Work_in_Progress:_Finding_Gold_with_FLOSS
title: Work in Progress: Finding Gold with FLOSS permalink: /Work_in_Progress:_Finding_Gold_with_FLOSS/
CLICK HERE FOR THE CURRENT VERSION: "HOW TO FIND GOLD WITH FLOSS"
Nicolas Robidoux, John Cupitt and Chantal Racette (2009).
(Title inspired by Gepostet von Stani's "How to make money with free software.")
This document discusses an ongoing research project which concerns the automatic and human-assisted identification of 5 nm gold nanodisks in Transmission Electron Microscope photographs of live cell membranes. Nip2 is a key component of the project's workflow. Stanly Steinberg of the Center for the Spatiotemporal Modeling of Cell Signaling (STMC), a NIH/NIGMS Center of Excellence in Complex Biomedical Systems Research, brought to our attention the need for better software in this context. Following is an excerpt of Characterizing the Topography and Interactions of Membrane Receptors and Signaling Molecules from Spatial Patterns Obtained using Nanometer-scale Electron-dense Probes and Electron Microscopy by J. Zhang, K. Leiderman, J. Pfeiffer, B. Wilson, J. Oliver and S. Steinberg, Micron, Volume 37, Number 1, pages 14-34 (2006):
The flow of information through a cell requires the constant remodeling
of cell signaling networks. Thus, spatially- and temporally-resolved
microscopy of signaling components is needed to understand the behavior
of normal cells as well as to uncover abnormal behavior leading to human
disease. Nanoprobe labeling and transmission electron microscopy of
cytoplasmic face-up sheets of cell membrane has been developed as a high
resolution approach to map the interactions of proteins and lipid during
cell signaling. Membrane sheets are labeled with 3-15 nm electron-dense
probes for receptors, signaling proteins and lipids and micrographs
record the distributions of the probes relative to each other and to
surface features.
Hence the need for identifying and finding the locations of gold nanodisks, the "3-15 nm electron dense probes," with Free Libre Open Source Software.
3330_cropped.tif is a typical input image. According to the gold standard prepared by Nick Andrews of the University of New Mexico, this test image contains 235 gold disks, each of which is identified by approximate location. (An uncropped version of the test image, 3330.tif, is the actual input of the Nip2 workspaces.)
Here is the final result, with approximate disk locations tagged with single pixel red squares: 3330_all_disks.tif. (Warning: the red dots are VERY SMALL: Use a high magnification and a non-smoothing viewer.) There is handful of false positives, which we expect can be eliminated using a test which focuses on "mid-size" disks. (In the current version of the software, we only have a test tuned for the smallest disks, and a test tuned for the larger ones; see below.)
Without relying on the gold standard, I (Nicolas) see a false positive near [940,435], a false positive near [700,1100], a false positive near [845,1735], two false positives near [830,1975], and a false positive near [1450,1760]. Ambiguous locations include a possible false positive near [200,2350], a possible false negative near [2250,1430], a possible "siamese twin" false negative near [2115,1730], and between one and four possible false negatives near [1150,1500],
Note that a conjunction ("this location is deemed to identify a gold disk if it passes the first test AND the second test AND ... the last test") of no false negative tests is also a test with no false negatives provided one ensures that the criteria are sufficiently well aligned, that is, that they all give a "pass" at every pixel location where they should. If alignment is not taken into account, one test could give a pass at one "true positive" pixel location, and another test could give a pass at a nearby, but not identical, pixel location, with none of them being in agreement at exactly the same location, leading to an avoidable false negative.
Likewise, a disjunction of no false positive tests is also a no false positive test. The goal of this project is to produce a robust nanodisk detector with no false positive and no false negative.
The justification for the "no false negative" policy is that it is much easier for a human operator to identify false positives (which the program gives coordinates for) than hunt for false negatives (which may be anywhere and which, by virtue of being false negatives, are generally hard to see).
The Nip2 workspace gold_disks_small.ws, programmed by N. Robidoux, J. Cupitt and C. Racette, identifies smaller disks. (Note that Nip2 stores workspace information in xml format: some browsers may complain.) The resulting small disk "centers" are shown in 3330_small_disks.tif, mostly as single black pixels on a white background. (This image is a "save" of output A2 in the workspace.) This first workspace implements a no false positive test which tags a location to be at the "center" of a "small disk" if the average near this location, computed with gaussian blur with a radius comparable to the radius of the smallest visible disks, is sufficiently small (dark) compared to each of eight averages computed over the intersection of a tight concentric ring and an octant. This "peak detector," being similar to an edge detector, is easily contaminated by groups of very bright pixels near very dark pixel regions. Such white/black alternations often occur in the TEM photographs near cell membrane "folds," not near gold disks, triggering false positives. To minimize this contamination, the white point of the image is manipulated prior to the above computation to make very bright pixels "grayer."
The Nip2 workspace gold_disks_large.ws, also programmed by N. Robidoux, J. Cupitt and C. Racette, identifies larger disks. The resulting larger disk "centers" are shown in 3330_large_disks.tif as one or two pixel black dots. This second test deems a location to be at the "center" of a "large disk" if a criterion similar to the one used to detect small disks is satisfied with larger gaussian and ring radii.
Each of the above two tests are combined (by "AND"ing) with the following, no false negative, tests (without which the black islands would be much larger):
**-**Is the (blurred) center dark enough? (The blur radius is small for the small disk test, larger for the other.)
**-**Is the (blurred) center darker than every immediate (blurred) neighbor?
**-**Is the (blurred) center darker than every (blurred) neighbor at a distance of 2 (in max norm)?
These tests help "separating" nearby gold disks.
Care must be taken when combining the results of the last two tests with the others because they yield one or two pixel wide positive "islands," and consequently they are very sensitive to alignment. One way of mitigating the "perfect alignment" issue is to use 8-connected then 4-connected "erosion" to enlarge the mostly single pixels to 5x5 rounded black squares. (Note that the current versions of the two workspace each compute identical versions of these two tests, instead of combining "all at once." This is a waste.)
The results are then "OR"ed in a third workspace, gold_disks_all.ws, programmed by John Cupitt which yields a no false negative test with "centers" highlighted with single pixel red dots for visualization purposes. This workspace also produces a matrix of "center" coordinates, which we'll include on this wiki when we're done eliminating false positives. (Warning: gold_disks_all.ws requires operations only available in the very latest Nip2 and VIPS.)
Chantal Racette is currently trying to eliminate the last few false positives. Note that there is some ambiguity in the gold standard: some experts see more gold than others.
The "red dot software" has a left/up bias, which we intend to fix.
The workspaces need to be put through their paces on other micrographs in the same, and other, batches.
This document will be updated as additional criteria are added, parameters are further optimized, and the whole process is made both more robust and easier to adapt to other situations.