Ex vivo classification of spinal cord tumors using autofluorescence spectroscopy and immunohistochemical indexes

Spinal cord tumors are complicated and infrequent, which poses a major challenge to surgeons during neurosurgery. Currently, the intraoperative identification of the tissues’ pathological properties is usually difficult for surgeons. This issue influences the decision-making in treatment planning. Traditional pathological diagnoses can facilitate judging the tissues’ properties, but the diagnosis process is complex and time-consuming. In this study, we evaluated the potential of autofluorescence spectroscopy for the fast pathological diagnosis of specific spinal cord tumors. The spectral properties of six types of spinal cord tumors were acquired ex vivo. Several peak intensity ratios were calculated for classification and then associated with the pathological immunohistochemical indexes. Our results revealed the spectral properties of three types of intramedullary tumors different from those of the other three types of extramedullary tumors. Furthermore, some peak intensity ratios revealed a high correlation with the immunohistochemical index of glial fibrillary acidic protein (GFAP). Thus, we believe that autofluorescence spectroscopy has the potential to provide real-time pathological information of spinal cord tumors and help surgeons validate tumor types and perform precise tumor resection.


Introduction
Tumor diseases are well-recognized as a severe threat to the human health. It is estimated that tumor diseases may result in almost 7.6 million deaths annually all over the world in 2013 [1]. Among all tumor diseases known, the nervous system tumor is one of the most difficult tumors to be treated, such as brain tumors and spinal cord tumors. For most nervous system tumors, surgical resection is the most effective treatment method. However, currently, intraoperative identification of tissues' pathological information is difficult for surgeons. In neurosurgery, this pathological information is critical for surgeons because different types of diseases may require different treatment plans [2]. If surgeons are unsure about the tumor type, it is possible to develop an inadequate treatment plan and influence the treatment effect. Therefore, accurate pathological information of the tissues is crucial for neurosurgery.
Conventional medical imaging techniques, such as computed tomography (CT) and magnetic resonance imaging (MRI), provide preoperative information with high spatial resolutions for surgeons to develop a treatment plan; however, these two imaging methods do not provide real-time information. Intraoperative MRI (iMRI) and ultrasound imaging (US) possess the potential for real-time diagnosis, whereas iMRI is expensive and has a high demand for the operating room, and the use of US is limited because of the low quality images produced [3]. Furthermore, these abovementioned imaging techniques rarely provide accurate pathological information. Currently, frozen pathological biopsy is widely preferred for intraoperative tissue identification; however, this technique is time-consuming and complex. Therefore, it is deemed necessary to develop a more convenient method that can provide accurate pathological diagnostic information in a short time.
Optical biopsy is a promising method for fast intraoperative diagnosis. This method applies the phenomenon of interaction between light and the examined tissues, such as scattering and reflection, to obtain pathological information. The common optical diagnostic methods include Raman spectroscopy, optical coherence tomography, fluorescence imaging and spectroscopy [4]. In this study, we focused on fluorescence imaging and spectroscopy in consideration of its convenience for medical application [5][6][7]. Fluorescence may be sourced from extrinsic or intrinsic agents, and the corresponding imaging and spectroscopic methods can provide real-time diagnostic information. Fluorescence imaging by using extrinsic fluorescent agents such as 5-aminolevulinic acid (5-ALA)-induced protoporphyrin IX is widely studied and can provide intraoperative imaging of tumors [8]. Shimizu et al. [9] reported the application of photodynamic diagnosis in surgery for spinal ependymoma. Sabel et al. [10,11] used 5-ALA fluorescence-guided resection to treat intramedullary malignant glioma. Inoue et al. [12] applied 5-ALA fluorescence to the resection of intramedullary ependymoma. Millesi et al. [13] also investigated 5-ALA-induced fluorescence characteristics of spinal tumors. All these researches have illustrated that 5-ALA-induced fluorescence is able to help surgeons discriminate tumors better and thus improve the operation safety. However, the mechanism of 5-ALA-induced fluorescence remains unclear, and for some types of spinal tumors, 5-ALA-induced fluorescence cannot be observed. Besides, 5-ALAinduced fluorescence is hard to be used for the discrimination of different types of spinal tumors.
Autofluorescence arises from some intrinsic fluorescent agents when excited by specific wavelengths in biological tissues such as porphyrin, collagen, and flavin [14,15]. Autofluorescence spectroscopy is a kind of label-free diagnosis that may provide more information about tumors' pathological properties for fast classification, and some relevant researches have been carried out. Lin et al. [16] used autofluorescence combined with diffuse reflectance spectroscopy to measure the different parameters between the normal and tumorous brain tissues, and the corresponding algorithms based on these parameters were highly sensitive and specific. Saraswathy et al. [17] also used autofluorescence spectroscopy to test the optimum wavelength for the differentiation of brain tumors and thereby suggested an optimal excitation wavelength of 470 nm. Other researchers also applied autofluorescence spectroscopy to the liver, bladder, and colon tissues [18][19][20] and concluded that autofluorescence spectroscopy may be a fast and accurate method for intraoperative diagnosis.
Spinal cord tumors have a low incidence and are challenging to surgeons owing to the important functions of the normal spinal cord. However, few researchers have studied the optical properties of spinal cord tumors. In this study, we developed an autofluorescence spectral measurement system to acquire the spectral properties of different types of spinal  . 2), d in Table  onsidering only one ctra at the d of each measured recorded

