1.

Introduction

Brain tumors are the second most common malignancy and the leading cause of death from cancer in children. Yearly, more than 2000 children are diagnosed with a brain tumor in the United States, and 30,000 to 40,000 are diagnosed worldwide.^{1, 2} The three most common pediatric brain tumors are astrocytoma (∼ 42%), medulloblastoma (16.3%), and ependymoma (10.1%).³ Although dysembryoplastic neuroepithelial tumors are benign, congenital tumors, they occur frequently in children and have a high association with seizures. For each of these pathological entities, gross total surgical resection, when feasible, is the optimal initial treatment option.⁴ A strong correlation between the extent of tumor resection and clinical outcomes, such as 10-yr, progression-free survival, has been reported by different groups.^{5, 6, 7, 8, 9}

To achieve a maximal tumor resection, an accurate delineation of the margins of the brain tumor during the planning and surgical stages is required. Similar to the adult population, preoperative tumor diagnosis and volumetric analysis for the pediatric brain tumor population are achieved using advanced imaging modalities, especially magnetic resonance imaging¹⁰ (MRI). For certain tumor types, their physiological characteristics are assessed further by the use of magnetic resonance spectroscopy and positron emission tomography.^{11, 12, 13} Frameless stereotaxy is used commonly to relay this preoperative diagnostic information into a surgical field.^{14, 15} In addition to the fact that preoperative imaging studies are limited in their ability to define the true pathological margins of the brain tumor,¹⁶ the surgical navigation systems may become inaccurate during surgical procedures due to shifting of brain structures.^{17, 18, 19} Ultrasonography is another popular intraoperative means for guiding tumor resection.^{16, 20, 21, 22} Ultrasound, however, is less likely to differentiate tumor from peritumoral edema, which is also hyperechoic.^{20, 21} Intraoperative MRI can also be used to assess tumor resection intraoperatively. However, obtaining multiple MRI scans during tumor resections is impractical. Therefore, newer technologies are being developed continuously to make surgery safer and more efficient.

The concept of using the tissue differentiation ability of optical spectroscopy to guide brain tumor surgery has been explored over the past 15 yr. So far, several researchers have successfully demonstrated the feasibility of using optical spectroscopy to differentiate brain tumors from normal brain tissue in vitro and in vivo.^{23, 24, 25, 26, 27, 28, 29, 30, 31, 32} While the results are promising, all of these studies with one exception were conducted only on adult brain tumor patients. The in vivo applicability of such a technique for pediatric brain tumor patients has not been evaluated previously. Similar to the majority of in vivo tissue diagnostic studies, optical spectral acquisition from the in vivo brain often is achieved using a handheld optical probe. The potential spectral alterations induced by the movements of a handheld probe have not been investigated thoroughly. Here, a small-scale, clinical study was conducted on the children with brain tumors to evaluate the feasibility of using diffuse reflectance spectroscopy to differentiate brain tumors from normal brain tissue. Concurrently, the quality of in vivo diffuse reflectance spectra acquired using a handheld probe was evaluated via repeated spectral acquisition from a single investigated site.

2.

Materials and Methods

3.

Results

A total of 12 patients was enrolled in the clinical trial reported here (Table 1). All studies were conducted successfully, and there were no unexpected complications. Nine of the 12 patients were females, and three were males. Eight patients had tumors in the cerebrum, and four had tumors in the cerebellum. The tumor types studied encompassed the major pediatric brain tumor categories, including pilocytic astrocytoma, ganglioglioma, and medulloblastoma (Table 2). Overall, 32 measurements were performed on the normal sites (i.e., cerebral cortex and cerebellar cortex) and 27 measurements were performed on the tumor sites. Since only a few measurements were obtained from white matter, these measurements were excluded from the statistical analysis.

Table 1

Demographic information of studied patients.

Table 2

Distribution of studied tumor types and investigated sites.

The representative diffuse reflectance spectra from normal and tumorous brain tissues are shown in Fig. 1. The most noticeable difference observed between these spectra is that the intensities from normal brain tissues were greater than the intensities of tumor tissues. It was found that diffuse reflectance spectra from pediatric brain tissues had profile features that were similar to those from adult brain tissues. The dominant feature in these diffuse reflectance spectra is the absorption characteristics of hemoglobin; the absorption peaks of hemoglobin in the visible wavelength region yield the two valleys in the wavelength region between 400 and 600 nm.

