1.

Introduction

Head and neck cancer refers to malignancies arising from the mucosal surfaces of the oral cavity, pharynx, nasal cavity, and sinuses. Often these tumors metastasize to lymph nodes in the neck. Several methods for treatment of head and neck cancer are used including surgery, radiation therapy (RT), chemotherapy, and combinations thereof.¹ The vast majority of head and neck cancers are squamous cell carcinomas (SCC) and treatment for this type of cancer, especially when locally advanced, often uses radiation therapy.

Radiation therapy efficacy is known to be dependent on oxygen status.² Therapeutic treatment is less efficacious in patients with poorly vascularized/hypoxic tumors, and it is therefore desirable to identify and target such patients for special treatment.^{1, 3, 4} To date, some correlations between oxygenation status in human solid tumors and tumor response to therapy have been evaluated,⁵ but the mechanisms associated with tumor oxygenation and blood flow variation during chemo-radiation are poorly understood.⁶ Studies including head and neck carcinoma have exhibited an increase of positive response in tumors with high pretreatment oxygenation compared to poorly oxygenated tumors.^{6, 7} However, in these studies some well-oxygenated tumors failed to respond, while some hypoxic tumors responded, possibly due to changes in tumor oxygenation during treatment. One factor that modulates tumor tissue oxygenation is blood flow. Recent magnetic resonance imaging (MRI) and computed tomography (CT) investigations have demonstrated significant blood flow changes during therapy and have suggested that these early blood flow changes may have prognostic value.^{8, 9, 10} Clearly functional assessment of blood oxygenation and flow variation during the early weeks of treatment holds potential for assessment of therapy efficacy/outcome. Moreover, blood flow and oxygenation changes during therapy may enable clinicians to adjust treatment dosage.

Several methods exist for measurement of oxygenation and blood flow. The oxygen-sensitive microelectrode needle method provides a “reference standard” for measurement of tumor oxygenation.^{5, 11} However it is invasive and inconvenient for clinical use.¹² Thus, there remains a need for reliable noninvasive techniques that measure tumor hemodynamic responses. Tumor blood flow measurements are particularly attractive for this application, since blood flow has been correlated with tumor oxygenation.^{2, 13, 14} Blood flow has been measured in clinical studies by several imaging modalities, including positron emission tomography (PET),^{12, 15, 16} dynamic computed tomography (CT),^{10, 17} dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI),^{8, 9, 18} MRI with spin labeling,¹⁹ and ultrasound color Doppler.²⁰ Some of these techniques require contrast agent administration (PET, DCE-MRI) or ionizing radiation (CT); others are surface sensitive (laser Doppler),²¹ and most are difficult to employ routinely with high throughput. The near-infrared diffuse optical methods presented here offer a noninvasive, rapid, portable, and low-cost alternative for repetitive bedside monitoring of tumor therapies.

The concept of noninvasive repetitive blood flow and oxygenation measurements is particularly attractive in the context of recent research on vascular modulating and antiangiogenic agents, which affect the response and sensitivity of tumors to chemotherapy and radiotherapy.^{22, 23, 24} The work of Folkman²⁵ and other investigators, for example, has demonstrated potential therapeutic benefits of targeting tumor vasculature and tumor angiogenesis, and clinical trials of the anti-vascular endothelial growth factor (VEGF) monoclonal antibody bevacizumab have confirmed this new therapeutic paradigm. ^{26, 27, 28, 29} To facilitate clinical translation of agents that target tumor vasculature, an ability to frequently assess tumor vessel blood flow and oxygenation with repetitive measurements is desirable, and potentially might lead to a means for individualized radiation therapy.

