NEIRetinalRegistration

From MIPAV
Jump to: navigation, search

Screening for low macular pigment (MP) and monitoring supplementation requires robust methods to estimate the optical density of the MP. The method includes the measurements of MP that are performed by analysis of autofluorescence (AF) images and building the MP maps.

The technique, which is implemented in the MIPAV "NEI Build MP maps" plug in, uses AF images with two excitation wavelengths (yellow and blue) to calculate the MP optical density distributions over a central eight fifteen degrees diameter area radius centered on the fovea (in this plugin defined to be the center of mass of macular pigment optical density within the selected VOI, in typical cases of centro-symmetric pigment distributions, this corresponds to the maximum value of pigment that is anatomically typically at the center of the fovea (TBD).

Background

The algorithm takes as an input two groups of retinal images acquired at two different excitation wavelengths, and creates MP maps that show the distribution of macular pigment over the retina. The input images are obtained with a standard Fundus camera for retinal photography, equipped with an autofluorescence modality. Images are captured using different wavelength filters - yellow and blue (defined in Appendix I, TBD) and have a fixed black background. It is important for the user to know the weighted average of the normalized macular pigment absorption coefficient in the excitation.

Typical values for NEI filter sets are provided in Table ## (see Appendix I, TBD)

Outline of the method

  • The plug in takes two directories of retinal images - one with "yellow" images and another one with "blue" images.
  • The images from these directories are then converted to 32-bit.
  • Images are, then, registered to a common reference image. Registration employs the assumption that in all images pixels are in the same physical location on the retina. Therefore, it is possible to measure the amount of macular pigment in the fovea over time, comparing the pigment's absorption over time and how vitamins and other supplements quantitatively affect the retina, in addition to monitoring correlations between macular pigment levels and other microscopic changes in disease state and other local quantities such as lipofuscin levels and drusen.
  • The MP maps are, then, built for registered images using the user defined values for maximum and minimum intensity and relative extinction coefficients.
  • The MP map average and standard deviation images are also created.
  • All output images are converted back to 16-bit.
  • The method also provides the user with the opportunity to analyze the MP map using the user defined VOI(s).

Registration

In order to perform analysis on any retinal images, the pixel intensities on the spatial domain of the image must correlate to the actual physical domain on the participant's eye.

Because of the acquisition process, very often retinal images are non-uniformly illuminated, and therefore exhibit local luminosity and contrast variability. This may seriously affect the algorithm performance and its outcome. To contest the pigment's light absorption, images from the "blue" and "yellow" groups and a reference image are, first, processed to emphasize vascular regions, and then, registered using vascular structures as reference points. To accentuate vascular regions, the algorithm applies gradient magnitude combined with Gaussian (with a scale of 6 by 6) to "blue" and "yellow" images and to the reference image.

The algorithm removes the black border from all gradient magnitude images leaving the eyeballs that will be used for registration. The algorithm registers gradient magnitude images to the reference image (also gradient magnitude). Registration is performed using the AOR algorithm (see MIPAV, Volume 2, Algorithms, "Optimized Automatic Registration 3D") with the following parameters:

  • A rotation angle sampling range is set to -10 to 10 degrees;
  • A coarse angle increment of 3;
  • A fine angle increment of 2;
  • The degrees of freedom are set to 3 (as for Rigid transformation);
  • The correlation ratio cost function is used.

Removing the black border from images

Removing the black border from the GM image: A1 and B1 are the far left and right non border pixels, O is the center pixel

To extract the eyeball part of the image for registration, the algorithm removes the black border from the gradient magnitude (GM) images using the following technique:

  • In the GM image, it finds the most left pixel A with coordinates (0, imageHeight/2);
  • Then, it searches from the pixel A towards the center of the image until it finds the first non background pixel A1 on the image's left side. The search is done based on the hardcoded threshold value;
  • The algorithm finds the most right image pixel B with coordinates (imageWidht, imageHeight/2);
  • It searches from the pixel B towards the center of the image until it finds the first non background pixel B1 on the image's right side;
  • The algorithm connects these points A1 B1 and calculates the center pixel and the diameter;
  • It creates a circular binary mask using the region grow algorithm;
  • It applies the mask to the image removing the border.

