GC Image (LCxLC Edition) Users' Guide

Peak Detection

After correcting the baseline, the various groups of datapoint pixels associated with the separated analytes can be detected and accurately quantified. In image processing, each such group of pixels is called a blob and the operation of segmenting pixels into blobs is called blob detection. LC Image implements blob detection as described in this chapter. This chapter also describes tools for blob display, selection, local blob detection, drawing, excluding, editing, and deconvolving/unmixing. The next chapter describes related tools for analyzing blobs.

Detect Blobs

To detect and quantify blobs in the chromatogram, invoke the Detect Blobs operation, either by clicking the Detect Blobs button on the Image Viewer tool bar or selecting Filter -> Detect Blobs from the menu. The blob detection operation takes a brief time (typically no more than a few seconds) and does not change the image data.

The blob detection algorithm uses a greedy dilation that successively attaches the largest-valued unassigned pixel to a neighboring blob or forms its own blob if no neighboring peak has been established. Configure -> Configure Settings on the Image Viewer menu bar provides parametric settings for Blob Detection:

• Before the algorithm is run, the data values are smoothed to reduce spurious variations associated with noise. The smoothing is considered only during the detection algorithm and no pixel values are actually changed. The widths of the Gaussian blurring filter is configurable. The default width parameters are 0.5 pixels in the first column direction and 2.0 pixels in the second column dimension. The minimum values are 0.0. Larger values tend to produce more errors where two adjacent peaks are detected as a single peak, whereas smaller values tend to produce more errors where a single peak is detected as two separate peaks. If the chemical peaks have many pixels in one or both dimensions of the separation, the corresponding smoothing parameters can be increased.
• The blob detection algorithm implements a post-processing filter to eliminate detected blobs that have either (a) a temporal footprint or area (i.e., number of included pixels) that is too small, (b) volume (i.e., total integrated response) that is too small, or (c) a maximum value (i.e., apex) that is too small. The default minimum area is 22 pixels. The default minimum volume is 0. The minimum apex can be configured either as Relative or Absolute. Relative sets the minimum apex threshold to the parametric value times the estimated standard deviation of the noise, whereas Absolute sets the minimum apex threshold to the parametric value. The default minimum apex is Relative and 5.0, i.e., five times the estimated standard deviation of the noise. The minimum value for all three blob detection filter parameters is 0.

Figure 1 illustrates the Configure -> Configure Settings -> Blob Detection dialog.

Figure 1: The Blob Detection configuration pane.

The Detect Blobs operation may detect hundreds or even thousands of separate blobs in the image. There are a number of important tasks in analyzing the large volume of blob data.

• A user may wish to identify and report on only a fraction of the chemicals in the sample, so one of the tasks of analysis is to designate those blobs that are to be included for reporting.
• In a report, the blob data usually should be associated with the name of the chemical that produced the blob and/or the chemical group that produced the blob, so one task of analysis is to identify the blobs by associating chemical names and/or chemical group names with the blobs to be included for reporting.
• Calibration of the system and accurate quantification of peaks may require the identification of internal standards (chemicals of known quantity injected into the sample mixture) and the association of sample chemicals with specific internal standards. So, analysis may require the identification of blobs produced by internal standards and the correct associations between sample blobs and internal standard blobs.
• Ideally, blobs are well separated and correctly distinguished in the Detect Blobs operation. In practice, imperfections due to noise and small proximity between peaks can lead to detection errors that require editing. For example, one type of error is the splitting of what should be one peak into two blobs and another type of error is the inclusion of what should be more than peak in a single blob. So, another analysis task may be to correct mistakes in the detection of blobs to be included for reporting.
• Ideally, blobs are produced by only one chemical peak. In practice, chemicals may co-elute, producing blobs that combine the response to more than one chemical peak. Another analysis task may be to deconvolve or unmix the co-eluted peaks.

The large number of separated chemicals in LCxLC data makes manual analysis tedious and time-consuming, so the automated tools provided by LC Image are essential for efficient processing of large LCxLC data sets. LC Image has tools, described in Chemical Identification, that automate much or all of the analysis process. The approach is based on pattern matching: a template that describes the pattern of chemicals peaks in one image is matched automatically to blobs of subsequent images. Similar patterns of blob peaks are seen in runs of similar samples under similar conditions, so templates can be used to make accurate chemical identifications based on previously observed patterns. Then, the chemical names, chemical groups, and internal standard associations in the template are copied automatically to blobs in the target image. For multispectral chromatograms, perform automatic identification can be performed by searching a library of multispectral signatures for best matches, as described in LCxLC-MS Data Processing and Analysis.