Diffuse optical spectroscopy and imaging has very recently emerged as a candidate for tumor therapy monitoring. In a case study, Jakubowski ³⁰ showed that the greatest breast tumor physiological (hemoglobin concentrations, water content, lipid content) changes occur within the first week of neoadjuvant chemotherapy. In a similar vein, combined diffuse optical imaging with ultrasound localization by Zhu ³¹ demonstrated changes in the heterogeneous hemoglobin distribution in breast tumors during chemotherapy, and in a case study with comparison to DCE-MRI, Choe ³² used the diffuse optical imaging technique to quantify optical contrast of breast tumors during chemotherapy.

In this contribution, we use noninvasive diffuse optical methods to investigate tumor responses to chemo-radiation therapy in a new class of patients with head and neck tumors. In contrast to previous work, our instruments concurrently incorporate diffuse correlation spectroscopy (DCS) as well as the more traditional diffuse reflectance spectroscopy (DRS). The DCS methodology permits assessment of tumor blood flow before and during radiation therapy, and the DRS measurements enable quantification of the concentration of tissue chromophores such as oxy- and deoxyhemoglobin. The DCS method detects moving blood cells, and has been successfully employed in animal studies,³³ for example, burn depth estimation in pigs³⁴ and cerebral blood flow in rats.^{35, 36, 70} Very recently, the technique has been applied in human brain,^{37, 38} human muscle functional studies,³⁹ and in breast cancer patients.⁴⁰ Moreover, validation of DCS in some cases has been provided by comparison to the power Doppler ultrasound,⁴¹ laser Doppler,⁴² and arterial spin labeled MRI.⁴³ DRS provides information about oxygen saturation and total hemoglobin concentration and has been widely used in tumor and normal tissue functional studies. ^{44, 45, 46, 47, 48, 49, 50, 51, 52}

The investigation of a limited number of patients in this pilot study reveal that weekly relative blood flow $\left(rBF\right)$ , tissue oxygen saturation $\left(St{\mathrm{O}}_2\right)$ , and total hemoglobin concentration $\left(THC\right)$ kinetics exhibit different patterns for different individuals, including (on average) a significant early increase in $rBF$ followed by a significant decrease in $rBF$ . The averaged $St{\mathrm{O}}_2$ exhibits an increase in the early weeks, while averaged $THC$ tends to decrease continually during therapy.

2.

Methods

2.1.

Basic Principles of Diffuse Photon Correlation Spectroscopy

Near-infrared photons diffuse through thick living tissues.^{33, 53} When diffusing photons scatter from moving blood cells, they experience phase shifts that cause the intensity of detected light on the tissue surface to fluctuate in time. These fluctuations are more rapid for faster moving blood cells. Therefore, one can derive information about tissue blood flow far below the tissue surface from measurements of temporal fluctuations impressed on diffusing light.

Details of the diffuse photon correlation method can be found elsewhere. ^{34, 35, 54, 55, 56} Briefly, the normalized temporal intensity autocorrelation function of the diffused light,

Eq. 1

$$g_2(r,\tau )=\frac{⟨I(r,t)\bullet I(r,t+\tau )⟩}{{⟨I⟩}^2},$$

is measured on the tissue surface. Here, $I(r,t)$ is the diffuse light intensity at position $r$ , and time $t,⟨\dots ⟩$ denotes a time average, and $\tau$ is the autocorrelation time delay. The electric field of the diffusing light $E(r,t)$ is also characterized by a temporal autocorrelation function, $G_1(r,\tau )=⟨E^{*}(r,\tau )E(r,t+\tau )⟩$ . Usually it is derived from measurements of $g_2(r,\tau )$ using the Siegert relation,⁵⁷ $g_2(r,\tau )=1+\beta \mid G_1(r,\tau ){\mid }^2∕⟨I{⟩}^2=1+\beta \mid g_1(r,\tau ){\mid }^2$ ; here $\beta$ is a constant that depends on source and detection experimental parameters such as the number of detected speckles, and $g_1(r,\tau )=G_1(r,\tau )∕⟨I⟩$ is the normalized field correlation function.