Results
A total of 41 prototype of o  Fig. 6 R 7 , and R 8 in t ose in CNSET erage spectral cur (c) 350-nm excita tion wavelength. R classified into two ther the above be validated f of several n ratios includi wavelength of he same manner (R 5 ), were ob significant dis in Fig. 6  To further analyze the correlation between the clustering distribution of peak intensity ratios and the spinal cord tumor type, the immunohistochemical results were considered. Several immunohistochemical indexes, such as S-100, GFAP, and Epithelial Membrane Antigen (EMA), if present, were acquired. Correlation analysis was applied to analyze the relationship between the peak intensity ratios and immunohistochemical indexes ( Table 2). The results of correlation analysis indicated that the relative peak intensity ratios R 6 , R 7 , and R 8 shared a high and significant correlation with the GFAP expression. The tumors with negative GFAP are likely to achieve high R 6 , R 7 , and R 8 values (>1.0), in contrast, the tumors with positive GFAP are likely to achieve low R 6 , R 7 , and R 8 values (<1.0). GFAP exists in mature astrocytes of the human central nervous system; it is a type of intermediate filament and plays an important role in cell regulation [21]. GFAP also exists as a type of biomarker that indicates some kinds of tumors' malignancy and prognosis. The correlation between the GFAP expression and the values of relative peak intensity ratios possibly indicates that the GFAP expression may be correlated with the content of some intrinsic fluorescent agents in the tumorous tissues.
Furthermore, we also found that all ependymoma, CNS embryonal tumor and glioblastoma multiforme in our study were intramedullary. On the contrary, all lipomyoma, meningioma and schwannoma were extramedullary. This observation corresponded with the common performance of spinal cord tumors. In other words, most intramedullary tumors are likely to perform positive GFAP and possess low R 6 , R 7 , and R 8 values (<1.0), while most extramedullary tumors are likely to perform negative GFAP and possess high R 6 , R 7 , and R 8 values (>1.0).

