Innovation and Partnerships Office

Classification of Tumor Composition and Delineation of Cellular Regions via Dictionary and Sparse Coding IB-2013-173

APPLICATIONS OF TECHNOLOGY:

  • Digital histopathology
  • Precision medicine in oncology
  • Oncology drug development and R&D

ADVANTAGES:

  • Impervious to biological and technical batch variation
  • Quantifies tumor composition via cell-by-cell profiling
  • Measures cellular processes such as mitosis
  • Extensible to a variety of other tumor types

ABSTRACT:

Berkeley Lab researchers Bahram Parvin, Hang Chang, and Yin Zhou have developed software enabling pathologists to quantify tumor composition via cell-by-cell profiling. The software assists with pre-identification of aberrant regions in tissue, saving pathologists’ time. It can assess frequency and organization of cellular processes such as mitosis and quantify tissue composition for assessing tumor-stroma interactions. The technology enables precision medicine by sifting through large amounts of data and will lead to improved predictive models.

This is the first attempt to use spatial pyramid matching (SPM) to classify tumor histopathology using engineered or learned features. Using a learned dictionary, excellent performance has been achieved even with a small number of training samples across independent data sets of brain and kidney tumors. The Berkeley Lab technology is extensible to other tumor types.

Existing techniques for image-based classification from a large cohort are hindered due to large technical and biological variation. The Berkeley team’s algorithms to classify tissue histology — based on strong representations of morphometric context and built on nuclear-level morphometric features at various locations and scales within the SPM framework — have been found to be extensible to different tumor types; robust despite wide technological and biological variation; invariant to different nuclear segmentation strategies; and effective across multiple tumor types with a limited number of training samples.
DEVELOPMENT STAGE: Software available for license. The technology has been evaluated on distinct data sets collected from The Cancer Genome Atlas. Details of these tests are described in the publications linked below.

STATUS: Patent pending. Available for license or collaborative research.

FOR MORE INFORMATION:

Chang H., Nayak N., Spellman P., Parvin B., “Characterization of tissue histopathology via predictive sparse decomposition and spatial pyramid matching,” Medical Image Computing and Computer Assisted Intervention. 2013;16 (Pt 2):91-8.

Nayak N., Chang H., Borowsky A., Spellman P., Parvin B., “Classification of Tumor Histopathology via Sparse Feature Learning,” Proc IEEE Int Symp Biomed Imaging. Apr. 2013. doi: 10.1109/ISBI.2013.6556499.

Chang H., Borowsky A., Spellman P., Parvin B., “Classification of Tumor Histology via Morphometric Context,” Proc IEEE Comput Soc Conf Comput Vis Pattern Recognit. Jun. 2013. doi: 10.1109/CVPR.2013.286.

REFERENCE NUMBER: 2013-173