Building the MP map

To measure the optical density of the MP, the autofluorescence (AF) method uses two excitation wavelengths (blue and yellow) that are differentially absorbed by the MP, accounting thereby for the non-uniform distribution of lipofuscin in the retinal pigment epithelium (RPE).

The algorithm calculates the distribution of the optical density of MP based on the following theory:

The autofluorescence IF(excitation wavelength, emission wavelength), measured at a given emission wavelength for a given excitation wavelength, for any location F in the field of interest can be derived by following equation:

Equation 1

In equation 1, absorbers other than the MP and fluorophores other than RPE lipofuscin are neglected to simplify the process. Typically, contributions from rhodopsin absorption will confound these measurements. In typical Fundus camera measurements, rhodopsin is bleached within the first flash and so does not affect the measurement. When using varying flashes and sensitivities, it is a good idea to verify that the intensity of the AF image is at steady state, i.e., rhodopsin is bleached.

Note that in practical situations, neither the intensity of light incident on the retina, nor the detection sensitivity is known as they depend upon many factors. Therefore, equation 1 is applied to a reference location P, generally located in the periphery of the field.

NEIPlugInEQ1.jpg

Note that in practical situations, neither the intensity of light incident on the retina, nor the detection sensitivity is known as they depend upon many factors. Therefore, equation 1 is applied to a reference location P, generally located in the periphery of the field.

A ratio of two versions of equation 1 applied for two locations F and P creates the following equation 2:

Equation 2

NEIPlugInEQ2.jpg

A ratio of two versions of equation 2 applied for two locations F and P creates the following equation 3:

Equation 3

NEIPlugInEQ3.jpg

Since the optical density of the MP is generally expressed at its wavelength of maximum absorption which is 460 nm; therefore,

Equation 4

NEIPlugInEQ4.jpg

Where, K is the known extinction coefficient of the MP at the excitation wavelength, relative to the extinction coefficient at 460 nm.

equation 5 after several substitutions, elimination of Io and logarithmic transformation appears as:

Equation 5

NEIPlugInEQ4.jpg

Where,

dDFP(460) is the optical density difference between a field location F and the reference location P, log (FF(Lambda)/ FP(Lambda) is an unknown constant.

In terms of the user defined parameters it can be rewritten as equation 6

Equation 6

MPmap = (1/(epsBlue-epsYellow))*[(log((bMax-bMin)/(bImage-bMin)) - log((yMax - yMin)/(yImage-yMin))]Italic text

Where:

variables epsBlue and epsYellow are used for relative extinction coefficients Kmp for blue and yellow light (in the case of broad band excitation, they are weighted by the spectrum of the excitation light, see Appendix I, TBD);

bMax is the maximum intensity value in the "blue" image and yMax is the maximum intensity value in the "yellow" image;

bMin is the minimum intensity value in the "blue" image and yMin is the minimum intensity value in the "yellow" image.

Analyzing MP maps

The plug in analyzes grey scale values in the resulting MP density maps using the following procedure:

  • It takes the used defined circular VOI and creates the concentric circles inside that VOI. The circles are created using the user defined value for degrees/pixels;
  • For each pair of circles, it defines the image area and calculates the average intensity for that area;
  • It outputs the result into a file in Tab-delimited format.

Image types

You can apply this algorithm particularly to 2D retinal images.