The electric field autocorrelation function traveling through dynamic turbid media obeys a diffusion equation.^{34, 54, 58} Data collected in the reflectance geometry are readily analyzed by solving this diffusion equation using, for example, the semi-infinite medium approximation.⁵⁹ The analytical form of the autocorrelation function within the semi-infinite approximation can be obtained from the image source approach following Kienle and Patterson.⁶⁰ In particular, for semi-infinite homogeneous fluctuating turbid medium and for point sources of the form $S\left(r\right)=S_0\delta \left(r\right)$ , the electric field autocorrelation function on the tissue surface is

Eq. 2

$$G_1(r,\tau )=\frac{3{\mu }_s^{\prime }{S}_0}{4\pi }[\frac{\exp (-k{r}_1)}{r_1}-\frac{\exp (-k{r}_2)}{r_2}],$$

where $k^2=3{\mu }_s^{\prime }{\mu }_a+6{\mu }_s^{\prime 2}{k}_o^2\Gamma \tau$ . Here, ${\mu }_a,{\mu }_s^{\prime }$ are the average absorption and scattering coefficients of the underlying medium. These can be obtained from DRS measurements, $k_o$ is the wavenumber of light in the medium, and $r_1\left(r_2\right)$ is the distance between source (image source) and the detector on the surface. Detailed explicit forms of these solutions are given in the Appendix. $\Gamma =\alpha D_B$ characterizes temporal fluctuations in the medium due to scatterer motions such as blood flow. Here, $\alpha$ is a factor representing the probability that a scattering event in tissue is from a moving scatterer such as a red blood cell. It is proportional to tissue blood volume fraction. $D_B$ is an “effective” diffusion coefficient for the blood cells. It should be noted that the effective diffusion coefficient need not (and is generally not) be the “thermal” Brownian motion predicted by Einstein;⁶¹ nonthermal random forces in the vasculature can also give rise to diffusive particle motions. It is assumed herein that measured relative changes of $\alpha D_B$ are proportional to relative changes in tissue blood flow. Larger $\Gamma$ implies faster electric field (or intensity) autocorrelation function temporal decay and higher blood flow (Fig.1 ). A detailed microscopic model relating tissue blood flow to $\Gamma$ is not available; it is the subject of current research. Nevertheless, the proportional relationship has been verified^{41, 42, 43} against a variety of traditional blood flow/perfusion measurement methods in a variety of different physiological contexts. We adopt this relationship as a fundamental assumption in our experimental approach. We note here that another related microscopic interpretation, which takes “bulk” blood displacement into account, explicitly has been proposed recently in the context of laser Doppler flowmetry.^{62, 63}

Fig. 1

Normalized temporal diffuse electric field autocorrelation function $\left[g_1\right(r,\tau )=G_1(r,\tau )∕⟨I⟩]$ data for patient 2 (P-2) before, and 1 week after radiation therapy. The faster temporal decay in the autocorrelation curve corresponds to an increased blood flow.

Using the Nelder-Mead Simplex algorithm in Matlab (Mathwork, Inc.) software, the flow parameter $\Gamma$ and the experimental constant $\beta$ are obtained by minimizing the difference between the predicted analytical form of the autocorrelation function in the reflectance geometry $g_{1,th}(r,{\tau }_i)$ , and the measured autocorrelation function $g_{1,\exp }(r,{\tau }_i)$ , i.e., ${\chi }^2={\sum }_i\left[g_{1,th}\right(r,{\tau }_i)-g_{1,\exp }(r,{\tau }_i){]}^2$ . The exact form of the autocorrelation function depends on measurement geometry as well as on tissue optical properties. Here we report $rBF$ . $rBF$ is a blood flow index defined as the blood flow parameter $\left(\Gamma \right)$ relative to its pretreatment value (in percent units with 100% implying no change). $rBF$ is thus unitless.

2.3.

Instrumentation

2.3.1.