Fig. 1

Representative Rd _{m, site}(λ) from cerebral cortex (thick solid line), cerebella cortex (thin solid line), and brain tumor (dotted line). The error bars represent the STD_site(λ).

The Student's t test (two-sided, equal variance) was used to detect any significant differences between the average intensities of nRd _m,site(λ) obtained from normal brain tissue and that obtained from tumor tissues at a given wavelength. The results of the statistical comparisons, presented by p values, are shown in Fig. 2. Statistically significant differences (p < 0.05) were observed in three spectral regions: 450 to 530, 590 to 690, and beyond 800 nm. Because of the high noise level, the differences beyond 800 nm were considered to be of questionable significance. The same statistical comparison was performed between the average intensities of nRd_m, _site(λ) from the cerebral and cerebellar cortex. However, no significant differences were detected (Fig. 2).

Fig. 2

Results of the Student's t test of the average diffuse reflectance intensities between (a) the normal and the tumorous groups (dashed line) and (b) cerebral and cerebellar cortex (solid line). The horizontal line indicates the significance level α = 0.05.

The outcome of the area under the ROC curve analysis is summarized in Fig. 3. It was noted that, in general, greater A _ROC values were found in the longer wavelength region (i.e., 600 nm to 830 nm). The maximum A _ROC obtained was 0.77, which was located at ∼820 nm.

Fig. 3

Graph of A _ROC(λ) from the area under the ROC curve analysis.

To demonstrate the effectiveness of the identified spectral features (i.e., normalized diffuse reflectance intensity) for differentiating brain tumor tissue from normal brain tissue, one representative spectral feature, nRd _{m ,site} (700 nm), was used to construct an empirical discrimination method. This discrimination method produced a sensitivity of 95% and a specificity of 66%, using an arbitrary cutoff of 0.65 (Fig. 4). A similar performance was obtained from the discrimination algorithms, using a single nRd _{m ,site}(λ) between 600 and 830 nm as the input.

Fig. 4

Results for nRd _{m ,site}(700 nm) from (a) normal brain samples and (b) brain tumor samples. The error bars represent STD_site(700 nm). The black line is an arbitrary cutoff (0.65) established for the purpose of discrimination; brain samples with nRd _{m ,site}(700 nm) greater than the cutoff are classified as normal.

The analysis of %STD_site(λ) provides insights into the spectral variations induced by the handheld probe. As shown in Fig. 5, the average %STD_site(λ) varied from 10 to 50% between 400 nm and 850 nm, and the maximum %STD_cat(λ) was observed close to 400 nm. It was also noticed that the values of %STD_site(λ) below 600 nm were, in general, much greater than the values above 600 nm. Interestingly, the profile of the average %STD_site(λ) of each tissue category strongly resembles the absorption spectrum of oxyhemoglobin. According to the Student's t test, the difference in the average %STD_site(λ) between the normal brain category and the brain tumor category was not statistically significant.

Fig. 5

Average %STD_site(λ) of cerebral cortex (thick solid line), cerebellar cortex (thin solid line), and brain tumor (dotted line). The upper dashed line represents the t test results between the normal brain (i.e., cerebral+cerebellar cortex) and brain tumor groups.

4.

Discussion

The results from the clinical study reported in this paper suggest that pediatric brain tumors may be differentiated from the normal brain in vivo using diffuse reflectance spectroscopy alone. Statistical comparisons identify several wavelength regions where diffuse reflectance intensities from normal brain differ from those of brain tumors. Without employing sophisticated spectral analysis and classification tools, we were able to construct an empirical discrimination method, based on one single spectral feature, with reasonably high accuracy. Despite the small number of patients enrolled in this study, we believe the results strongly support the feasibility of using diffuse reflectance spectroscopy to identify brain tumors in vivo and, hence, guide a pediatric brain tumor surgery.

