Those of us working in biomedical informatics are well aware that one of our omnipresent challenges is overcoming the disparity between our capacity to use high-throughput research technologies to produce huge quantities of data, which is now very high, and our practical ability to store, aggregate, analyze, and integrate the resulting mountains of heterogeneous data, which stubbornly remains quite limited. If we are to translate informatics capabilities into the everyday clinical-care setting—for example, to inform diagnosis and support clinical decision-making—we must meet that challenge. And nowhere is that challenge more acutely felt than in the domain of digital pathology imaging, which has enormous potential to serve as a diagnostic tool.
As part of the ongoing CBIIT Speaker Series, we recently invited Dr. Ulysses Balis, director of the Division of Pathology Informatics in the Department of Pathology at the University of Michigan Medical School, to tell us how his research group is addressing this challenge. Over the past few years, pathology has been moving toward the adoption of digital whole-slide imaging for research, education, and routine clinical workflow. Yet, as Dr. Balis argues, we are only at the midpoint of the technology adoption curve, with 100 percent digital workflow penetration remaining elusive and no use yet of whole-slide imaging data for diagnostic pre-screening in the clinic.
For some time, Dr. Balis and his research team have been developing novel image analysis techniques based on Vector Quantization. To overcome the enormous degree of complexity inherent in utilizing square vectors, they have perfected the use of circular vectors, which exhibit continuous symmetry. They have termed this new approach “Spatially Invariant Vector Quantization” (SIVQ). This transformation radically reduces the degrees of freedom in the analysis to the point where a clinical pathologist can easily map and compare features through the resulting precise pattern-recognition capability. The spatial reduction achieved through the SIVQ algorithm facilitates the rapid clinical interpretation of image data, making its use practical in the clinical setting and obviating the need for an image-analysis expert.
Dr. Balis also reported on his research group’s work combining SIVQ analysis with laser-capture microdissection (LCM), a technique pioneered at the National Cancer Institute. The cellular and molecular heterogeneity present in tissue samples, which consist of tumor cells and stroma, means that pathology slides typically exhibit a highly unfavorable signal-to-noise ratio. By enabling the isolation of defined cell populations from tissue samples, LCM improves the proportion of signal in comparison to noise. But isolating cell populations today is time-consuming and requires a high level of expertise, thus hampering its implementation in the clinic. Using SIVQ to guide LCM allows for cell identification to become automated, enabling rapid and efficient large-scale tissue extraction and reducing the need for the high degree of expertise required by manual LCM, thus making clinical adoption practical.
To listen/view a screen cast of Dr. Balis presenting his slides, please visit http://www.youtube.com/watch?v=j1yh-Oak528&feature=player_embedded..
Key publications cited in this presentation include
- JD Hipp, JY Cheng, M Toner, RG Tompkins, UJ Balis, Spatially Invariant Vector Quantization: A pattern matching algorithm for multiple classes of image subject matter including pathology. J Pathol Inform 2011;2:13.
- J Hipp, J Cheng, JC Hanson, W Yan, P Taylor, N Hu, J Rodriguez-Canales, J Hipp, MA Tangrea, MR Emmert-Buck, U Balis, SIVQ-aided laser capture microdissection: a tool for high-throughput expression profiling. J Pathol Inform, published online Jan 28 2012.
Jose Galvez, MD is Director of NCI Enterprise Informatics, Center for Biomedical Informatics and Information Technology (CBIIT), National Cancer Institute (NCI).You may connect with Jose via LinkedIn.
Laura Kay Fleming, Ph.D, Scientific Writer, CBIIT, NCI.