Diffuse correlation spectroscopy instrument

We constructed a portable four-channel system for use in the clinical study [Fig. 2a ]. The instrument used a long coherence length laser (Crysta Laser, Nevada) operating at $785\mathrm{nm}$ , an optical switch (DiCon Fiberoptics, California), four photon-counting fast avalanche photodiodes (Perkin-Elmer, Canada), and a custom built four-channel autocorrelator board (Correlator.com, New Jersey). The source light delivered to the neck was switched between two multimode source fibers. Single-mode detector fibers were used to collect the light. All fibers were inserted into a soft pad so the operator could place this single hand-held probe onto the patient’s neck [Fig. 2b]. Photons transmitted into the neck were collected by the single-mode detector fibers in reflectance. The shortest and largest separations between source and detector fibers were 2 and $3\mathrm{cm}$ , respectively. When our signals were small, we increased the number of fibers at large separations to increase the measurement signal-to-noise (SNR) ratio.⁷⁰ In the fitting process, the largest separation $\left(3\mathrm{cm}\right)$ was used to extract blood flow parameters. Typically the average photon penetration depth into the tissue is one-third to one-half of the source-detector separation;^{71, 72} thus, we believe the signal originates largely from superficial neck tumor nodes.

Fig. 2

(a) Diagram of the flow instrument (only one source fiber and one detector channel is shown for simplicity). The instrument consists of a 785-nm coherent laser, an optical switch, photon-counting avalanche photodiodes (APDs), and an autocorrelator board. The data are stored in a computer for postprocessing. (b) Hand-held optical probe: source and detector fibers are inserted into a soft pad. The maximum source detector separation is 3 cm.

2.3.2.

Diffuse reflectance spectroscopy instrument

A four-channel frequency domain instrument was used in the clinical study [Fig. 3a ]. The details of the instrument are described elsewhere.^{73, 74} Briefly, the instrument uses 690-, 785-, and 830-nm laser diodes (Thorlabs Incorporated, New Jersey), each of which were modulated at $70\mathrm{MHz}$ . Two $1\times 4$ optical switches (DiCon Fiberoptics, California) were used to switch the wavelength and source fiber positions. The light was collected by two avalanche photodetectors (APD, Hamamatsu, C5331-04) and two PMTs (R928, Hamamatsu), which were all coupled onto the tissue surface via 3-mm fiber bundles. After amplifying and filtering, signals from the detectors were mixed with a reference signal in an in-phase and in-quadrature (IQ) demodulator (Mini-Circuits, New York), thus generating the I and Q signal components. After the low-pass filter, the dc components of the I and Q signals were used to calculate the amplitude and phase of the diffuse photon density waves (DPDWs) that passed through the tissue. In the measurements, an optical probe using two detector fiber bundles and four source fibers were employed [Fig. 3b]. Source and detector fibers were arranged such that at least four different source detector separations (1.8, 2.2, 2.6, and $3\mathrm{cm}$ ) were used for each patient to quantify oxygenation parameters with fidelity. To calibrate the unknown source-detector coupling and to normalize the instrument response, the symmetry of the source-detector fibers and an Intralipid solution with known optical properties were used (see the Appendix). The absorbance of the ink was determined and calibrated using an Ocean Optics spectrometer; for scattering, we used the well-known formula for Intralipid from the literature.⁶⁴

Fig. 3

(a) Diagram of the frequency-domain instrument. Laser diodes (LD1,LD2,LD3) were modulated with 70-MHz rf signal. Two $1\times 4$ optical switches were used to switch the colors and source fiber positions. The light was collected by two APDs and two PMTs (only one source, one detector fiber, and one channel of the instrument is shown for simplicity). (Amp=amplifier, BPF=bandpass filter, LPF=low-pass filter, IQ=in-phase and quadrature demodulators, and A/D=analog-to-digital converter.) (b) Optical probe: two detector fiber bundles and four source fibers are arranged with shortest separation of $1.8\mathrm{cm}$ , and longest separation of $3\mathrm{cm}$ . S1, S2, S3, and S4 are source fibers and Dl and D2 are detector fibers.

