Innovation and Partnerships Office

Software to Analyze Morphometric Trait Variations in Tumor Samples IB-3251


  • Cancer treatment and drug development
  • Solid tumor imagery
  • Personalized medicine


  • Prognostic and predictive
  • Batch invariant, quantitative
  • Reveals heterogeneity lost in molecular profiling
  • Morphology can be linked to genome-wide studies
  • Incorporates data from very large cohorts from The Cancer Genome Atlas


Berkeley Lab researcher Bahram Parvin and colleagues have developed a package of scientific software to analyze variations in structural traits of tumor samples and reference them against data in the Cancer Genome Atlas project. Because cancer progression varies significantly among individuals, a major goal of cancer research is to identify tumor subtypes within a population. Subtyping enables identification of molecular markers and may predict response to therapy and outcome.

The Berkeley Lab software analyzes variations in morphometric traits of tumor samples originating from and imaged by most any laboratory. This package addresses and overcomes the batch effect, caused by variations in fixation and staining, that may result from different sample preparation protocols. With this batch-invariant design, the system summarizes each histology image into a data matrix for further bioinformatics analysis and molecular association. The software allows for quantification of cellular morphometric features and also can link computed morphometric representations with clinical data to predict outcome.

The program enables the processing of the massive amounts of data associated with a large cohort: each individual sample image is on the order of 10 billion pixels (100k-by-100k). The software implementation currently runs on the Berkeley Lab Lawrencium clusters and has revealed morphometric subtypes in Glioblastoma Multiforme (GBM) and Low Grade Glia from large cohorts of patients. Extensions to other cancer types, such as kidney clear-cell carcinoma and lung adenocarcinoma, are in the evaluation stage.

The software package detects and extracts each cell from the imaged histology section.  Subsequently, each cell is represented multiparameterically, e.g., size, cellularity, shape. Additional analysis also captures gross histology features, i.e., stroma area versus tumor area, to establish distinct contexts for nuclear signature. As a result tumor heterogeneity can also be quantified.

The system exploits the enormous capabilities of high performance scientific computing systems, and dovetails neatly with data-intensive information generated through genome-wide association studies. By linking large data sets of morphometric traits to large-scale genomic data, new insights for prediction of survival and response to therapy may emerge. The Berkeley Lab system facilitates access to a reference library providing information on tumor subtypes, mix grading, molecular correlates, and prognostic information about each subtype. The system is adaptable to a range of solid tumor histology imagery, and could become an important tool for the emerging field of personalized medicine.

DEVELOPMENT STAGE:  Proven principle.

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


Chang, H., Han, J., Spellman, P.T., Parvin, B. “Multireference Level Set for the Characterization of Nuclear Morphology in Glioblastoma Multiforme,” IEEE Transactions on Biomedical Engineering, Electronically published Sept.10, 2012.

Chang, H., Han, J., Borowsky, A., Loss, L., Gray, J.W., Spellman, P.T., Parvin, B. “Invariant Delineation of Nuclear Architecture in Glioblastoma Multiforme for Molecular and Clinical Association,” IEEE Transactions on Medical Imaging.

Bahram Parvin, The Cancer Genome Atlas’ 1st Annual Scientific Symposium, November 17-18, 2011.


Genetic Profile Determines Prognosis and Response to Cancer Therapy, WIB-2709