Running the NEI Retinal Registration plug in

The plug in's input and output catalogs: given the Blue and Yellow as the input catalogs, the plug in created Output1, BlueReg1 and YellowReg1 catalogs after the first run and Output2, BlueReg2 and YellowReg2 after the second run
(a) - the registered image, (b) - the MP map Yellow1 to Blue1 image, (c) - the MP map average image, (d) - the MP map standard deviation image, and (e) the Output directory

To run the plug in, complete the following steps:

  • Install the plug in;
  • Run MIPAV;
  • In the MIPAV main menu, navigate to PlugIns > General > NEIRetinalRegistration;
  • The NEIRetinalRegistration dialog box appears. The dialog box has to tabs - Create MPmaps and Analyze MPmaps;
  • In the Create MPmaps tab:
    • Specify the directory where "yellow" images are stored;
    • Specify the directory where "blue" images are stored;
    • Specify the reference image;
  • Enter values for the following parameters:
    • epsBlue and epsYellow, the values should be in a range (0,1);
    • bMax, which is the maximum intensity value in the "blue" image;
    • yMax, which is the maximum intensity value in the "yellow" image;
    • bMin, which is the minimum intensity value in the "blue" image;
    • yMin, which is the minimum intensity value in the "yellow" image;
  • Specify the value type for entered parameters - select either Percentile or Intensity;
  • Press OK.

The algorithm begins to run and the list of steps performed appears in the algorithm's window. When the algorithm finishes running, the following new catalogs appear in the images directory:

  • BlueRegN that contains the registered "blue" images;
  • YellowRegN that contains the registered "yellow" images;
  • OutputN that contains the MP maps for each pair of "yellow-blue" images and also MPmapAverage and MPmapStDev images.

Where "N" is the number of the plugin's run.

Examples

If you are running the plug in for the first time, it will create the following directories: BlueReg1, YellowReg1, and Output1;

If you are running the plug in for the second time over the same set of images, it will create the following directories: BlueReg2, YellowReg2, and Output2.

Creating MPmaps tab

Create MP maps tab

The tab options are as follows:

Yellow directory - use this box to select the directory where "yellow" images are stored;

Blue directory - use this box to select the directory where "blue" images are stored;

Reference image - use this box to select the reference image for registration;

epsBlue and epsYellow are parameters that used for relative extinction coefficients Kmp for blue and yellow light, refer to Building the MP map. The entered values should be in the range (0, 1). For the first run, use 0.007 for epsYellow and 0.815 for epsBlue (see Appendix I, TBD);

bMax is the maximum intensity value in the "blue" image;

yMax is the maximum intensity value in the "yellow" image;

bMin is the minimum intensity value in the "blue" image;

yMin is the minimum intensity value in the "yellow" image;

Don't register the image - if this option is selected, the plug in doesn't run the registration algorithm;

Make concatenated image - if selected, the concatenated image is displayed;

Min/Max value type (percentile or intensity) - use this list box to select the type of values for yMin, yMax, bMin, and bMax.

To run the MP map analysis algorithm, do the following:

Before running the MP map analysis algorithm, do the following:

  • Run MIPAV.
  • Open an MP map file of interest;
  • Delineate a VOI;
  • Save the VOI.
  • In the MIPAV main menu, navigate to PlugIns > General > NEIRetinalRegistration;
  • The NEIRetinalRegistration dialog box appears;
  • Open the Analyze MPmaps tab;
  • In the Analyze MPmaps tab:
    • In the mpMap File box, select the MP map file you would like to analyze;
    • In the VOI File box, select the previously saved VOI file;
    • Specify the Degrees/Pixel value;
    • Press OK.

The algorithm begins to run and the steps performed appear in the plug in's output window. When the algorithm finishes running, the result file appears in the OutputN directory.

Analyze mpMap tab

Analyze mpMap tab

The tab options are as follows, see Figure on your right:

  • mpMap file - use this box to select the MP map file. The file is in TIFF format;
  • VOI File - use this box to select the VOI file. The VOI file should have an extension xml or voi;
  • Degrees/Pixel - this parameter represents angular pixel density in terms of pixels per degree.

Any constrains on this value?

TBD.

References

Delori F.C., Autofluorescence method to measure macular pigment optical densities fluorometry and autofluorescence imaging, Arch. Biochem. Biophys., 430(2), (2004), pp. 156-162.

Trieschmann M., Heimes B., Hense H. W., Pauleikhoff D., Macular pigment optical density measurement in autofluorescence imaging: comparison of one- and two-wavelength methods, Graefe's Arch. Clin. Exp. Ophthalmol., (2006) 244, pp. 1565-1574.

See also: