As shown in Fig 1(a), two standard SLEDs centered at 750nm and 840nm are combined together by a 50/50 fiber coupler to form a broadband source. The measured source spectrum (red) and interference spectrum (blue) is shown in Fig 1(b).

The source spectrum is not Gaussian. Hence, there would be sidelobes in the A-scan signal, which will cause images to blur rather than increasing the image resolution. Fig 1(c) shows the result by directly inverse Fourier transforming the interference signal in Fig 1(b). It is clear that the deconvolution of the spectrum needs to be performed in order to achieve a higher resolution.

Fig.1.(a) FD-CP-OCT setup      

(b) measured source and interference spectrum     

 (c) OCT A-scan without deconvolution

Various deconvolution algorithms were applied to obtain higher resolution in OCT. Here, we used a filter which is designed to minimize the error between the restored and the original signal in the mean squared sense, to enhance the resolution while suppressing the noise. The method is also referred as Wiener deconvolution .

In OCT, the effectiveness of deconvolution algorithm is based on the phase continuity between the interference spectra of the two sources. If the SNR is too low at the valley between sources, this continuity won’t exist, and the two sources are absolutely independent and any deconvolution wouldn’t work well.We quantitatively studied the effectiveness of deconvolution algorithm as a function of ΔΛ and SNR. We assume two sources are centered at 800nmΔΛ/2, and each with the bandwidth (Full Width Half Maximum) of 35nm. In our simulation, the sample is only a reflectance layer with a delay of l’ from the reference surface. In this case, an A-scan OCT signal can be treated as random variable, with mean l’, and its standard derivation is proportional to the axial resolution. α, the ratio of resolution after deconvolution to resolution b, is used as a measure of the effectiveness the deconvolution algorithm. β is defined as ΔΛ/δλ , hence α is a function of β and SNR. Fig 2 shows how α changes under different β and SNR. A certain combination of β and SNR determines a specific α value and corresponds to a pixel in Fig2.

Fig.2. α versus SNR and β

        Fig 4(a)             Fig 4 (b)

Images of cellophane films in Fig 4(a) and (b) are obtained by using single source and dual source respectively for comparison. Fig 4(b) shows an improved OCT image which exhibits a much better axial resolution than the one with the single source. 

Hemoglobin oxygen measurement

The basic experimental setup is described as above. Interference spectra were recorded for OCT image reconstruction and SO2 analysis. This can be expressed as I(k)=η|E0(k)|2∫rs(k,l)cos(2kl)dl, where η is a constant; rs(k, l) is the depth dependant reflectance; E0(k) is the electrical field of source output. For homogenous blood sample, rs(k,l) decays exponentially as a function of the scanning depth, l. The attenuation coefficient of blood sample contains scattering coefficient αs and absorption coefficient αa, which is a function of wavelength as well as a function of SO2.

rs(k, l)= rs0exp{-[αs+αa(k)]l}         αa(k)=SO2αHbO2(k)+[1- SO2] αHb(k)                

For a biological sample such as chicken embryo, αa, varies as tissue composition changing at difference depth, therefore rs(k,l) has a much more complex dependence of l.

Fig. 5. Flow chart of signal processing

R(l’) is the difference of spectrum between short and long wavelength range. In Fig 6(b), in the lower image R map and OCT image overlaped. And the upper curve is obtained by averaging R(l’) along the depth direction.

In our work, R(l’) does not directly provide this ratio, nevertheless the technique is effective in identifying the blood vessels and gives relative value of the SO2 level in tissue.

 Fig 6(a)

Fig 6(b)