Patent Application
20160139051
Embodiments relate to an apparatus and method for distinguishing between different tissue types, such as brain tissue, using specific Raman spectral regions.
Glioblastoma (GBM) is an extremely aggressive primary brain tumor. Despite multimodality therapy, including maximal surgical resection and adjuvant radiation and chemotherapy, the prognosis for GBM patients remains dismal, with an average life expectancy around 12-18 months. A significant factor in determining patient outcomes is the completeness of resection of the malignant tissue. However, GBM is a diffusely infiltrating glioma, and the tumor margins are difficult to identify intraoperatively, even with the assistance of intraoperative neuronavigation
Establishing a histopathological diagnosis of GBM is essential for initiating therapy. Intra-operative consultations with frozen sections are often performed to help confirm the presence of tissue diagnostic of GBM in the biopsy samples. However, performing frozen sections is limited by the time taken during ongoing neurosurgery, the need for an experienced neuropathologist to interpret the frozen sections, and various processing artifacts that may lead to sub-optimal histological evaluation. In areas of eloquent brain where glioma cells infiltrate normal tissue, patient functionality outcomes may be improved by sparing normal brain while leaving small populations of residual cells, but available tools are often too imprecise for this level of discrimination.
Intra-operative magnetic resonance imaging (iMRI) has been suggested as a potential tool for identifying tumor tissue along the surgical margins to aid in resection. However, uptake of contrast enhancement in areas of diffuse tumor is not as robust as in the tumor core. Furthermore, iMRI-assisted surgery is limited by its significant cost, the time of imaging, and its accessibility confined to major cancer centers.
In one embodiment, a portable apparatus for distinguishing between different tissue types in a tissue sample is provided, the apparatus including a housing, and a plurality of Raman spectrometers disposed within the housing, each spectrometer having a different spectral region. A processor is provided in communication with the plurality of spectrometers, the processor analyzing output from the plurality of spectrometers to identify the tissue type of the tissue sample.
In another embodiment, a method for distinguishing between different tissue types in a tissue sample is provided, the method including providing a portable apparatus having a housing, a light source disposed within the housing, and a plurality of Raman spectrometers disposed within the housing, each spectrometer having a different spectral region. The method further includes illuminating the tissue sample using the light source, receiving light from the tissue sample with the plurality of spectrometers, and analyzing output from the plurality of spectrometers to identify the tissue type of the tissue sample.
In another embodiment, a method for distinguishing between different tissue types in a tissue sample using different Raman spectral regions is provided. The method includes (a) selecting a first spectral region which provides a best classification accuracy between the tissue types, (b) selecting a next spectral region that provides a next best classification accuracy between the tissue types, (c) repeating step (b) until a plurality of spectral regions are selected that, when combined, provide a desired combined classification accuracy, and (d) analyzing the tissue sample with the plurality of selected spectral regions to determine the tissue type in the tissue sample.
As required, detailed embodiments of the present invention are disclosed herein; however, it is to be understood that the disclosed embodiments are merely exemplary of the invention that may be embodied in various and alternative forms. The figures are not necessarily to scale; some features may be exaggerated or minimized to show details of particular components. Therefore, specific structural and functional details disclosed herein are not to be interpreted as limiting, but merely as a representative basis for teaching one skilled in the art to variously employ the present invention.
Embodiments include a portable Raman spectroscopy apparatus and method for the in vivo identification and distinction between normal tissue, necrotic tissue, and tumor tissue and their boundaries in real time, such as during surgery. Raman spectroscopy is a non-destructive surface technique which provides a molecular signature of the region under examination. When light is incident on a sample, most of the light is scattered back at the same energy and wavelength. However, in rare cases (1 in 107 photons), there is an energy exchange between the incident photon and the molecule under examination causing the scattered photon to shift its wavelength, termed the “Raman effect”.
As is known in the art, Raman spectrometers use focused laser light and highly accurate optical systems to rapidly measure a molecular signature of a region under examination. Raman spectroscopy can be performed at several points within a region of tissue to provide a molecular map of the tissue. Because Raman spectroscopy is non-destructive and is not significantly impacted by water, it is an ideal tool for mapping regions of tumor and necrosis in the brain. Preliminary in vivo studies of brain tissue have been performed using fiber optics connected to full-size (benchtop) Raman spectrometers. However, these spectrometers are large and expensive, and the output spectrum must undergo significant processing to provide a diagnosis.
