Ying (Grace) Cui Research Assistant Department of Electrical and Computer Engineering Northeastern University Boston, MA 02115 USA Office: 216 Egan Research Center Phone: 617-373-3975 E-mail: cui.yi@neu.edu

Advisors

Prof. Jennifer G. Dy , and Prof. Steve B. Jiang .

Brief Biography

Ying Cui received her B.S. degree from the Department of Electrical Engineering, Xi'an Jiaotong University , and M.S. degree from the Department of Computer Engineering, North Carolina State University in 2003. She is currently a Ph.D. student in the Department of Electrical and Computer Engineering, Northeastern University. She has been a research assistant in the data mining and machine learning laboratory since fall 2003, working on the project of clustering high-dimensional data. Also, from summer 2005, she has been a research fellow in Massachusetts General Hospital, associated with Harvard Medical School. Her work is concentrated on tumor respiratory motion analysis, gating, tracking and prediction in image guided radiation therapy.

Research Interests

• Data mining and machine learning. I am currently working on non-redundant feature selection and multiview clustering for high dimensional data, such as face, gene and text data.
• Biomedical research. I am currently working on tumor respiratory motion modeling, analysis and prediction in real-time imaged guided radiation therapy.
• Statistical pattern recognition.

Publications

• Y. Cui, X. Fern, and J. G. Dy, "Non-redundant Multi-view Clustering Via Orthogonalization, " Proceedings of the IEEE International Conference on Data Mining, Omaha, NE, October 2007.
• Y. Yan, Y. Cui, and J. G. Dy, "A Gradient Descent Based Continuous Optimization to Select Non-Redundant Principal Features," ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2008, submitted.
• Y. Cui, J. G. Dy and et. al., "Learning Methods for Lung Tumor Markerless Gating in IGRT," ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2008, submitted.
• Y. Cui, J. G. Dy, B. Alexander, and S. B. Jiang, "Fluoroscopic Gating for Lung Cancer Radiotherapy Based on Support Vector Machine," Physics in Medicine and Biology, 2008, submitted.
• Y. Cui, J. G. Dy, G. C. Sharp, B. Alexander, and S. B. Jiang, "Multiple Template Based Fluoroscopic Tracking of Lung Tumor Mass without Implanted Markers," Physics in Medicine and Biology, Vol. 52, October 2007, pp. 6229-6242.
• Y. Cui, J. G. Dy, G. C. Sharp, B. Alexander, and S. B. Jiang, "Robust fluoroscopic respiratory gating for lung cancer radiotherapy without implanted fiducial markers", Physics in Medicine and Biology (PMB), 52, 741-755, 2007.
• Y. Cui, J. G. Dy, G. C. Sharp, B. Alexander, and S. B. Jiang, "Correlation score based respiratory gating for lung cancer radiotherapy without implanted fiducial markers", 48th Annual Meeting of the American Association of Physicists in Medicine (AAPM), 2006.
• Y. Cui, J. G. Dy, G. C. Sharp, B. Alexander, and S. B. Jiang, "Fluoroscopic tracking of lung tumor mass without implanted fiducial markers", 48th Annual Meeting of the American Association of Physicists in Medicine (AAPM), 2006.

Projects

• Non-redundant Multi-view Clustering for High Dimensional Data. (This research is supported by NSF CAREER grant No. IIS-0347532. )
  • Clustering is the process of grouping "similar" objects/samples together. Typical clustering algorithms output a single clustering of the data. However, in real world applications, data can often be interpreted in many different ways; data can have different groupings that are reasonable and interesting from different perspectives. This is especially true for high-dimensional data, where different feature subspaces may reveal different structures of the data. Why commit to one clustering solution while all these alternative clustering views might be interesting to the user.
  • we propose a new clustering paradigm for explorative data analysis: find all non-redundant clustering views of the data, where data points of one cluster can belong to different clusters in other views. We present a framework to solve this problem and suggest two approaches within this framework: (1) orthogonal clustering, and (2) clustering in orthogonal subspaces. In essence, both approaches find alternative ways to partition the data by projecting it to a space that is orthogonal to our current solution. The first approach seeks orthogonality in the cluster space, while the second approach seeks orthogonality in the feature space.
  • Clustering results for real-world data in different iterations showed that our method can provide the user with different alternative partitionings:
    • Digit data: Find different digits not discovered in previous iterations automatically.
    • 
    • Face recognition: Find the 20 different persons in the first iteration. In the next iteration, by removing the information based on person, we are able to find the 4 different poses.
    • 
    • Confusion matrix for news group data. Different structures of the news data are revealed in each iteration.
    • 
    • Confusion matrix for Webpages. First, we group the webpages by their content. Then next, they are automatically grouped by universities.
    • 

• Lung tumor motion analysis and prediction for image-guided radiotherapy (IGRT) in real time. Develop tools for gated treatment of lung cancer without implanted fiducial markers. (The project is partially supported by an NCI grant (1 R21 CA110177 A 01A1) and by NSF CAREER IIS-0347532, as well as by Varian Medical Systems, Inc.)
  • Introduction:
    • Treatment errors related to respiratory organ motion may greatly degrade the effectiveness of conformal radiotherapy for the management of thoracic and abdominal lesions. This has become a pressing issue in the emerging era of image-guided radiation therapy (IGRT). Respiratory gated radiotherapy holds promise to precisely deliver a lethal dose to the tumour, while minimizing the incidence and severity of normal tissue complications, for mobile tumours in the thorax and the abdomen.
    • Acquire fluoro images

      A dual-head on-board x-ray imaging system, called Integrated Radiotherapy Imaging System (IRIS), developed at MGH. The system consists of two x-ray tubes, two flat panels, and two x-ray generators.

    • Pre-processing the image

      A frame of setup fluoroscopic images. Left: original image. Right: motion-enhanced image. The selected ROI is shown in both images as a rectangular region containing tumor positions of all phases.

• Score-based gating
  • Proposed clinical procedure
  • Gating by template matching:
    • Single template. All the images in the gating window is averaged to be a template.
    • Multiple templates. Each image in the gating window is a template.
    • Find the best representative template by clustering. Template Building: Gaussian mixture model for clustering and Bayesian Information Criteria (BIC) to automatically find the number of clusters. Using cluster means as representative templates.
    • Result, smooth correlation scores and good gating signals.

 

• Location-based gating.
  • Precise target localization in real time is particularly important for gated radiotherapy due to the reduced clinical tumor volume to planning target volume (CTV-to-PTV) margin and/or escalated dose. Implanted fiducial markers can help localize the tumor however for many lung cancer patients the implantation of markers is not a preferred procedure due to risk of pneumothorax. Therefore, direct tracking of lung tumor mass without implanted markers is necessary.
  • Build multiple templates
  • Multiple-template matching
  • Allow small shift for each template matching and take the weighted average of locations derived by those templates with high scores
  • Tracking result in absolute distance with the clinician marked tumor location.
• More to come. Solve the gating problem by support vector machines.

For more detailed information, refer to our publications.