Comparison with manual measurements

Automatic measurements of individual bifurcations were compared with manual measurements for 17 randomly chosen bifurcations from red-free retinal subimages. Original images were collected with the subjects seated throughout the study. The right pupil was dilated using a topical mydriatic (1England). Retinal photographs were taken using a fundal camera with a $30^\circ$ field of view (Kowa FX-50R, Kowa, Tokyo, Japan). For red-free images a green filter was incorporated into the camera to enhance definition of retinal blood vessels. For fluorescein images a fluorescein dye was previously injected intravenously. Ilford FP4 (125 ASA) photographic film (Ilford Imaging UK Ltd., Knutsford, England) was used. Only superior temporal views were analysed. Photographic negatives were digitised using a Nikon 35 $mm$ film scanner (LS-1000, Nikon, Tokyo, Japan). Digitised images were 2800 x 2400 pixels in size, and segmented with an integer interval of $1 \le s \le 25$ pixels. Figure 12 shows some examples of the individual bifurcations; (a - c) are the gray scale red-free subimages and (d - f) are the automatic segmented and measured segments.

Figure 12: Some examples of individual bifurcations in gray scale red-free images. (a - c) are the original subimages and (d - f) are the respective automatic segmented and measured segments. (d) angles are measured along the skeleton lines within a distance from the bifurcation point centre to a fixed circle. Subimages are taken from gray scale images of size $2400 \times 2800$ pixels.

The black areas in the segmented images represent the areas and lengths measured per vessel segment, an average diameter is calculated as $d=area/length$. Bifurcation angles, $\omega$, were measured along the skeleton lines within a distance from the bifurcation centre to a fixed circle of $5R$ where $R$ is the radius of the maximum circle centered on the bifurcation point that fits inside the boundary of the bifurcation, as shown in Figure 12(d).

The skeleton of the vascular tree is obtained from the segmented binary image by a thinning process where pixels are eliminated from the boundaries towards the centre without destroying connectivity in an 8-connected scheme [24]. A pruning process is applied to eliminate short, false spurs, due to small undulations in the vessel boundary.

In the absence of a true measure, we defined a normalised difference between measures $X_a$ and $X_b$ as:

$$\Delta X = \frac{X_a - X_b}{<X>} \mbox{\,\,,\hspace{+10pt} where \,\,} <X>=\frac{1}{2} (X_a+X_b)$$ (9)

$X_a$ corresponds to automatic measurements and $X_b$ to manual [25]. Table I summarises the results.

Table I:

For $d$, automatic were smaller than manually measured diameters, whereas for $\omega$ automatic were larger than manual angles. Since the manual measurements involved the average of 5 diameters measured close to the bifurcation and the angles between straight lines fitted by eye, these significant differences can not be taken as error but only as indications of the variation between different measurement techniques.