Although LC Image provides a high level of automation, sometimes manual, interactive processing is required. As described below, LC Image allows users to interactively assign chemical names, chemical group names, internal standard associations, reporting status, and other metadata to individual blobs or groups of blobs. Other LC Image tools, also described below, support locally variable blob detection and interactive editing of detected blobs.

Display Blobs

LC Image graphically highlights detected blobs in the Viewport of the Image Viewer, e.g., as shown in Figure 2.

Figure 2: Graphical highlights show the unflagged blobs (circles in red), included blobs (circles in light blue), internal standards (circles in yellow), and associations between included blobs and internal standards (lines in green).

Generic Blob Display configuration settings are accessible via the Configure -> Configure Settings on the Image Viewer menu.

• A configuration setting determines whether graphical highlights for blobs are shown in other cursor modes (e.g., graphics mode). The default is to show blob highlights in all modes except template mode.
• A configuration setting determines whether graphical highlights are displayed for each of the three classes of blobs:
  • Unflagged — blobs that are detected, but neither designated as included nor designated as excluded for reports. (The meanings of inclusion and exclusion are described next.) This is the initial state for all blobs.
  • Included — blobs that are designated to be included in reports. Reports are to individually characterize (in the manner appropriate for the type of report) all included blobs.
  • Excluded — blobs that are designated as excluded in reports. Reports may include in computations blobs that are not designated as included. Excluded blobs are excluded from such computations. For example, a fractional response report characterizes the fractional response of each included blob to the total responses of all included and unflagged blobs, but does not count excluded blobs in the total.
  The default is to graphically highlight all blobs using different colors for each class.
• A configuration setting determines whether the graphical highlighting for each blob is a thin outline, a thick outline, a filled outline, a bounding box, or a bubble. The bubble indicates both retention time (by image location) and volume (by disk size). The default is to highlight with a thin outline.
• A configuration setting determines whether or not associations between sample chemicals and associated internal standards are highlighted with a line between the blob peaks. The default is to highlight the internal standard associations with a line.
• Configuration settings determine the default colors used for included blobs (default cyan), excluded blobs (default gray), unflagged blobs (default red), internal standard blobs (default orange), internal standard association lines (default green), and selection box (yellow).
• Multiple blob display settings can be configured. Then, typing the 't' (or 'T') key while the Image Viewer is active, cycles through the blob display settings.

Figure 3 illustrates the Configure -> Configure Settings -> Blob Display dialog.

Figure 3: The Blob Display configuration pane.

After blob detection, when the cursor lingers over a blob in the Image Viewer, a tool tip appears with information about the blob. A configuration tab for blob mouse-overs, pictured in Figure 4, determines which blob information is visible in blob mouse-overs. The default mouse-over information configuration is: Blob ID, Compound Name, and Peak Value. At most five fields can be shown in the blob mouse-overs.

Figure 4: The Blob Mouse-Overs configuration pane.

The Text View tool, accessible from the View -> Text View menu option or the tool bar, has a flag for displaying blob assignments at each pixel. This tool provides a quantitative view of each pixel in each blob and is very useful for closely examining blobs.

Selecting Blobs

The Image Viewer palette has four cursor modes for selecting blobs and three modes for drawing blobs:

• Point blob selection,
• Polyline blob selection,
• Rectangle blob selection,
• Polygon blob selection, and
• Draw blob free hand.
• Draw blob with oval tool.
• Draw blob with rectangle tool.

The Blob Mode button, located on the Image Viewer palette, shows only the current or most recent blob-selection mode. The Blob Mode can be changed from the pull-down menu that is adjacent to the Blob Mode button:

In point blob-selection mode, a blob is selected by moving the cursor over any pixel inside the border of the blob and clicking the left mouse-button. Once a blob is selected in pixel blob-selection mode, other blobs can be added to the selection set by depressing the control key and clicking of the left mouse-button while the cursor is inside the next blob to be selected. To select from multiple overlapping blobs (e.g., as generated by the Deconvolve/Unmix operation described below), first click the desired location, which selects the overlapping peak with the smallest blob ID. Other overlapping peaks can be selected, in a cycle ordered by blob ID, using the <TAB> key (which deselects the previous peak in the cycle) or the <CTL>+<TAB> keys (which maintains selection of the previous peak in the cycle).