2.4.

Measurement Protocol

The study protocol was approved by the review board of human subjects of the University of Pennsylvania and informed consent was obtained from all patients. In our measurements, CT and MRI scans provided additional information about the location and size of each tumor. DCS and DRS measurements were carried out consecutively. The protocol (Fig. 4 ) consisted of preradiation measurements as baseline data. Subsequently, weekly measurements were carried out for each individual until his/her treatment was completed. Each patient received daily fractionated irradiation from Monday through Friday, and the optical measurements were completed just before treatment began each week. Patients were concurrently treated with weekly Carboplatin (area under the $\mathrm{curve}=2\mathrm{mg}∕\mathrm{ml}\times \min$ ) and Paclitaxel $(30\mathrm{mg}∕{\mathrm{m}}^2)$ . Daily fractionated radiotherapy was administrated with an intensity-modulated parotid-sparing radiotherapy technique. A simultaneous in-field boost prescription technique was used prescribing 7040 centiGray (cGy) in $220\mathrm{cGy}$ per fraction over 32 fractions to both primary and gross neck disease. The overall treatment time was about 6.4 weeks. Standard response evaluation criteria in solid tumors (RECIST)⁷⁵ response criteria was applied to classify tumor responses. A responder is defined as a patient with no evidence of residual cancer in the neck dissection specimen at the end of treatment. Formal assessment of treatment response was conducted 6 weeks after completing therapy. Postchemoradiotherapy neck dissections were evaluated using pathologic response.

Fig. 4

Treatment and measurement schedule. See Methods in Sec. 2 for details.

The optical measurements were carried out by three different operators to assess the repeatability of the method. Each operator placed the probe onto the neck and arm muscle (for control purposes) three times. The data reported in this work represent an average ( $\pm$ standard error) of the three operator measurements. Placement of the probe and consecutive measurements of both blood flow and oxygenation took $\sim 15\min$ in total, including the different observers. The largest nodal mass was selected for weekly optical measurements. Palpation and measurements with a ruler were the main tools used for identifying tumors during therapy. A trained radiology nurse was present during each measurement to assist in the identification of tumors. Tumor locations were measured with the ruler and systematically noted with respect to the ear and chin of the patients. Diffuse optical measurements placed the probe at the same center location of each tumor, and the measurements were repeated at that particular location. The repeatability error was small $(\sim 5\%)$ . No obvious trends with respect to operators were found. Only the diffuse optical method was available for weekly measurements. Structural images such as from CT were available at pretherapy only.

3.

Results

A total of eight patients were examined weekly. The patient and tumor characteristics are given in Table 1 . Tables 2, 3, 4 summarize data from tumor $\left(t\right)$ and arm muscle $\left(m\right)$ of eight patients (labeled P-1, $\mathrm{P}\text{-}2,\dots \mathrm{P}\text{-}8$ ). Table 2 exhibits $rBF$ [%] at the end of week 1, week 2, week 3, and week 4 of chemo-radiation therapy. The pretreatment value at week 0 was defined as 100% in all patients. Similarly, Tables 3, 4 summarize weekly changes of $St{\mathrm{O}}_2$ [%] and $THC\left[\mathrm{\mu }\mathrm{M}\right]$ , respectively. For patient 2 (P-2), optical measurements were stopped at the end of third week because the tumor was no longer palpable. For patient 3 (P-3), optical measurements could not be acquired after the third week of the therapy because of scheduling difficulties. It is clear from the tables that our individual results varied greatly. This has also been the case in animal experiments¹⁶ and clinical trials² employing radiation therapy; apparently, this variation is only partly a result of methodological factors such as differences in probe handling and positioning on the tissue.

Table 1

Characteristics of patients with head and neck cancer (SCC=squamous cell carcinoma, TNM=tumor, node, metastasis stage76).