Discussion
This study involved a total of 41 patients with six different types of spinal cord tumors. We conducted autofluorescence spectral measurements to evaluate the diagnostic potential of autofluorescence spectroscopy in these patients. To the best of our knowledge, this is the first application of autofluorescence spectroscopy to spinal cord tumors, mainly due to the relative low incidence of spinal cord tumors. In order to differentiate the six types of spinal cord tumors, several peak intensity ratios were calculated. The different peak intensity ratios of spinal cord tumors possibly indicated the different contents of endogenous fluorophores. We found that the values of R 6 , R 7 , and R 8 were larger in extramedullary tumors like lipomyoma, meningioma, and schwannoma than in intramedullary tumors like CNSET, ependymoma, and GBM. Furthermore, we analyzed the correlation between the autofluorescence properties and immunohistochemical indexes and found that the values of R 6 , R 7 , and R 8 were inversely proportional to the GFAP index. In other words, the GFAP expression may influence the relative intensity of autofluorescence spectrum, possibly indicating that GFAP is related to the amount of endogenous fluorophores in spinal cord tumors. This finding is believed to be extremely useful for clinical application. In neurosurgery, surgeons usually require different treatment plans for intramedullary or extramedullary tumors. Our proposed technique can provide fast pathological information of tumors, which will assist surgeons in judging the tumor type and therefore in planning a corresponding treatment plan to achieve better clinical outcomes.
From the measurement data, we now know that the major UV-VIS endogenous fluorophores in spinal cord tumors are mainly coincident with those in other human tumor tissues [16][17][18][19][20]. The emergence of the three common fluorescence peaks may be attributed to the emission of NAD(P)H, lipopigments, and porphyrins, respectively [14,22,23]. The particular fluorescence peak centered at 520 nm may arise due to the emission of flavin adenine dinucleotide and flavins [14,15], while the fluorochrome responsible for the fluorescence peak centered at 394 nm remained uncertain. The fluorochromes close to this excitation-emission band include structural proteins like collagen and elastin as well as vitamin B6 compounds like pyridoxine and pyridoxal 5′-phosphate [14]. Generally, Collagen does not exist in the brain tissues [22]. However, Saraswathy et al. [17] indicated that collagen might be responsible for one of the emission bands at 320-nm excitation in the brain tumor autofluorescence. To clarify which fluorochromes are responsible for the emission of spinal cord tumors and to further explain the correlation between different fluorochromes and GFAP expression, we will investigate the spectra and composition of more different spinal cord tumors by using additional other methods like mass spectrometry [24,25].
Spinal cord tumors occur in several types, and each type may have a unique autofluorescence spectral property. In fact, even the same tumor type may possess different optical properties when their growth environment is different. Therefore, it is deemed essential to validate the generalizability of our observations. In the future, we plan to analyze more kinds of spinal cord tumors for more precise tumor analysis, validate our conclusion that intramedullary and extramedullary tumors have different autofluorescence properties, and further explain the correlation between immunohistochemical indexes and autofluorescence spectroscopy. Our present results indicate that different types of intramedullary (or extramedullary) tumor demonstrate similar EEM and that it is difficult to distinguish within the intramedullary (or extramedullary) tumor group based on the currently available methods. Therefore, more excitation wavelengths like near-infrared (NIR) band and the corresponding techniques such as time-resolved spectroscopy and fluorophore localizing are believed to provide detailed information about tissues' properties [26][27][28], and thereby further improve the diagnostic efficiencies for precision medicine [29][30][31].
The grading of tumors is important for surgeons to decide the appropriate treatment plan. Currently, the grading of tumors is acquired by pathological diagnosis. Mitosis, cellular pleomorphism and cell morphology are the common methods used to grade a tumor. Some immunohistochemical indexes such as S-100 and Ki-67 may assist in judging the grading of a tumor. In this study, we found that autofluorescence spectroscopy of spinal cord tumors was related to the immunohistochemical indexes that were relevant with tumor grading. Thus, we believe that our method has the potential to assist tumor grading. In the future, we plan to expand the tumor samples for analyses with different grading in order to investigate the quantitative relationship between autofluorescence spectroscopy and tumor grading so as to develop new processing methods for faster grading of spinal cord tumors.
All spectral measurement experiments in this study were performed ex vivo. Due to the significance of their neurological functions, the normal spinal cord tissues were not collected. An ex vivo study can preliminarily validate the effectiveness of autofluorescence spectroscopy for tumor classification. In order to realize the distinction of spinal cord tumor tissues and normal tissues for intraoperative tumor boundary division, new detecting device will be developed for in vivo autofluorescence spectral measurement of the spinal cord. Furthermore, due to the limited data size, other statistical methods such as PCA or SVM could not be applied in this study. In the future, we plan to enlarge the data size to realize better classification results.

Conclusion
In conclusion, we developed an autofluorescence spectral measurement system and acquired the spectral properties of different types of spinal cord tumors ex vivo by it. Our results show that extramedullary tumors with negative GFAP possess significantly different autofluorescence spectral properties from intramedullary tumors with positive GFAP. The autofluorescence spectroscopy analysis method in our research is believed to act as an additional diagnostic tool for the intraoperative validation of tumor types and thereby help further surgical treatment in consideration of its real-time performance and result accuracy.