In polyline blob-selection mode, blobs are selected by intersection with a drawn polyline. To draw a polyline, locate the cursor at the starting point of the polyline and click the left mouse-button. Then, similarly locate each successive point of the polyline and click the left mouse-button. At the ending point of the polyline, double-click the left mouse-button. All blobs intersected by the polyline are selected and the polyline disappears. To create a polyline that persists in the image, draw a graphic polyline, as described in Graphics and Areas .

In rectangle blob-selection mode, a group of blobs can be selected by moving the cursor to a point above (or below) and left (or right) of the peaks of all blobs to be selected, depressing the left mouse-button, moving the cursor below (or above) and right (or left) of the peaks of all blobs to be selected and releasing the mouse button. All blobs with the peak pixel (apex) in the rectangular bounding box are selected and the rectangle disappears. To create a rectangle that persists in the image, draw a graphic rectangle, as described in Graphics and Areas .

In polygon blob-selection mode, a group of blobs can be selected by moving the cursor to a point near the blobs to be selected, clicking the left mouse-button, repeatedly moving the cursor around the blobs to be selected and clicking the left mouse-button (drawing a polygon), and double clicking the left mouse-button (to close the polygon). All blobs with the peak pixel (apex) in the polygon are selected and the polygon disappears. To create a polygon that persists in the image, draw a graphic polygon, as described in Graphics and Areas .

The blob(s) that are selected are highlighted with a colored box. Figure 5 shows a selected blob with a mouse-over that shows information about the blob.

Figure 5: A selected blob is highlighted by a box.

All blobs may be selected by first setting the cursor mode to Blob Mode and then choosing Edit -> Select All (or clicking in the image and typing the <CTL>+A keys). Three additional tools for selecting blobs by user-specified criteria — Find Blobs, Visualize Blobs, and CLIC tool are described below.

Local Blob Detection

Automated blob detection sometimes fails to correctly detect blobs. For example, detection may fail to indicate a blob or detection may split what should be a single blob or detection may join what should be two separate blobs. If blob detection results in many errors, it is best to undo blob detection and try different detection parameters. If only a few blobs are detected incorrectly, it is possible to perform local blob detection on a sub-region or a subset of blobs using different detection parameters. (Alternatively, as described in the next section, a user can manually redraw incorrectly detected blobs.)

Local blob detection may be performed either in Graphics mode with one graphic selected or in Blob mode with one unexcluded blob selected. Select "Edit -> Local Blob Detection". Then, in the Local Blob Detection dialog, pictured in Figure 6, enter the detection parameters and click "OK". In Graphics mode, all blobs indicated by the graphic are excluded and blob detection is performed on the region. In Blob mode, the selected blob is excluded and local blob detection is performed in its place.

Figure 6: The Local Blob Detection dialog.

Drawing, Excluding, and Editing Blobs

If blob detection fails to correctly detect blobs, a user can manually draw outlines of blobs that were missed, to exclude incorrect blobs (removing them from reporting) and then redraw the outline, or to edit blobs in various other ways. Note that such editing does not change the data, only blob detection.

One of the blob cursor modes is Draw Blob mode. In draw blob mode, the user delineates the blob outline in the Viewport by depressing the left mouse-button on the border of the new blob, moving the cursor along the desired outline of the blob, and releasing the cursor at the end of the outline (causing the outline to be closed and all unassigned interior pixels to be assigned to the new blob). If a new blob is drawn to overlap another blob, the pixels in the overlapping region(s) will not be assigned to the new blob unless the blob that is overlapped has been flagged as excluded from reporting or the existing blob(s) percentage intensity at the pixel is less than 100%.

To flag blob(s) as excluded, select the blob(s) to be excluded and then click the Delete / Exclude Object button on the palette. Confirmation for excluding blob(s) is required. A blob that is flagged as excluded will be treated in reporting as if it was not detected, but excluded blobs are maintained in the data, so it is possible to remove the exclusion flag or to examine the data for the presence and history of excluded blobs. Blobs flagged as excluded lose their claim to pixels. For example, suppose that Blob A is flagged as excluded and then Blob B is defined to contain pixels that were contained in Blob A. Then, if the exclusion flag is removed from Blob A, Blob A does not reclaim pixels that now are contained in Blob B.

To facilitate scripting in which blobs are excluded and then redrawn, LC Image supports exclusion of blobs selected by a graphic. So, a user can define a graphic (with optional CLIC function), then select "Edit -> Exclude Blobs in Shape". This operation can be applied without selecting specific blobs, which may be different from image to image, which allows the script to be used without reference to specific blobs.

The Edit Blobs tool, available in the Edit menu in the Image Viewer, is a specialized interface for drawing, excluding, and/or editing blobs. Figure 7 illustrates the Edit Blobs tool.