A typical Raman spectrum provides hundreds, or even thousands, of data points, each representing a specific wavelength or energy shift. Traditional statistical methods are not suited for this type of data. Compression methods such as principal component analysis have been used to reduce data to a few significant variables. However, this ignores the wealth of molecular data present in the Raman spectrum. Frequently, compressed data is then used for clustering methods to identify like regions within areas of tissue. These unsupervised methods are then correlated with histology, and classification methods are developed based on the clusters. While other blind methods, such as support vector machines, have provided high accuracy, these continue to ignore the molecular significance of the Raman spectra.
In an embodiment of the disclosed apparatus and method, a selected group of peaks or regions in a Raman spectrum which provide specific biological (molecular) information, rather than the entire Raman spectrum, may be used to identify GBM tumor tissue, necrosis, and normal brain tissue and their boundaries. A study was performed to distinguish between normal brain (grey matter), necrosis, and GBM regions in banked frozen tissue samples using Raman spectroscopy. Discriminant function analysis was used for spectral identification, to allow for biologically relevant interpretation of the model structure. Homogenous regions of normal grey matter, necrosis, and GBM were identified. Using data from these ‘known’ areas, a select group of Raman peaks was directed into discriminant function analysis for tissue identification. Using discriminant function analysis allows for rapid, accurate identification of neural tissue without loss of meaningful biologic data.
In the study, Raman spectroscopy was performed with an InVia Raman microscope (Renishaw, Gloucestershire, UK), using a 785 nm excitation laser, 1200 l/mm grating, and a 576×400 pixel thermoelectric-cooled charge-coupled device (CCD). A 50× Nikon plan-fluor objective with a numeric aperture of 0.45 and working distance of 4.5 mm was used for measurements, with an approximate spot size of 5×30 μm when focused on the sample. When optimized, laser power at the sample is approximately 115 mW at 100% power, and spectral resolution is 0.7 cm−1. Actual resolution varied from 0.82 to 0.98 cm−1. Prior to daily measurement, system calibration was performed using a silicon control sample. Data was measured over a spectral range of 600-1800 cm−1. Each spectrum consisted of 1 accumulation with an integration time of 10 seconds and a laser power of 50%. At least two distinct regions were measured on each tissue. For each measurement, a region of interest was identified based on the following criteria: a) the region was level to ensure consistent focus throughout the area, and b) the region was of recognizable features or near identifiable orientation markers for easy correlation with the H&E section. Renishaw Wire software then automatically subdivided the region into a 25-μm grid. A Raman measurement was performed at each grid point in the selected measurement region.
Following Raman measurement, each region measured with Raman spectroscopy was identified and photographed at 20× and 40× magnification on the adjacent H&E slide, when possible. Some regions could not be correlated due to folding or stretching in the frozen sections, or lack of orientation markers within the tissue. An experienced neuropathologist examined each H&E slide and marked distinct regions of normal grey matter, tumor, or necrosis. Areas of tumor were further noted as suspicious for tumor, diffuse glioma, and GBM. Freeze artifact was noted when it was present. Images of each map region were reviewed, and the location of each region was compared to the marked regions on the H&E slide to reach a final gold-standard diagnosis for each area studied. For each region measured, the H&E slide and recorded images of each region measured by Raman spectroscopy were reviewed by an experienced neuropathologist.
Spectra were preprocessed using proprietary software by spike elimination, background subtraction, vector normalization, and Whitaker smoothing prior to statistical analysis. Following processing, spectra were individually reviewed to remove spectra containing obvious measurement error, such as missing data, failure of cosmic ray removal, or CCD overload.
Raman data from homogenous regions of normal grey matter, necrosis, and GBM which did not display freeze artifact were identified based on histologic validation. The data was further randomly divided into a model training group and a validation group. Three regions each of normal grey matter, necrosis, and GBM were included in the training group. A third group was established for further validation, which consisted of regions identified as homogenous but displaying freeze artifact. The overall mean of each type of tissue in the training group was calculated and plotted using the Statistical Analysis Software (SAS) procedure PROC MEANS. All distinct Raman peaks and shoulders were identified within the GBM, normal grey matter, and necrosis spectra. These distinct peaks were input to discriminant function analysis using SPSS (i.e., a statistical analysis package from IBM) to develop a tissue identification algorithm. This algorithm was then applied to the validation groups (with and without freeze artifact).
