Biomedical Photo-chemistry and Photo-imaging In Vivo

Brian W. Pogue, Ph.D.
Thayer School of Engineering,
Dartmouth College
Tel (603) 646-3861
Fax (603) 646-3856
Email: pogue@dartmouth.edu

Overview

Medical use of light and lasers has evolved into an enormous field of study. Photons can be used to both affect tissue photochemically, as well as acquire diagnostic measurements about the tissue. In basic chemistry, the methods for spectroscopy and photochemical reaction are mature fields of study, which have well established techniques. Photon interaction with molecules in tissue is dominated by elastic scattering off of the intra-cellular compartments of the tissue, so that traditional methods for spectroscopy and photochemical analysis must be altered for use in vivo. Interpretation of light interaction with tissue requires a careful blend of model-based analysis together with new technology to produce and measure light signals, which together can be used to deconvolve the effects of scattering or compensate for them. Some of the basic methods developed for investigating therapeutic tissue photochemistry and imaging of molecular concentrations in vivo are outlined here.

1. In Vivo Photo-chemistry : Photodynamic Therapy for Cancer Treatment

It has been known for some time that light can cause photochemical changes in the body, and especially ultraviolet light can induce acute burns or chronic changes related to DNA damage. In more recent years , medical applications of photochemistry have been applied to medical use. Specifically for cancer treatment, a method called photodynamic therapy (PDT) is now well established, and has very recently received FDA approval for use in treating esophageal cancer , as well as some forms of skin malignancies. This therapy is essentially a light activated type of chemotherapy, where the photo-sensitizer is injected or topically applied to the tissue, and light which is resonant with the absorption band of the drug is applied to the tissue. For some drugs and tissues, there is preferential uptake in the malignant tissue, and then the light can be confined to the region of tissue to be destroyed.

The photochemical pathways responsible for PDT are well established from solution chemistry measurements, as well as from basic cell culture studies. In general, the photosensitizer molecule absorbs light which excites it to the first excited singlet state (see Figure 1), which then has a large probability of intersystem crossing to the triplet state. The sensitizer molecules are chosen to be near resonant with the transition of oxygen from ground to excited singlet state, such that the excited sensitizer molecule is quenched by molecular oxygen with high yield. In solution and in cells, singlet oxygen is efficiently produced and causes cellular death through reaction with membranes and organelles. In living tissue, there is also significant blood vessel occlusion which leads to further destruction by starvation of treated regions.

Figure 1.1 The Jablonski diagnram of energy level pathways for a photosensitizer molecule in the photosensitation process (indicated by dotted lines). The molecules are excited from the ground state, S0, to the excited singlet state, S1, with probability proportional to the product of the absorption coefficient ma and irradiance, y. Once in the S1 state, the molecule can relax by fluorescent photon emission (with quantum yield Ffl) or intersystem cross to the first triplet state, T1 (quantum yield Fisc) From triplet state, the molecule can either relax by phosphorescent photon emission (quantum yield Fph), or be quenched by interaction with a ground state oxygen molecule, O2, to produce singlet state oxygen, O2 (1Dg) (the quantum yield for singlet oxygen generation, FD, is the ratio of O2 (1Dg) molecules produced per photosensitizer molecule excited. Photobleaching of the molecule can come from many paths in this process including directly from S1 or T1 or from S0 , S1, T1 in combination with O2 (1Dg), or from photosensitizer intermolecular interaction, resulting in destruction of the photosensitizer molecule.

In order to better understand and plan PDT treatments, it is important to understand in detail how the singlet oxygen dose is deposited in tissue, and which treatment parameters can be altered to optimize the malignant tissue killing, while potentially limiting the effect in normal tissues. In general, the main controllable parameters are simply the drug concentration and the light dose, yet it is important to examine subtler parameters such as drug biodistribution, localization, aggregation, oxygen supply and consumption and tissue optical properties. It is challenging to directly measure many of these in vivo, and sometimes these parameters are not the same when examined ex vivo or in vitro. The best method of understanding PDT dosimetry is to combine in vitro, ex vivo, in vivo measurements together with computational models of the relevant transport and photochemical processes. Modeling of the dose deposition is an important part of understanding how PDT works in vivo, and can be used to combine pieces of information obtained in separate experiments. Since the heterogeneity of tumors is very high, modeling of the PDT dosimetry must incorporate this type of microscopic complexity.