Table 2

Weekly relative blood flow changes (rBF[%]) of individual tumor (t) and arm muscle (m) tissue at the end of week 1, 2, 3, and 4 of chemo-radiation therapy. The pretreatment value at week 0 was defined as 100% in all patients. For patient 2 (P-2), optical measurements were stopped at the end of the third week, since the tumor was not palpable anymore. For patient 3 (P-3), optical measurements could not be acquired after the third week of therapy because of scheduling difficulties.

Table 3

Weekly blood oxygen saturation (StO2) [%] during chemo-radiation therapy for both tumor (t) and arm muscle (m) tissue.

Table 4

Weekly total hemoglobin concentration (THC)[μM] during chemo-radiation therapy for both tumor (t) and arm muscle (m) tissue.

A representative optical response of one of the complete responders is given in Fig. 5 (corresponding to P-l). In this case, $rBF$ increased in the early weeks, and a subsequent decrease followed; $St{\mathrm{O}}_2$ exhibited a small decrease with a subsequent small increase at the second week; and THC exhibited a continuous drop-off during therapy. Patient 8 (P-8), a partial responder, was excluded from the statistical analysis, since P-8 exhibited substantially different tumor hemodynamic response during the therapy (Fig. 6 ); in this case $rBF$ exhibited a continual increase, while $St{\mathrm{O}}_2$ and $THC$ also tended to increase over the course of treatment. For this patient, pretherapy CT showed a large necrotic nodal mass measuring $\sim 5\mathrm{cm}$ diam, and the tumor was still relatively large and palpable at the end of the therapy. Postsurgical pathology confirmed the existence of residual tumor, and so the patient was considered to be a partial responder.

Fig. 5

(a) Tumor relative blood flow changes $\left(rBF\right[\%\left]\right)$ during chemo-radiation therapy for P-1. Pretreatment value at week 0 was defined as 100%. Tumor (b) blood oxygen saturation $\left(St{\mathrm{O}}_2\right)$ and (c) total hemoglobin concentration $\left(THC\right)$ during chemo-radiation therapy.

Fig. 6

(a) Tumor relative blood flow changes $\left(rBF\right[\%\left]\right)$ during chemo-radiation therapy for P-8. Pretreatment. value at week 0 was defined as 100%. Tumor (b) blood oxygen saturation $\left(St{\mathrm{O}}_2\right)$ and (c) total hemoglobin concentration $\left(THC\right)$ during chemo-radiation therapy.

3.1.

Average Relative Blood Flow Response

Figure 7 shows the trend of $rBF$ , $St{\mathrm{O}}_2$ , and $THC$ averaged over patients 1 to 7 (P-l, P-2, $\dots$ P-7). A significant $(p=0.0002)$ increase $\left[\right(52.7\pm 9.7)\%]$ was observed in $rBF$ during the first week of the therapy [Fig. 7a]. Our data also showed that tumor blood flow decreased $\left[\right(42.4\pm 7.0)\%,p=0.007]$ during the second week of the therapy and remained low in the third and fourth weeks. The changes measured in the third $(p=0.52)$ and fourth $(p=0.92)$ weeks were not significant. Arm muscle levels had a tendency to decrease in the early weeks, but overall the changes were not statistically significant ( $p=0.54,$ 0.25, 0.30, and 0.19, respectively). Our observations are in reasonable agreement with other studies using different methods. Mantyla ¹⁶ reported absolute blood flow changes in 43 patients (including squamous cell carcinoma of head and neck); a 56% (mean) increase at the end of the first week, and a statistically significant decrease was observed at the end of the second week using the ${}^{133}\mathrm{Xe}$ clearance method. MRI studies have also reported an enhancement in blood flow after the first week of the therapy; ^{8, 9, 18, 19, 77} De Vries⁹ quantified the blood flow changes and found a statistically significant increase of 21% after the first week and 25% after the second week of the radiation therapy in patients with rectal carcinoma.

