Bahram Khoobehi and James M. Beach

Fig. 4.15. Image of the primate ONH showing analysis regions for vessels (artery and vein segments) and areas of the ONH that include from the rim: superior (S), inferior (I), temporal (T), and nasal (N) areas; and from the cup: temporal and nasal areas, as marked in the figure. Image is oriented with temporal aspect to the left and superior aspect to the top. Vessel types are identified. See Sec. 4.3.1. Reprinted with permission from Beach, J., Ning, J., and Khoobehi, B. Oxygen saturation in optic nerve head structures by hyperspectral image analysis. Curr Eye Res 32:161–70, 2007. ©2007 Informa Medical and Pharmaceutical Science.

of the ONH tissue (combinations of cyan, yellow, and red). Arteries are dark red. Saturation is, thus, lowest in the vein, intermediate in the ONH, and highest in the artery. At 55 mmHg, the order of saturation is the same; however, desaturation of structures is evident: arteries (yellow), ONH tissue (cyan-blue), and veins (deep blue, with more of the vein structure visible). Yellow-red codes in the temporal cup area indicate that this area has a relatively higher saturation at the high pressure. In all maps, stray light reflected from the ONH causes the reflectance from the disk surround to differ from the disk interior, hence, color codes on and off the disk cannot be directly compared.

4.3.3. Discussion

The current study is the first, to our knowledge, to report blood oxygen saturation in the ONH structures and map its distribution. At 55 mmHg, the

154

Hyperspectral Image Analysis for Oxygen Saturation Automated Localization of the Eye

Table 4.4. Percentage saturation of vessels and optic nerve at low and high IOP.

aMean ± SE.

bPercentage deviation about the average optic nerve head (ONH) saturation at 10 mmHg.

cN = 336 at 10 mmHg, 120 at 55 mmHg. Saturations at 55 mmHg differed significantly from those at 10 mmHg in all structures (p < 0.05).

Note: Reprinted with permission from Beach, J., Ning, J., and Khoobehi, B. Oxygen saturation in optic nerve head structures by hyperspectral image analysis. Curr Eye Res 32:161–70, 2007. ©2007 Informa Medical and Pharmaceutical Science.

pressure on the disk and vessels is high enough to occlude the blood flow into the disk and retina partially. At this pressure elevation, lowered saturation in the disk tissue and the veins at high IOP could result from decreased blood flow, which results in a greater desaturation of the blood in the presence of a fixed oxygen consumption in tissue. The lowered saturation in arteries was not expected and is not well understood. A possible mechanism, which could reduce the oxygen carried in retinal arteries that are visible on the disk, is oxygen leakage from the central RA. The proximity of the large central artery and vein in the sheath of the optic nerve, behind the eye, may play a role in the effect of high IOP on the retinal arterial SO2.

Our method evaluates areas under spectral curves with respect to a baseline that does not change with saturation. The advantage of using areas is noise reduction, as significant additive noise appears on spectral curves from

155

Bahram Khoobehi and James M. Beach

Fig. 4.16. Plots of the OSC of the saturation algorithm (left panel), and the RSI after blood-volume correction (RSIv, right panel), against percentage saturation values found from curve fits. For IOP of 10, 30, 45, and 55 mmHg, goodness of fits to lines were respectively (OSC) 0.204, 0.362, 0.741, and 0.639; (RSIv) 0.966, 0.983, 0.883, and 0.743. See Sec. 4.3.2. Reprinted with permission from Beach, J., Ning, J., and Khoobehi, B. Oxygen saturation in optic nerve head structures by hyperspectral image analysis. Curr Eye Res 32:161–70, 2007. ©2007 Informa Medical and Pharmaceutical Science.

single pixels. The curves shown in Fig. 4.14 were averaged over several hundred pixels and, thus, are virtually noise free. However, the map algorithm operates on a single pixel of the image. Between 8 and 10 spectral points are averaged in each band; hence, there is still noise visible in the saturation maps.

The change in color codes at the edge of the ONH is likely due to different optical properties of nerve and retinal tissue. Significant light is reflected back through vessels from a more robust scattering in the nerve. In the retina, structures behind vessels contain pigments and tend to absorb the light before it is scattered into vessels. Our algorithm does not correct for the different effective optical path lengths of ONH and retina.

4.3.4. Conclusions

In animal studies, where the eye can be immobilized, this method should provide new information about oxygen supply and use in the retina and optic disk. Hyperspectral recordings are not yet practical in humans, as periods of 10–30 seconds are required to scan the fundus. Multispectral methods are being developed to reduce the time required to collect imagery from human subjects.

156

Hyperspectral Image Analysis for Oxygen Saturation Automated Localization of the Eye

Fig. 4.17. Saturation maps of the ONH and overlying retinal vessels. Top row: OSC of the RSI. Middle row: BVC. Bottom row: RSI map corrected for blood volume. Left panels: 10 mmHg. Right panels: 55 mmHg. See Sec. 4.3.2. Reprinted with permission from Beach, J., Ning, J., and Khoobehi, B. Oxygen saturation in optic nerve head structures by hyperspectral image analysis. Curr Eye Res 32:161–70, 2007. ©2007 Informa Medical and Pharmaceutical Science.

157