Yang Meng
Pronouns: He/Him/His
Research Mentor(s): Yanglong Lu
Co-Presenter:
Research Mentor School/College/Department: Department of Radiology / Medicine
Presentation Date: April 20
Presentation Type: Oral5
Session: Session 1 – 10am – 10:50am
Room: Breakout Room 5
Authors: Yang Meng , Jaslin Garcia-Morales, Yanglong Lu
Presenter: 4
Abstract
Imaging techniques have been widely used in medical applications to diagnose and prognose different diseases. The ability to share clinical data including medical images across different organizations and locations is significant to improve the healthcare system. However, storage and sharing of image data are challenging because the volume of the image data grows rapidly. Thus, there is a pressing need to develop a framework to share and store medical images efficiently. In this work, a physics-constrained dictionary learning is developed to improve the efficiency of data storage. Instead of using the original image, only a few pixels in the two-dimensional medical image or voxels in the three-dimensional medical image are used to reconstruct and classify the original images simultaneously. Physics-constrained dictionary learning can optimize the measurement matrix, the basis matrix, and the classification matrix simultaneously. The traditional compressed sensing (CS) technique is then used to reconstruct the original image with the limited pixels or voxels, and the optimized basis matrix. The locations of the limited pixels or voxels are identified in the optimized measurement matrix. The class of each image is then obtained by the linear combination of the classification matrix and the coefficient vector from CS. With the developed approach, the memory required for data storage and the amount of data exchanged through communication channels are significantly reduced. To demonstrate the proposed approach, a free open-access database including medical images with labels is used. In addition to the original image, features such as the area and the shape of detected tissues are extracted to improve the classification accuracy as needed. The developed physics-constrained dictionary learning method is also compared with other existing classification algorithms such as support vector machine and random forest. However, these existing classification algorithms cannot be used to classify images with the reduced amount of data.
Biomedical Sciences