1.1 Light Dosimetry In Vivo

In most treatment conditions, the total light dose delivered to the surface of the tissue is measured (typical treatments may involve 50 to 250 Joules/cm²), but since the propagation of light in tissue is dependent upon both the tissues absorption and scattering coefficients, then it is important to estimate the fluence within the tissue (Wilson 1989). Measurement of the light dose in vivo can be achieved by sampling the light with interstitially implanted fiber optics, which capture some fraction of the delivered light fluence rate (Stringer, Hudson et al. 1995)(light energy passing through a unit spherical surface, per unit time, i.e. Watts/cm²). As photodynamic therapy becomes used in more types of treatment, accurate measurement of the in vivo dose will become a more important issue, and patient specific dosimetry may become a requirement (Marijnissen, Baas et al. 1993; Wilson, Patterson et al. 1997).

1.2 Photosensitizer Dosimetry In Vivo

The concentration of photosensitizer in vivo should ideally be measured for each patient as well. While the injected dose it generally calculated relative to the patient surface area (i.e. mg/m2), or less accurately to the patients weight (i.e. mg/kg). Typical injected doses are between 0.1 and 5 mg/kg, depending upon the sensitizer and the treatment. Newer drugs have been developed to be applied to malignant regions topically or orally, rather than through injection. One such drug is a mixture of aminolevulinic acid, which causes over production of protoporphyrin IX in rapidly growing cells, and this latter molecule is itself the photosensitizer. However the uptake or production of photosensitizers in vivo is variable, and the exact concentration cannot be known from the injected dose with high accuracy. Early studies indicate that variations up to a factor of 5 can occur for the human esphageal tumors depending upon the malignancy stage and size (Braichotte, Wagnieres et al. 1997).

In vivo measurement of these drugs can be determined either through absorption spectroscopy of the tissue (Farrell 1992), or through quantitative fluorescence spectroscopy (Wilson, Patterson et al. 1997). Fluorescence spectroscopy devices have been developed which can measure signals from tissue, where the signal is not significantly affected by the tissue absorption and scattering coefficients (Pogue and Burke 1998). These new type of tools will also patient specific dosimetry of the photosenstizer uptake, and can be used both on surface tumors as well as an intra-operative device.

1.3 Oxygen Dosimetry In Vivo

PDT can cause rapid depletion of oxygen within tissue, due to the photochemical production of singlet oxygen, which irreversibly reacts with lipids and proteins within its environment. However, measuring oxygen partial pressure within tissue is a difficult task, and in general every measurement system has its own artifacts associated with the measurement. Only by comparing measurements taken with different systems can be fully understand the spatial and temporal changes in oxygen that occur during PDT. In our studies we have examined polarographic electrode measurement, diffuse reflectance flash photolysis, electron paramagnetic resonance oximetry (Pogue, O'Hara et al. 1999), and micro-electrode measurements. These measurements are interpreted with a model-based theory of the oxygen diffusion and consumption effects (Pogue 1997; Pogue, O'Hara et al. 1999).

2. Photo-Imaging In Vivo: Near-infrared Diffuse Tomography of Breast Cancer

2.1 Near-infrared Spectroscopy In vivo

Imaging of photobiologically relevant chromophores in tissue is also now possible, and the means to quantify important metabolites from measurements of diffuse light from tissue has been established within recent years (1994; 1995; 1996). Yet while quantitative tissue spectroscopy is now possible over moderate regions of tissue (1-100 mm), spatially localizing specific spectral changes is somewhat more challenging because of the diffuse path that light travels in tissue. However, it is possible to reconstruct images of the absorption coefficient of tissue from multiple tomographic measurements of light propagation at the surface, assuming that an appropriate model of the propagation process can be used. Beyond this, it is then possible to estimate the concentrations of specific chromophores present within tissue, such as oxy-hemoglobin, deoxy-hemoglobin, water, lipids, cytochromes and potentially exogenous dyes to assess different tissue functions.

Figure 2.1 Spectrum of the dominant absorbing molecules present in breast tissue, weighted to their approximate relative concentrations. The absorption coefficient is the molar extinction coefficient times the concentration, and converted to a natural log exponent.

