sutures newsletter

PRODUCED BY AND FOR MEMBERS OF THE DEPARTMENT OF SURGERY July 2016 | Archived Issues

Pharmacy and Therapeutics Committee Approvals

Pharmacy Focus

Highlights of the June meeting of the Pharmacy and Therapeutics Committee are summarized in the PDF link below.

P and T Approvals - June 2016 (PDF)  


Mark Your Calendar


Grand Rounds

Click here to view a schedule of all upcoming grand rounds.


Education Schedule

Click the PDF link below to see the Department of Surgery's education schedule.

Education Schedule - July 2016 (PDF)


Surgery Scheduling

Click the "read more" for hours and contact information for surgery scheduling.

Share Your News

Know an interesting colleague we should profile? A story we should tell? Submit your ideas, meetings and events for consideration.

Click here to submit your news to Sutures

Seeing Tumors Through a Machine's Eye

By Arkadiusz Gertych, PhD

Tumor cells arise from genetic alterations in cells leading to their abnormal behavior and malignant transformation. Tumors are diagnosed by pathologists through the microscope based on morphologic criteria. Aberrant morphologic features that are caused by genetic and epigenetic changes in cancer cells seen by pathologists can be quantitated by digital image analysis.

Example of a modern image analysis workflow. A machine learning classifier is trained to recognize different components of a tissue. For instance, epithelial cell nuclei (blue), glandular lumen (red), immune cells (green) and other components (no color) can be automatically recognized in a whole histology slide with colon tissue.

  • Analysis of tumors through this technique provides multiple benefits and complements human vision:
  • Digital image analysis yields quantitative data that report on effects of the gene function on cancer cell morphology.
  • It can be used to measure interactions between cells in the tumor microenvironment. It can determine the composition of the tumor heterogeneity.
  • It is objective and can provide an opinion when pathologists disagree.

Yet technologies that enable seeing and distinguishing details of tumors that aren't readily available to an expert's eye need to be developed. Our lab has a longstanding interest in building and validating such technologies for research and pre-clinical utilization.

We recently developed an image analysis technique to analyze biopsy specimens from patients with high-grade prostate cancer who developed metastatic disease or were metastasis free. The pathologist cannot distinguish between the groups, and no pathologic features are available today to identify patients at risk of metastasis. However, our analyses of thousands of cancer cell nuclei revealed significant morphological differences in chromatin structure and cell topology that separate patients with lethal cancers from those with more indolent tumor types. We built a model based on nuclear features to predict the severity of the cancer that possesses accuracy greater than 80 percent.

We also devised a tool to quantitate branching of the vascular tree in clear cell renal cell carcinoma specimens. Some patients diagnosed with this disease receive anti-angiogenic therapy to prevent the spread and recurrence of the tumor, but selecting the most effective drug for a patient is difficult. Utilizing images of renal tumors from Cedars-Sinai and The Cancer Genome Atlas, we identified features of the vascular tree branching that lead to the identification of subjects with different survival characteristics. Future studies will apply this approach to the development of biomarkers for treatment of patients with anti-angiogenic drugs.

Despite significant advances, the field of quantitative image analysis to support pathologists in their daily routines is still in its early days of development. A large number of annotated images (usually whole histology slides — see figure) is required to train and validate single approach, and collecting such data is time-consuming. Even with these hurdles, the field slowly pushes the boundary of science. What's not easily observed by humans but can be seen and quantified through a machine's eye provides an opportunity to vastly increase biomarker discoveries. It also leads to answering exciting algorithmic and healthcare-related questions and to build a modern team science.