In vivo diffuse reflectance signals from cerebral cortex were found to be higher than those from brain tumors over the spectral region recorded. This observation is inconsistent with the findings of the adult brain tumor studies.^{31, 33} Both Gebhert and Eggert and Blazek reported that the scattering properties of adult brain tumors were stronger than those from cerebral cortex. Therefore, it is expected that the diffuse reflectance signals between 600 and 800 nm from adult brain tumor should be stronger than those from cerebral cortex. The discrepancy, in our opinion, should be attributed to the structural/composition differences between adult and pediatric brain tumors. Therefore, optical surgical guidance developed for adult brain tumors would not be directly applicable to the pediatric brain tumor population.

In this study, diffuse reflectance signals were also measured from the cerebellar cortex, and they were comparable with those from cerebral cortex in terms of spectral intensity and profile. Therefore, a separate classification algorithm for cerebellar cortex, cerebellar tumors did not have to be created. Because the number of measurements from white matter was relatively low, they were not included in the statistical comparisons disclosed in the result section. Nevertheless, we are certain about the feasibility of using the same spectral features, namely, nRd _{m ,site}(λ) for λ = [600 nm, 800 nm], to differentiate cerebral and cerebellar white matter from pediatric brain tumors with high efficiency. This assessment is supported by several previously published studies that showed that diffuse reflectance signals from white matter are notably stronger than those from the cerebral cortex due to its high scattering characteristic. The strong scattering properties of white matter have been attributed to the dimension and density of neural axons as well as neural myelination.³³

The two spectral regions showing statistically significant differences between normal cortex and tumor tissue are 450 to 530 and 590 to 690 nm. However, the analysis of A _ROC suggests that diffuse reflectance signals beyond 600 nm are more suitable discriminants. This discrepancy can be explained by the distribution characteristics of nRd _{m ,site}(λ) from the tumorous group. As shown in Fig. 4, the nRd _{m ,site}(λ) of two samples significantly deviate from the group mean. In other words, these two samples would introduce a strong bias to the distribution of nRd _{m ,site}(λ) of the tumorous group, thus degrading the results of mean comparison (i.e., increasing the p value). According to the histological evaluation, both samples belong to the ganglioglioma tumor category, which suggests a separate tissue discrimination algorithm may be developed for this particular tumor type. Note also that diffuse reflectance spectral signal from ganglioglioma is stronger than those from the other specimens over the entire spectral region. Future studies with a larger number of patients in each histological category will be required to ensure this conclusion.

One issue associated with the utility of the handheld probe is its stability. The investigated site as well as the probe contact pressure vary because of the movements of the investigated organ (e. g., brain pulsation) and/or the operator's hand. This instability may affect in vivo diffuse reflectance spectral acquisition in two distinct ways. First, the lateral movement of the handheld probe causes a shift in the investigation location; thus, the resulting spectrum represents the characteristics of a tissue volume greater than expected. This in turn introduces more errors to the process of correlating the histologic findings with the spectral data. Second, the axial movement of the probe induces irregularity in the probe contact pressure, which primarily leads to a significant change in tissue hemodynamics. As reported by several research groups, ^{34, 35} excessive probe contact pressure can lead to significant changes in local tissue hemodynamics, which explains the high%STD in the region of hemoglobin absorption, as shown in Fig. 5. Therefore, the spectral features below 600 nm are not reliable inputs for the tissue discrimination purpose. In a separately study which is not yet published, we noticed that the usage of a handheld probe in conjunction with a mechanical holder can significantly reduce %STD. Such an improvement enables continuous recording of diffuse reflectance spectra from the brain, which is important for applications such as studying the temporal dynamics of epileptic cortex.

5.

Conclusion

We investigated the utility of diffuse reflectance spectroscopy for in vivo pediatric brain tumor detection using a small-scale clinical trial. The results of the study found that pediatric brain tumors may be differentiated from normal cortex using spectral features such as diffuse reflectance intensity between 600 and 800 nm. Because of the high scattering properties of the white matter, the same differentiation efficient should be expected between white matter and pediatric brain tumors. Spectral artifacts induced by the handheld probe also were quantified by studying the standard deviations of diffuse reflectance spectra continuously acquired at the same investigated site. It was found that large intensity variations exist in the spectral region of 400 to 650 nm in the repeatedly measured diffuse reflectance spectra. This in turn diminishes the applicability of the spectral features from this wavelength region to the discrimination algorithm development.