2.2 Near-infrared Imaging In vivo

Before quantitative spectral imaging can be reliably achieved, it is important to work out a robust method for image reconstruction from projections of diffuse light in tissue, and to ensure that the confounding effects of multiple scattering are adequately separated from the estimated absorption coefficient. In our work, we have made use of frequency-domain (or intensity-modulated) light at frequencies near 100 MHz, to allow simultaneous measurement of the intensity attenuation as well as the temporal phase shift of the signal due to multiple elastic scattering of the photons. See figure 2.2 for an illustration of this process. Since the multiple elastic scattering dominates the travel of photons in tissue, their propagation is best described as a diffusive process, and indeed many researchers have now demonstrated that the relative light propagation in tissue can be predicted by the diffusion equation.

-Ñ×DÑF(r,w) + (iw/c + ma)F(r,w) = S₀(r,w)

where D is the diffusion coefficient, m_a is the absorption coefficient, w is the frequency of light modulation, c is the speed of light in tissue, F(r,w) is the fluence rate of light in the tissue at any point r, and S₀(r,w) is the light source term. Note that D is related to the bulk transport scattering coefficient, m_s^/, by the equation D = 1/(3m_s^/). Considerable further discussion can be found in the following references (Paulsen 1995; Pogue 1995; Pogue, Testorf et al. 1997; McBride, Pogue et al. 1999; Pogue, McBride et al. 1999).

(A)	(B)
Figure 2.2 Illustrations demonstrating that when light intensity is modulated, intensity waves propagate through the tissue. In (A), a short intensity pulse (picoseconds) travels through tissue, and is delayed and widened by the effects of multiple elastic scattering of the photons. When cyclically modulated light is used (B) at frequencies above 50 MHz, the phase shift of the signal is an indirect measure of the pathlength that the photons have taken through the tissue. The phase shift is approximately related to the average time of propagation through the tissue, and thus allows an indirect estimation of the total absorption due to Beer’s law type attenuation.

(A)	(B)
Figure 2.3 (A) Pictorial representation of the light fluence (logarithm of light fluence) in a circular tissue volume, where the light source is positioned in the upper right corner, here red is high light fluence (100 at the highest), and blue is low fluence (10–15 at the lowest). In practice, 16 source locations are used in our imaging system, but only one image is shown here for simplicity. In (B) the fluence is perturbed by a small round object, denoted by a circular region of black dots in the image. This object has an increased absorption coefficient relative to the background by 5X. The dimension of the total region is 86 mm diameter, and the simulated optical properties were absorption coefficient of 0.01 mm-1 and transport scattering coefficient of 1.0 mm-1.

Using multiple light source positions to measure diffuse light patterns in tissue (as shown in Figure 2.3), it is possible to consider computed tomography of the interior, however since the diffusion equation cannot be solved for inexact or arbitrary boundary conditions, the inverse problem must be solved with non-linear optimization methods. We use an iterative Newton-Raphson method, which is a stable and robust means to recover the image from a set of forward propagation calculations. Using measurements of light propagation through the tissue (see Figure 2.3), we begin our calculation by solving for the least square difference between our measurements and our calculated signal from the diffusion equation, based upon some assumed optical property distribution (which in practice is always assumed to be homogeneous at the average tissue optical properties). Assuming that the measurements of phase and amplitude are represented by the set, ym, and the calculated values are F(D,m_a) then our objective is to minimize the following function:

c² = (y_m - F)^T R² (y_m - F)

where R² is a diagonal matrix with the square variances of each measurement on the diagonal. Minimizing this diagonal is achieved by taking the derivative and equating it to zero, resulting in the following:

0 » J^T R² (y_m - F)

where J = F^/ is the derivative matrix (often referred to as the Jacobian matrix), where each element is given by ¶F_i ¤¶D_j for measurement i, and image pixel j, or for the derivative with respect to D, the elements are ¶F_i ¤¶D_j. This Jacobian matrix provides an estimate of how much each property value at each point in the image effects the detected fluence rate value, and these calculations are pictorially demonstrated in Figure 2.4 below. In the next few equations we will eliminate the derivative with respect to D, in order to simplify the derivation, however the same formalism can be applied to calculating images of D. Since the equality is only approximate in the last equation, and the true solution must be derived iteratively, we can expand this latter equation in a Taylor’s series, about the initial set of optical properties m_a^k, such that it becomes:

0 = J^T(m_a^k) R² (y_m - F(m_a^k) ] - Dm_a^k [J^T(m_a^k) R² J(m_a^k) ] + … . (Higher order terms negligible).

