SIVQ Image Analysis: A High-Throughput Morphology Discovery Tool for Surgical Pathologists.
Jason Hipp, Jerome Cheng, Jeff Hanson, Wusheng Yan, Jaime Rodriguez-Canales, Jennifer Hipp, Michael Tangrea, Michael R Emmert-Buck, Shing Han, Stephen Hewitt, James Monaco, Anant Madabhushi, Ulysses Balis. University of Michigan, Ann Arbor; National Institutes of Health, National Cancer Institute, Bethesda, MD; Rutgers University, Piscataway, NJ
Background: Spatially-Invariant Vector Quantization (SIVQ) is an image analysis algorithm based on morphologic and histopathologic pattern matching. Training to identify features is not required and is as simple as clicking on a particular morphologic feature. It has very high selectivity as compared to conventional morphometric approaches; and can be performed in a matter of minutes on a laptop.
Design: SIVQ was used to identify breast calcifications, microorganisms, breast epithelium and stroma, normal esophagus and tumor.
SIVQ was also integrated with an existing high-performance algorithm (probabilistic pairwise Markov model, PPMM) for the identification of prostate cancer.
SIVQ was integrated into the workflow of the Arcturus XT laser capture microdissection (LCM) instrument to dissect frozen esophageal normal and tumor tissue.
Results: SIVQ was used to analyze breast biopsy cases of patients with ultrasound densities, and count the percentages of stroma and epithelial components. The total run-time to identify and “paint” the epithelium and stroma of 500 breast images with SIVQ was under 15 hours.
SIVQ was integrated it into an existing algorithm (PPMM) for the identification of prostate cancer. By combining SIVQ's ability to identify the malignant morphology with PPMM's ability to identify abnormal luminal architecture, we show that the synergistic algorithm's cancer detection capability is superior to that of either algorithm individually. This approach is unique in that it models the “thought process” of a pathologist working up a prostate cancer case.
SIVQ was then applied to LCM. Samples of normal and tumor epithelium were profiled on the Affymetrix Genechip and compared to the gold standard, LCM by hand. Dendrograms, were highly similar to each other, thus supporting equivalency between the two dissection methods (manual vs. automated).
Conclusions: Here we demonstrate that SIVQ serves as a high-throughput morphologic analysis and selection tool to assist surgical pathologists.
SIVQ enables high-throughput clinical studies, such as extraction of percentages of specific cell types from whole-slide images.
Integration of SIVQ with LCM significantly improves the work-flow of microdissection by: 1) enabling the collection of large preparative amounts of morphologically-constrained biologic material for high-throughput expression studies and 2) significantly decreasing human-contact machine time.
Monday, February 28, 2011 1:00 PM
Poster Session II # 281, Monday Afternoon