Fig. 7

(a) Tumor relative blood flow changes $\left(rBF\right[\%\left]\right)$ during chemo-radiation therapy averaged over all patients excluding P-8. Pretreatment value at week 0 was defined as 100% in all patients. (b) Average tumor blood oxygen saturation $\left(St{\mathrm{O}}_2\right)$ during chemo-radiation therapy. (c) Average tumor total hemoglobin concentration $\left(THC\right)$ during chemo-radiation therapy.

The biologic significance of an increase in tumor blood flow is not well understood. It is possible that the increase may improve tumor oxygenation and therefore tumor radiosensitivity.^{1, 2, 8} The mechanism for such a favorable response might be reflected in the observations of Sonveaux ⁷⁸ These investigators concluded that clinically relevant doses of radiation elicit a vascular stress response with increased secretion of tumor endothelial nitric oxide, which in turn can cause vasodilation, increased blood flow, and increased vessel permeability. Alternatively, this early increase in blood flow may reflect a corresponding decrease in the interstitial fluid pressure affecting the tumor vessel distensibility, and consequentially, blood flow.⁷⁹ Preferential damage to a subpopulation of oxygenated cells may lower the interstitial pressure on microvessels within the tumor, thus opening capillaries and increasing tumor blood microcirculation.⁸⁰ This effect can facilitate improved chemotherapy delivery to tumors as has been demonstrated in preclinical xenograft models.^{81, 82, 83}

3.4.

Average ${\mu }_s^{\prime }$ Changes

Weekly mean ${\mu }_s^{\prime }$ for P1-P7 is plotted in Fig. 8 . The results show $\sim 5\%$ variation in ${\mu }_s^{\prime }$ , suggesting that changes in functional parameters are more significant than changes in the structural parameter ${\mu }_s^{\prime }$ .

Fig. 8

Tumor scattering coefficient changes $\left[{\mu }_s^{\prime }\right({\mathrm{cm}}^{-1}\left)\right]$ during chemo-radiation therapy for an average of patients P-1 to P-7.

4.

Discussion

We have demonstrated the feasibility of diffuse optics for chemo-radiation therapy monitoring in head and neck cancer patients. In this section, we outline some of the limitations of the current approach. We also indicate variations in approach that will facilitate future improvement.

In our measurements, it is possible that different observers may have applied different probe-tissue pressure and introduced different probe positioning; both of these effects can induce variations in quantification. These variations were quantified and reported as error bars in the figures. In addition, the head and neck region consists of different anatomical tissues such as muscle and fat, which can vary across patients. Thus the semi-infinite homogeneous medium approximation is unlikely to be exactly valid. For example, our multidistance fitting scheme uses short separations $(\rho =1.8,2.2,2.6\mathrm{cm})$ and one long separation $(\rho =3\mathrm{cm})$ . On the other hand, $rBF$ was extracted using the longest separation $(\rho =3\mathrm{cm})$ only. Therefore, $St{\mathrm{O}}_2$ and $THC$ measurements are likely to be more affected by near surface tissues compared to blood flow measurements. Nevertheless, we have confined to employing the semi-infinite approximation because it simplifies our analysis enormously and enables us to extract trends from our weekly measurements. Better quantification, as well as increased ability to distinguish different tissue structure (tumor, muscle, fat), can be obtained from similar measurements using larger numbers of sources and detectors, and also by image segmentation based on other available anatomical information.^{31, 89} However, in practice the larger number of sources and detectors also introduces some difficulty interfacing the probe to the tissue. From experience we have found that the relatively large probe having many fibers had disadvantages, especially when trying to contact all fibers to the tissue surface with equal pressure. Therefore, a bulkier single probe containing many blood flow and oxygen saturation fibers was avoided. In the future, a better probe design may enable better quantification by preserving good probe-tissue contact.