Figure 2.4 Calculation of the predicted diffuse photon paths from a single source location (upper right) to each of 16 detector locations, which are equally spaced around the periphery, going from immediately beside the source (top left image), to across the circle from the source (bottom left image), and back to the source on the opposite side (bottom right image). The path is calculated as the logarithm of the intensity for illustration purposes. Intensity images would not show any detail in the interior because the intensity drops by orders of magnitude as the distance increases from either the source or detector. These calculated paths are used in the image reconstruction to provide the sensitivity map of photon path, contained mathematically in the Jacobian matrix. The same calculations must be repeated for all 16 source locations.

Therefore, if an initial optical property distribution can be estimated m_a^k, then the update to this distribution at each iteration, k, can be calculated by solving the following equation for Dm_a^k :

[J^T(m_a^k) R² J(m_a^k) ] Dm_a^k = J^T(m_a^k) R² y_m - F(m_a^k)]

Unfortunately, inversion of the matrix on the left hand side of this equation is often ill-posed in these non-linear problems, so that it must be solved in an approximate manner, by regularization of the matrix. In our work we use a modified Tikhonov regularization method, where some minimal scalar value is added to the diagonal (i.e. a scalar value, l, times the identity matrix, I) to provide stability in the inversion of the matrix, while minimally altering the inversion:

Dm_a^k = [J^T(m_a^k) R² J(m_a^k) + lI]-1 J^T(m_a^k) R² (y_m - F(m_a^k)]

Thus, our image is initially estimated by the optical distribution given by m_a^k, and at each iteration, it is updated as m_a^k+1 = m_a^k + Dm_a^k. This iteration process is continuously repeated until a minimum of c² is reached or when some predetermined number of iterations is reached.

2.3 Experimental Testing and Calibration

Some test reconstructions of tissue-simulating phantom objects are shown in Figure 2.5 below. Tissue phantom testing has been completed and in vivo testing is underway. The current clinical tests indicate that breast abnormalities can be detected and imaged, and preliminary data suggests that there is a significant blood volume difference between tumor tissue and background normal breast tissues. Further testing and spectral analysis on these patients is ongoing.

(A)	(B)	(C)
(D)		(E)
Figure 2.5 Images of a test phantom used to calibrate the system for hemoglobin concentration and oxygen saturation imaging. (A) shows the locations of two test objects placed within a homogeneous solution of dilute lipid emulsion, where the left object has a 4x increase in blood volume, and the right object has the same blood volume but is de-oxygenated. The reconstructed images are shown in (B) to (E), where (B) is the absorption coefficient (mm-1) at 750 nm, (C) is at 800nm, (D) is the hemoglobin concentration image (m) derived from (B) and (C), and (E) is the oxygen saturation (%) image. Troy McBride, a Ph.D. candidate at Dartmouth, is gratefully acknowledged for producing these images in our lab (McBride, Pogue et al. 1999).

2.4 Clinical Development

At the present time, our tomography research is funded through Dr. Keith Paulsen’s Breast Imaging Program Project to examine the potential of near-infrared optical spectral tomography to diagnose breast abnormalities. This is a multi-disciplinary project involving collaboration with the Departments of Radiology and Pathology at the Dartmouth-Hitchcock Medical Center, and the Biostatistics Resource at the Norris Cotton Cancer Center. More details of this project can be found at our web sites (http://www-nml.dartmouth.edu/biomedprg/biomed.html#breast and http://www-nml.dartmouth.edu/biomedprg/NIR/index.html).

(A)	(B)
Figure 2.6 (A) The clinical implementation of the imaging system where the source fibers (small 2 mm plastic fiber) and the detection fiber bundles (6 mm black bundles in image) are attached to a central radial translation system to allow circular coupling to diameters of breast tissue between 5 and 10 cm. The entire system can be manually translated vertically to image within different planes along the breast. (B) The patient imaging room is shown in the photograph, containing a bed (at left) where the optical fiber array is placed below a single hole in the table. The rack-mounted instrument panel is shown in center with computer control for all data acquisition, and a laser table is positioned nearby (right) where light can be generated from either an argon-ion-pumped titanium sapphire laser or from individual diode lasers.

Acknowledgements

These research programs involve important collaborations and the leadership of colleagues who have contributed significantly to this work, including Keith Paulsen, Ph.D. who directs the Breast Imaging Program at Dartmouth, Ulf Österberg Ph.D., Troy McBride and Shudong Jiang Ph.D. The photobiology work is a collaborative effort with Julia A. O’Hara Ph.D., P. Jack Hoopes D.V.M. Ph.D., Harold M. Swartz M.D., Tayyaba Hasan Ph.D., Claudia Lee B.Eng., and Gregory C. Burke B.S. This research is funded by the National Cancer Institute through grants RO1CA78734 and PO1CA80139.

