, we aimed to classify patients with Alzheimer's disease and health control based on the MRI scans of brain.
- Extract shape,cortical thickness, and appearance features from images.
- Super-high-dimension classification between disease of Alzheimer's type and normal aging cases.
- Explored deep learning, sparse coding and manifold learning applied to medical images.
- Matlab, bash script, C++, Linux.
This project studies the problem of embedding healthy control (HC) and Alzheimer's disease (AD) subjects into an anatomical shape space by computing a dissimilarity representation between subjects. Manifold learning techniques are applied on the dissimilarity representation to obtain embeddings for different subjects followed by classification in the embedded space. A widely-used framework in CA, large deformation diffeomorphic metric mapping (LDDMM) is used for dissimilarity measurement.
- Matlab, bash script, C++.
Developed a pipeline for full-body MRI registration by applying multi-channel Large Deformation Diffeomorphic Metric Mapping (McLDDMM). The registration is carried out from large-scale structures to fine details.
I created a pipeline for measuring gray matter thickness from MR images.
Bash Script, Matlab.
Implemented a prototype system of face-recognition. Occlusion and shadow are removed by Robust Principle Component Analysis.
Compared the performance of sparse representation classifier and SVM on Feret database.
Full-time intern, Siemens Corporate Research (SCR), 06/2012 - 09/2012
Accurate vessel detection is very important for surgery. This project explored supervised learning methods for vessel detection in CT images of liver. I implemented a number of feature extraction/evaluation and classification algorithms.
- Preprocessing of images to improve detection accuracy.
- Explored dictionary learning methods applied on medical images.
Due to the non-disclosure agreement, I cannot show any original image data from this project. Here are some detection result and 3-D basis learned via dictionary learning algorithm.
Figure. The result of vessel detection in CT image (liver).
Figure. The first 100 basises learned from dictionary learning.
Full-time research intern, Microsoft Research Asia (MSRA) , 08/2008 - 01/2009
Designed and implemented an unsupervised learning algorithm and developed a tool to automatically detect error data entries for Windows Live Local Search (predecessor of Bing maps).
Incorporated density-based clustering and natural language processing.
Full-time internship, Founder Co. Beijing, China, 03/2007 - 09/2007
Designed and developed image processing/enhancement algorithm specified for halftone image for improving printing quality of printers (collaborated with KYOCERA Co. JP). (C++/OpenCV)
Refined result using previous method.
Refined result using my proposed method.
Implemented a web-based prototype of CBIR, running on a 10,000-image database.
Developed a new Query-by-Multiple-Example method based on user's feedback to refine the result.
Written with StackEdit. Last update: 04/14/2014 by Jianqiao Feng