Additional benefits may be obtained by comparing the tumor to surrounding healthy tissue. Line scanning, as suggested by Jakubowski ,³⁰ across the tumor would be more favorable during therapy monitoring; however, it was not possible in our case due to time constraints. Ultimately, line scanning and/or imaging the whole tumor with rapid data acquisition should generate a richer dataset. Although radiation therapy science has improved with recent technological developments to better optimize beam localization in the tumor, normal tissue damage near the tumor may still exist. In the long term, it would be interesting to coregister radiated volume with the diffused photon path to better discriminate radiation effects on normal and tumor tissues.

Finally, limitations such as uncertainties in tumor boundaries can potentially be eliminated by coregistering the diffuse optical methods with other structural imaging modalities such as hand-held ultrasound³¹ and MRI,^{89, 90} and by correcting for tumor shrinkage. Since CT is available only at pretherapy, adding a structural imaging methodology such as ultrasound into our protocol in the future would enable us to assess tumor size changes weekly, during therapy. Tissue heterogeneity effects may also be investigated with imaging techniques. ^{31, 89}

In this study we have primarily focused on the changes of hemodynamic responses of the tumor as a result of a perturbation, i.e., chemo-radiation therapy. Because assessment of early response could potentially improve treatment outcome, the results we have presented encourage one to focus on early weeks. Indeed, as suggested by Jakubowski, ³⁰ one can carry out more frequent measurements (e.g., on a daily basis) to extract trends within the first week. It might even be interesting to focus on pretreatment conditions, possibly targeting patients for special protocols. In its current form, however, normalization of the blood flow to the first week precludes use of DCS before therapy begins. Extraction of absolute measures of blood flow will require better absolute calibration. In principle, DCS can be calibrated with other techniques at particular physiologic conditions, as was done recently with the MRI spin labeling technique,⁴³ but further work remains.

Our results suggest early clinical tumor response to radiation therapy can be detected and quantified by diffuse optical spectroscopies. The data clearly exhibit significant changes within two weeks of therapy. The early flow changes may be significant in affecting drug delivery efficacy and/or tumor oxygenation during chemo-radiation therapy, and the early tumor oxygenation changes may be related to tumor response. The responses of patients P-1 to P-7 were similar, and different from that of the partial responder patient P-8, but our statistics are not sufficient to draw significant physiological conclusions. Since a primary aim of therapy diagnostics is to predict the response as early as possible, the early blood flow and oxygenation changes observed here suggest the potential utility of daily measurements during the first two weeks of treatment. Due to very low accessibility of most other diagnostics methods, diffuse optical techniques have advantages for daily-based therapy monitoring.

5.

Conclusion

Several techniques such as MRI, CT, and PET have been employed for monitoring tumor therapies, but the desire for a noninvasive, real-time bedside monitoring device makes diffuse optical techniques very attractive for clinical applications. Future clinical applications might also include concurrent use of optical methods with established modalities.^{31, 89, 90}

With possible clinical requirements in mind, we quantify tumor $rBF$ , $St{\mathrm{O}}_2$ , and $THC$ changes noninvasively during chemo-radiation therapy using diffuse optical spectroscopies. These techniques do not require contrast agent administration and are suitable for bedside examinations with rapid data acquisition. Our preliminary data show that patients exhibit significant changes of $rBF$ , $St{\mathrm{O}}_2$ , and $THC$ , even in the first two weeks of treatment. In one patient (P-8), a different trend was observed with a preliminary indication that it coincided with a different treatment outcome. This anecdotal observation should be further studied with better statistics. At this point, more statistics are required for assessment of the prognostic value of these new methods. During treatment, different types of functional parameters may have different sensitivity to the therapy for different individuals. In other words, one parameter alone may not be a good predictor for a particular patient; multiparameter analyses such as those we present potentially enable clinicians to better discriminate and predict responses, and to continuously re-evaluate the treatment plan according to the probabilistic sensitivities of various hemodynamic functional parameters. Such experiments should be a next step.