A total of 17,138 Raman spectra were collected from 95 distinct regions of 40 brain tissues. An average of 180 individual spectra were measured from each region (minimum 35, maximum 494, standard deviation 76). The 40 tissues were extracted from 17 donors; 12 with a GBM diagnosis and five with a non-tumor (epilepsy) diagnosis. Among the GBM donors, 58.3% were male ( 7/12), with an average age of diagnosis at 63.9 years (range: 47 to 76 years), with diagnosis occurring between years 1993 and 2000, and an average time of 353.5 days (range: 8 to 806 days). Among the non-tumor donors, 60% were males (⅗), with surgery occurring between 1993 and 1996 at an average age of 31.8 years (range: 19 to 45 years).
Because of the sensitivity of the discriminant function analysis algorithm, regions which contained a mix of cell types (i.e., diffuse glioma, tumor/necrosis border, etc.) were excluded from this study. H&E review showed that a majority of the normal tissue provided for this study was grey matter. Previous studies have shown a distinct biochemical and Raman difference between grey and white matter, and other brain structures, such that the system and method disclosed herein may also be used to distinguish between these tissue types. However, for this study, regions of white matter and leptomeninges were excluded from analysis. Likewise, regions of severe hemorrhage were excluded, as they have been shown to have a unique Raman signature.
Three regions each of normal grey matter, necrosis, and GBM were randomly assigned to a model training set (1396 spectra), while the rest of the data was reserved as the primary validation set (1759 spectra). The mean Raman spectrum was calculated for normal grey matter, necrosis, and GBM in the training data. Raman shoulders and peaks were identified at 927, 934, 954, 958, 977, 1003, 1030, 1061, 1081, 1107, 1122, 1154, 1172, 1206, 1239, 1255, 1259, 1266, 1300, 1313, 1334, 1397, 1419, 1441, 1518, 1552, 1578, 1581, 1604, 1614, 1616, 1657, 1659, and 1735 cm−1.
There is a general consensus that the Raman spectrum of normal brain tissue is dominated by lipids (1063, 1081, 1127, 1268, 1298, 1313, 1397, 1440, 1657 cm−1) and cholesterol (1440, 1670-1675, 1735 cm−1). Necrosis is characterized by an increased protein content (phenylalanine at 1003, 1032, and 1208 cm−1, tyrosine [peaks below 900 cm−1], tryptophan at 1340 cm−1, amide I band 1645-1675 cm−1, amide III at 1225-1300 cm−1, and CH3, CH2 deformation of collagen at 1313, 1397, and 1440). Increased concentration of cholesterol esters (1739 cm−1), carotenoids (1159 and 1523 cm−1), calcifications (985 cm−1), and hemoglobin (1250 and 1585 cm−1) have also been reported. However, hemoglobin can be present in any excised tissue. GBM tissue has been shown to have lower lipid and cholesterol content than normal brain tissue, and higher nucleic acid content (1097 and 1580-1700 cm−1). When compared with necrosis, tumor tissue should display increased lipids and nucleic acids. The Raman data from this study followed those trends. Normal grey matter showed strong contribution from lipid peaks, especially at 1061 and 1081 cm−1, and necrosis had strong contributions from proteins, especially at 1003, 1206, 1239, 1255-1266, and 1552 cm−1. Peaks associated with carotenoids (1154 and 1518 cm−1), calcifications (977 cm−1), and hemoglobin (1250 and 1581 cm−1) were also elevated in necrosis. Necrosis also showed a broad shoulder at 1239 cm−1. This region is associated with the amide III band of proteins. As conformation of the protein structure changes, the peak becomes broader, suggesting necrosis has a higher concentration of α-helix and random chain structures than normal and GBM tissues. GBM had a lower protein content than necrosis, as evidenced at 1003, 1030, 1206, 1239-1266, 1313, 1552, and 1657 cm−1. The composition of 1061 and 1081 was lower in GBM than in normal grey matter, and higher than that of necrotic tissues. In the primary validation data set, overall accuracy was 97.8.
The 34 identified Raman peaks were split into combinations of peaks within 20 wavenumbers of each other. Discriminant function analysis was applied iteratively to each 20-wavenumber region to find the single region which provided the best overall classification accuracy between normal grey matter, necrosis, and GBM. The best region is shown as Region 1 in Table 1 below. Again, an iterative process was applied to each 20-wavenumber region to find the single region which added the next best classification accuracy to the training data when added to Region 1. This process was repeated until five key regions were identified. The specific peaks used are shown, as well as the increase in overall accuracy when each combination of peaks is added. The 34 original Raman peaks input to discriminant function analysis provided 99.6% accuracy, however, 98.5% overall accuracy can be achieved by using only five key regions. However, it is understood that any number of spectral regions can be combined that provide a desired combined classification accuracy.
