Abstract 255: Toward Autonomous Real-Time Microcirculation Video Analysis
Hypothesis: The microscan device, by Microvision Medical, makes it possible to image the sublingual microcirculation during different shock states. The device has not gained widespread clinical acceptance, in part because video analysis is cumbersome and time consuming, and is highly user dependent. We hypothesize that it is possible to have a computer autonomously recognize the capillaries and analyze the quality of the blood flow in the image.
Methods: An algorithm was written in the open source programming language Python (version 2.6). The algorithm locates capillaries in the first frame of the movie and chooses tracking points and data mining points central to the capillaries. The relative position of the tracking points in each image is used to track camera motion during the movie and the data mining points are moved accordingly. The recorded data is pixel intensity, and the data is saved for later analysis. To date, we have used the Discrete Fourier Transform (DFT) to analyze the flow rate.
Results: The algorithm works as intended and analysis takes roughly one second a frame. DFT results from approximately 20 movies indicate that the algorithm is sufficient to track changes in blood flow. An example DFT plot is shown. Continuing work focuses on completing analysis of more movies, with an expected number of 200 completed by November, and on designing data output that is more clinically understandable.
Conclusions: We conclude that autonomous video analysis is possible. Autonomous analysis removes concerns about user variability, and reduces the time required to complete the analysis.
- © 2010 by American Heart Association, Inc.