Home > Research Projects > Pathology Diagnosis

Computational Support for Cancer Pathology Diagnosis

Collaboration Project with:

Edmond Szabo MD and Murray Resnik MD PhD
Department of Pathology RIH and Brown University

People: Haynes Heaton, H. Can Aras

  • Digitally cloning Dr. Resnick
  • Determining the level of the cancer from
    the biopsy images using computational techniques

Our goal is to create a completely automated algorithm to determine the level of malignancy of images of tumor biopsies. We do this by first segmenting the nuclei out of the image, defining statistics about the "pleomorphism" of the nuclei, and then using machine learning techniques to categorize the images into normal, low, and high malignancy. Pleomorphism includes nuclei size, variability of size, complexity of shape, eccentricity, heterogeneity of color, heterogeneity and organization of orientation, and many more higher level features that all deal with complexity and heterogeneity that result from malignant tissue.

Approach: Segmentation
  • watershed segmentation
  • use of gradient of the image
  • intensity value at each pixel