Table 1.
Demographics of the experimental groups.
Fig 1.
Examples of MRIA maps showing the magnitude and distribution of the macular pigment in subjects from group 1 (left), group 2 (centre) and group 3 (right). Top row: original MRIA maps of MP with brightness proportional to MP level; middle row: visualisation of MP as contour maps, derived from the original MRIA maps; bottom row: visualisation of MP as 3D elevation maps. The bar on the right shows the MP concentration (A.U.). The maps represent a region with eccentricity of 4 degrees centred on the fovea. All the MP maps are publicly available at the University of Birmingham eData Repository, “Macular pigment maps and groups”, https://edata.bham.ac.uk/153/.
Fig 2.
Visualisations of Zernike polynomials.
Top: 2D visualisation of the first fifteen ZP on the unit circle. The numbers underneath the plots are ANSI sequential indices, see Eq (4). Bottom: 3D visualisations of ZP with indices 1, 3 and 14, representative of asymmetry, magnitude and irregularity of the periphery.
Table 2.
Performance of the optimisation algorithms.
Fig 3.
ROC plots of the results of the classifier selection.
Table 3.
Classification using PRNN classifier with LOO selection.
Table 4.
Classification using LDA classifier with LOO selection on the ranked coefficients.
Table 5.
First 15 ZP coefficients ranked highest according to their significance using a feature selection algorithm based on a two-way t-test.
Fig 4.
Mean MP distribution maps visualised as 3D elevations and 2D contours for three subject groups.
(top row) subjects under 50 without AMD; (middle row) subjects 50 and over without AMD; (bottom row) subjects 50 and over with AMD. The maps were generated from Eq (5) using mean ZP coefficient values for each of the groups on the basis of classification results from the LOO experiments.
Fig 5.
Fitting ZP to 2W-AF MP density maps.
Left to right: 3D representation of a 2W-AF image; 3D representation of the image data after Gaussian smoothing (σ = 4); 3D representation of the ZP fit to the original 2W-AF image; and 2D profiles of the unprocessed 2W-AF data (black), Gaussian smoothed 2W-AF data (blue) and ZP fit (red).