Figure 7: The Edit Blobs tool allows a variety of actions, including excluding and redrawing blobs.

The first step in using the Edit Blobs tool is to select a blob in the Image Viewer (as described above using one of the blob-selection modes from the palette). Then, select the Edit -> Edit Blobs option from the menu in the Image Viewer .

The Edit Blobs tool resembles the Multi-Projections tool in that it has a sub-image in the upper-left, graphs of slices of rows in the bottom-left, graphs of slices of columns in the upper-right, and a control pane in the lower-right. The sub-image can be shifted by click-and-drag in the sub-image. The Edit Blobs tool can be resized to show a larger or smaller sub-image and the region of interest can be changed by click-and-drag with the right mouse-button. The Reset Display resets the scale to the minimum resolution and the position to center the active blob (or, if no blob is active, the original active blob).

The Edit Blobs tool shows three row and three column slices in one-dimensional graphs below and to the right of the image. The cross-hairs indicate the three row and three column slices in the sub-image that are graphed. The center of the cross-hairs selects the active blob. (If there are overlapping blobs at the cross-hairs, the <TAB> key can be used to cycle the active blob.) The cross-hairs can be relocated in the sub-image by depressing the shift key and clicking the left mouse-button on the desired location. The cross-hairs can be moved incrementally with the four arrow buttons on the Edit Blobs control pane.

The graphs of the one-dimensional slices show the pixel values (in blue), the values of the active blob (in red, with knots), and the values of the other blobs (in black). The values of the active blob can be edited on the row and column slices by moving the graphical knot associated with the blob value up or down. The knots are moved by clicking with the left mouse-button on the knot and dragging it to the desired value. The sum of all blob values at a pixel may not exceed the pixel value.

The Edit Blobs control pane has five additional buttons for editing blobs. There are buttons for splitting the active blob at the center of the cross-hairs either horizontally or vertically along the cross-hair. Pressing these buttons will split the active blob in the specified direction. There is a button for excluding the active blob at the center of the cross-hairs. Pressing this button will flag the active blob as excluded in reporting. There is a button for merging two blobs. After pressing the button to merge blobs, the active blob at the center of the cross-hairs is merged with another blob. The other blob is indicated by pointing the cursor and shift-click with the left mouse-button, then clicking the merge button again to indicate the selection of the other blob is complete. There is a button for drawing a new blob. After pressing this button, locate the cursor on the desired border of the new blob, depress the left mouse-button, move the cursor along the desired outline of the blob (drawing an outline), and release the cursor at the end of the outline (which is closed automatically).

Deconvolve/Unmix Blobs

Although LCxLC has significantly greater separation capacity than conventional liquid chromatography, analytes still may co-elute creating blobs consisting of overlapping peaks associated with two or more compounds. LC Image has a Deconvolve/Unmix tool that can separate overlapping peaks. Like local blob detection, Deconvolve/Unmix has the ability to change how the minimum peak reference is evaluated. Deconvolution refers to processing single channel data (e.g., FID) and unmixing refers to processing for multi-channel data (e.g., MS). The process is the same for both.

1. The first step is to select one or more blobs in a small region that are presumed to contain two or more chemical peaks.
2. Next, choose the Edit -> Deconvolve/Unmix option from the menu bar. If the image has a single channel, then the peaks are extracted from the blobs based on a separable peak model with shift-variant peaks. If the image has multiple channels, then the peaks are extracted based on a tri-linear peak model.
3. When processing completes, a popup dialog asks the user to okay or cancel the operation. The dialog presents an image of the selected blobs and separate images of each extracted peak.

At pixels with overlapping blob peaks, the pixel value will be assigned on a percentage basis to each overlapping peak (with the total percentages no greater than 100%). The Text View tool and the Edit Blobs tools have facilities for presenting the relative assignments of pixel value to multiple peaks.

Separating overlapping peaks is a difficult problem and deconvolution and unmixing processing will not succeed in all cases. The operation works better if:

• the overlapping peaks are somewhat separated (e.g., two overlapping peaks detected as two adjacent blobs);
• each peak extends over several rows and columns of the data;
• each secondary chromatogram of a peak differs principally in scale (and does not change in shape or retention time); and
• for multispectral data, each peak has a discernibly different spectral signature.

Contents Previous: Preprocessing Next: Blob Analysis

GC Image (LCxLC Edition)™ Users' Guide © 2001–2011 by GC Image, LLC, and the University of Nebraska.