Cell detection workflow

501 (IDH_CellDetectorLoG) (Stereo Investigator does not have this workflow)

Purpose

 Use the cell detection workflow to automatically detect, count and map labeled cells in images. A marker is placed on each cell detected using this workflow.

How cell detection works

NeuroInfo software uses edge detection to identify roughly circular areas that are significantly brighter (or darker in brightfield images) than their surroundings. This is done using Gaussian smoothing for noise reduction, then finding the Laplacian of the image—this process is known as determining the Laplacian of Gaussian or LoG.

Cell Strength/Strength filter: When you run cell detection in NeuroInfo, the LoG response of each detected object is determined. This is a relative intensity measure; it’s a weighted average of how much brighter the center of the detected object is from its surroundings. The value is normalized across scales; larger objects pull the weighted average from a wider area, but the responses are normalized to be comparable across scales.

  • In Steps 3 and 4 of the workflow you specify the minimum cell "strength" required for NeuroInfo to classify a detected object as a cell.

  • If you don't have NI, will you ever actually see cell-strength data (numbers)? if yes, where? Would they be visible in Neurolucida Explorer?

  • When you Map Experiment to Atlas, the cell-strength data generated are the average LoG strength of cells detected in the brain region.

Cell detection workflow overview

The cell-detection workflow includes the following steps; click a step to jump to detailed instructions.

  1. Image Geometry

  2. Setup

  3. Preview

  4. Filter and Finalize

Commands available in all steps of the workflow

New workflow: Click the new workflow button to start over; the settings will revert to the defaults or those specified in the previous completed workflow.
Previous step / Next Step: Click to advance in the workflow or revisit a previous step. Alternatively, you can click the steps listed at the top of the workflow to jump to that step.

Presets: Click to select a preset and apply those saved settings to the current workflow. Note that you can save your settings as a preset at the end of the workflow.

  • Loading an existing preset: Click Presets and select a preset from the list.

  • Saving or deleting presets: Click Presets and select Edit Presets; the Save/Update dialog box opens. You can:

    • Type a name for the preset, and click Save.
    • Select an existing preset and click Delete to remove it.

Procedure

Before you start

  1. Open an image

  2. Click Detect cells from the Pipelines ribbon in the Main window.

    The Cell Detection workflow window opens.

Cell detection workflow

The numbered steps below reference the steps shown in the Cell Detection workflow window.

  1. Image geometry

    Click the radio button to indicate whether you want to detect cells in 2D or 3D and click the Next Step button:

    • Choose Detect cells in plane(s) for 2D images or to detect cells in a single plane of an image volume.

    • Choose Detect cells in a volume to detect cells in all planes of a 3D image volume. Note that you will jump to step 3 if you choose to detect cells in a volume.

  2. Setup

    Select portions of an image plane or an entire image plane for cell detection:

    Detect in image planes (available for image stacks)—choose one of the following:

    • Click the checkbox to detect cells in the Current plane only
    • Type plane numbers into the boxes or use the arrows to select multiple image planes for cell detection.

    Use image regions (available for 2D images)—choose one of the following:

    • Full image: cells are detected in the entire image

    • Use Region(s) of Interest: cells are detected in regions of interest

      • Click Draw ROIs to define regions of interest in which you want to detect cells;

        1. Trace contours in the Main window.

          If the cursor is difficult to see, increase its size by rolling your mouse wheel.

        2. Click Continue in the left panel when all contours have been traced.

      • Click Select ROIs to select existing contours for automatic cell detection

        1. Click contours to select them. Each contour selected is displayed within a white box.

        2. Click continue in the left panel when complete.

  3. Preview

    Define and fine-tune the settings used for cell detection.

    Click here to go to detailed instructions.

  4. Filter and finalize

    Refine filter settings, choose a marker, and conduct the final cell detection.

    Click here to go to detailed instructions.

 

Aidan: Lots of customers ask me how the cell detection strength value is calculated in our product.

Brian: When you run cell detection, each detected object has its LoG (Lapacian of Gaussian) response encoded as its “diameter”. This is a relative intensity measure--it’s a weighted average of how much brighter the center of the detected object is from its surrounding. The value is normalized across scales--larger objects pull the weighted average from a wider area, but the responses are normalized to be comparable across scales.

Strength - the LoG response above

Map data to atlas strength values are the average LoG strength of objects within the region.

 

Aidan: I also have lots of NI users ask about the cell detection false positive and negative rates. We put some of the info on an NI poster for a conference. It would be cool to add that to the help too.

Classification report:

 precision   recall  f1-score   support

false       1.00      0.99      0.99     69074

true       0.90      0.93      0.92      4536

accuracy                           0.99     73610

macro avg       0.95      0.96      0.96     73610

weighted avg       0.99      0.99      0.99     73610

Confusion matrix:

[[68614   460]

[  317  4219]]