References

(1994). Advances in Optical Imaging and Photon Migration. Opt. Soc. Am., Orlando, FL.

(1995). Optical-Thermal response of Laser-Irradiated tissue. New York, Plenum Press.

(1996). Advances in Optical Imaging and Photon Migration. Opt. Soc. Am. Trends in Optics and Photonics, Washington, DC, OSA.

Braichotte, D. R., G. A. Wagnieres, R. Bays, P. Monnier and H. E. van den Bergh (1995). “Clinical pharmacokinetic studies of photofrin by fluorescence spectroscopy in the oral cavity, the esophagus, and the bronchi.” Cancer 75(11): 2768-78.

Dougherty, T. J., C. J. Gomer, B. W. Henderson, G. Jori, D. Kessel, M. Korbelik, J. Moan and Q. Peng (1998). “Photodynamic therapy.” Journal of the National Cancer Institute 90(12): 889-905.

Farrell, T. J., Patterson, M. S., Wilson, B. C. (1992). “A diffusion theory model of spatially resolved, steady-state diffuse reflectance for the noninvasive determination of tissue optical properties.” Med. Phys. 19(4): 879-888.

Hasan, T. and J. A. Parrish (1996). Photodynamic Therapy of Cancer. Cancer Medicine. Baltimore, Williams & Wilkins. XIII: Chapt. 50.

Marijnissen, J. P., P. Baas, J. F. Beek, J. H. van Moll, N. van Zandwijk and W. M. Star (1993). “Pilot study on light dosimetry for endobronchial photodynamic therapy.” Photochemistry & Photobiology 58(1): 92-9.

McBride, T. O., B. W. Pogue, E. Gerety, S. Poplack, U. L. Osterberg and K. D. Paulsen (1999). “Spectroscopic diffuse optical tomography for quantitatively assessing hemoglobin concentration and oxygenation in tissue.” Appl. Opt. 38(25): 5480-90.

Paulsen, K. D., and Jiang H. (1995). “Spatially varying optical property reconstruction using a finite element diffusion equation approximation.” Med. Phys. 22(6): 691-701.

Pogue, B. W., Patterson, M. S., Jiang, H., Paulsen, K. D. (1995). “Initial assessment of a simple system for frequency domain diffuse optical tomography.” Phys. Med. Biol. 40: 1709-1729.

Pogue, B. W., M. Testorf, T. McBride, U. Osterberg and K. Paulsen (1997). “Instrumentation and design of a frequency-domain diffuse optical tomography imager for breast cancer detection.” Opt. Express 1(13): 391-403.

Pogue, B. W., Hasan, T. (1997). “A theoretical examination of light fractionation and dose rate effects in photodynamic therapy.” Rad. Res. 147: 551-9.

Pogue, B. W. and G. C. Burke (1998). “Fiber optic bundle design for quantitative fluorescence measurement from tissue.” Appl. Opt. 37(31): 7429-36.

Pogue, B. W., N. Chen, H. C. Luedemann, R. R. W. and T. Hasan (1988). Comparative measurement of photosensitizer triplet yield in vitro. Proceedings of Therapeutic Laser Applications, Orlando FL, OSA.

Pogue, B. W., T. McBride, U. Osterberg and K. Paulsen (1999). “Comparison of imaging geometries for diffuse optical tomography of tissue.” Opt. Exp. 4(8): 270-286.

Pogue, B. W., J. A. O’Hara, K. J. Liu, T. Hasan and H. Swartz (1999). Photodynamic treatment of the RIF-1 tumor with verteporfin with online monitoring of tissue oxygen using electron paramagnetic resonance oximetry. Proceedings of SPIE: Laser Tissue Interaction X, San Jose.

Stringer, M. R., E. J. Hudson and M. A. Smith (1995). “The Calibration of Fiber Optic Probes Used for Light Dosimetry Measurements During Photodynamic Therapy.” Laser Med Sci 10(1): 19-24.

Wilson, B. C. (1989). “Photodynamic therapy: light delivery and dosage for second-generation photosensitizers. [Review].” Ciba Foundation Symposium 146: 60-73.

Wilson, B. C., M. S. Patterson and L. Lilge (1997). “Implicit and explicit dosimetry in photodynamic therapy: a new paradigm.” Lasers Med. Sci. 12: 182-199.

Return to Table of Contents