To confirm the accuracy of the model, it was applied to the primary and secondary validation sets. Results are shown below in Table 2 for each region measured. Overall accuracy in the primary validation group was 95.3%. Overall accuracy in the secondary training group was 71.3%. This group contained regions which displayed significant freeze artifact. The presence of freeze artifact may have altered the composition of the tissue, contributing toward the lower accuracy. Freeze artifact will not be present in in vivo tissues.
Therefore, five Raman spectral regions have been identified as most significant for diagnosis and identification of normal grey matter, necrosis and GBM, including region 1 of 1657-1660, region 2 of 1153-1172, region 3 of 1002-1004, region 4 of 1106-1123, and region 5 of 1254-1268 wavenumbers. Based on these data, in one embodiment, the apparatus and method disclosed herein provides a spectral analysis of these five different, non-overlapping Raman spectral regions. It is understood that these specific spectral regions were selected for distinguishing between normal grey matter, necrosis, and GBM tumors in the brain, and therefore other spectral regions may be identified for distinguishing between other types of normal, necrotic, and tumor tissue. It is also understood that while five spectral regions are utilized herein, this number is not intended to be limiting, as more or fewer regions may be suitable for identifying alternative tissue types.
Embodiments of the apparatus and method disclosed herein use specific Raman regions, instead of a broad range of wavenumbers, to identify different tissue types. Any Raman peaks within a range of 20 cm−1 can be measured by a single Raman chip/CCD detector or micro-Raman spectrometer, and an array of these spectrometers is combined in a single, portable apparatus 10 which is illustrated schematically in
With reference to
In addition to the tracking system 20, a light source (such as a laser diode) 22 for tissue illumination, excitation fibers 24, collection fibers 26, a collimating lens 28, an optical dispersing element 30, a CMOS detector 32, and a mirror 34 may be integrated into the handheld Raman apparatus 10. The wavelength of the light source 22 and other optical parameters may be defined based upon the optimal excitation wavelength and resultant Raman shifts described above. User controls 36 and a power indicator 38 may be provided, and the apparatus 10 may also be configured to provide a visible or audible indication of tissue type to the user.
In operation, the surgeon places the probe 14 adjacent the tissue and acquires the Raman spectra for the tissue to determine the tissue type. Light scattered by the tissue is collected by optics to be transmitted to the spectrometers (CCD) 12a-12e. The processor 16 analyzes spectra from the five spectrometers (CCD) 12a-12e, determining signal intensity in order to identify specific spectral characteristics of the Raman spectra received in real time to determine tissue type. The surgeon can also use the probe 14 to locate margins of the tumor by determining the points nearest to the tumor where the apparatus 10 indicates that the tissue is normal. Once the margins of the tumor have been identified, the entire tumor can be removed without removing excess tissue.
Although the processor 16 is shown as being contained within the housing, the processor may alternatively be a personal computer external to and in communication with the apparatus 10. The processor 16 can execute software instructions stored in a memory module (not shown) in communication with the processor 16 which causes the processor to perform the method disclosed herein. The software instructions may be stored on a computer-readable medium such as, but not limited to, physical media or electronic data storage media.
The disclosed apparatus and method can be used to identify tumor, necrosis, and normal tissue regions in vivo without damaging tissue. This is a need of every neurosurgeon who operates on brain cancer. The portability of the disclosed spectrometer apparatus and its design for Raman bands uniquely suited to neurosurgical applications make it ideal for in vivo tumor detection, mapping tissue boundaries, and glioblastoma resection surgery.
While exemplary embodiments are described above, it is not intended that these embodiments describe all possible forms of the invention. Rather, the words used in the specification are words of description rather than limitation, and it is understood that various changes may be made without departing from the spirit and scope of the invention. Additionally, the features of various implementing embodiments may be combined to form further embodiments of the invention.
This application claims the benefit of U.S. provisional application Ser. No. 61/844,926 filed Jul. 11, 2013, the disclosure of which is hereby incorporated in its entirety by reference herein.
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/US2014/046391 | 7/11/2014 | WO | 00 |
| Number | Date | Country | |
|---|---|---|---|
| 61844926 | Jul 2013 | US |
