Phantom Imaging

Imaging an object with known optical properties allows a comparison between the reflectance spectra predicted by MCML modelling and the reflectance spectra evaluated by the multispectral imaging system.

This section details the Monte Carlo modelling of an optical phantom (INO biomimic optical phantoms), in comparison with the reflectance spectra recovered from the multispectral imaging process. The homogenous, cuboidal, phantom (Figure 38c) had known reduced scatter (μs’) and absorption coefficients (μa) for the wavelength range

590nm to 1000nm, recovered from a geometry-specific fitting using time-resolved diffuse optical imaging data (181). Figure 38a displays the optical properties of the phantom. For the MCML program input, the μs’ values were used for μs with an

anisotropy (g) value of 0, and the μa values were used directly, together with a depth

value of 25mm. The reflectance data from the multispectral imaging system were calibrated as above with imaging wavelengths from 480nm to 999nm in steps of 10nm, and an average reflectance spectrum was computed from 30 randomly selected areas of the image. Figure 38b displays the MCML modelled reflectance spectra with the average of the measured spectra. The experimental set-up was designed so that the surface of the block was at the same imaging distance as the corresponding image of the reflectance standard.

Figure 38b displays the recovered spectral reflectance of the optical phantom as compared to the reflectance spectra modelled using the MCML Monte Carlo method. There is a clear resemblance in the shape of the spectra, showing that the effects of the phantom’s optical properties are broadly recognisable in the reflectance spectra detected by the multispectral imaging system. However, the error between the modelled spectra and the imaged spectra is maximum at 4.7% for 750nm.

Figure 38: Spectral data from multispectral imaging and Monte Carlo modelling of an INO Biomimic optical phantom. a) Absorption and reduced scatter coefficients for INO Biomimic optical phantom sourced from reference (181). b) Comparison of the MCML modelled reflectance spectrum for the optical phantom vs the reflectance spectrum recovered from multispectral imaging. c) Photo of the INO Biomimic optical phantom.

Reasons for the difference between the modelled and imaged spectra is likely due to the difference between the geometry of the image in reality vs the point-source, semi- infinite set-up assumed in the MCML model. In addition, the MCML model assumes single wavelength data whereas the imaging system images across the bandwidth imposed by the use of the Varispec filters. Also, approximately 2% of the reflectance calculated in MCML is due to specular reflectance, which theoretically is not detected by the imaging system.

5.5 Discussion

5.5.1 Discussion of Results

The multispectral imaging system detailed above offers the ability to take multispectral data in both a reflectance and transmittance geometry at a flexible series of

wavelengths, across the visible and near infrared spectrum, for relatively small objects. In this case, the design and layout of the stage is especially suited to the imaging of the hind limbs of small animals, and the implementation of imaging meets the demands of the application. Comparing this system to commercially available imaging systems is apposite, although indirect, due to considerable variation in the method of data gathering in commercial multispectral and hyperspectral imaging systems. Relatively cheap, fast, solutions for multispectral imaging exist in the form of multispectral cameras with a small number of bandwidths fixed by filters or LED illumination. The advantage of these kinds of systems is the potential for fast data acquisition, ease of use, and low cost. The disadvantage is the inflexibility and low spectral resolution. However, this may represent a potential future avenue for this work if a small number of optimal imaging wavelengths could be found to work effectively in diagnosing symptom severity in mouse models of murine arthritis. The system has comparatively lower spectral resolution than other commercially available hyperspectral imaging systems such as push broom cameras, but such cameras are generally expensive, they rely on moving the object relative to the camera and require time to capture a full image.

5.5.2 Limitations

There are several caveats of the imaging system that influence the recovered reflectance and transmittance data. In reflectance imaging, due to the method of calibration used, it is assumed that the object of interest is flat and is imaged at the same distance as the reflectance standard. Given that intensity has an inverse proportional relationship with distance, if the surface height of the object varies, then

the calculated reflectance of the object may vary linearly in wavelength according to equation 16

I ∝ or = (16)

where I is intensity and d distance. Similarly, the angle of the imaged surface compared to the reflectance standard will also affect the calculated percentage of reflectance. For a lambertian surface, Lamberts Cosine Law states that the radiant intensity of a surface is directly proportional to the cosine of the angle between the surface normal and the measurement angle (equation 17). The same principle applies in reverse for illuminating a surface at a non-perpendicular angle. In the case of topologically heterogeneous objects such as the mouse hind paw, the incident illumination will not always be normal to the surface and hence detected reflectance will be reduced for uneven areas according to the cosine law

I = Icos (θ) (17) where Io is the detected intensity, I is the input intensity, and θ is the angle between

the direction of measurement and the surface normal.

5.5.3 Conclusion

This chapter describes the development of a multispectral imaging system made specifically for the purpose of imaging mouse paws in a reflectance and transmittance geometry. Combining multispectral reflectance and transmittance (or multispectral imaging with fluorescence imaging) is recognized as a method of increasing data output (182–186), but is still relatively rare in medical multispectral/hyperspectral imaging (187). Tuneable filters were used which are common in imaging stationary

objects and generally faster than point scanning methods (188). Overall, the elements of the design of the system are relatively standard methods used in multispectral imaging, but geared towards the novel function of imaging mouse feet. The following chapter introduces the results of multispectral imaging for mouse models of arthritis.