q 2006 International Society for Analytical Cytology
Cytometry Part A 69A:888–896 (2006)
Forensic Visualization of Foreign Matter in Human Tissue by Near-Infrared Spectral Imaging: Methodology and Data Mining Strategies Eunah Lee,1 Linda H. Kidder,1 Victor F. Kalasinsky,2 Joseph W. Schoppelrei,1 and E. Neil Lewis1* 2
1 Spectral Dimensions, Inc., Olney, Maryland Armed Forces Institute of Pathology, Washington, District of Columbia
Received 21 October 2005; Revision Received 14 February 2006; Accepted 2 March 2006
Background: Rapidity of data acquisition, high image fidelity and large field of view are of tremendous value when looking for chemical contaminants or for the proverbial ‘‘needle in the haystack’’ – in this case foreign inclusions in histologic sections of biopsy or autopsy tissues. Near infrared chemical imaging is one of three chemical imaging techniques (NIR, MIR and Raman) based on vibrational spectroscopy, and provides distinct technical advantages for this application. Methods: We have chosen to utilize and evaluate near infrared (NIR) imaging for studies of foreign materials in tissue because the experimental configuration is relatively simple, data collection is rapid, and large sample areas can be screened with high image fidelity and spatial resolution. Results: We have shown that NIR imaging can readily find and identify silicone gel inclusions in biological tissue
One of the challenges faced by pathologists as they examine histologic sections of biopsy or autopsy tissue to render medical diagnoses is to identify foreign materials that can be present in the tissue. These materials arise from two basic sources. Medical devices such as sutures, screws, hemostatic agents, and prosthetic implants can be present in a patient from documented medical procedures (1). Others such as drugs, excipients, and related materials are unexpected (2), and usually require a detailed clinical history to provide insight into potential sources. Some of these materials elicit significant tissue reactions in patients. Conversely, other materials appear to be completely biocompatible. The extent of an adverse reaction depends upon the individual patient and the particular foreign material but can range from mild inflammation to serious tissue fibrosis (3–6). One example of a foreign material in tissue is medical silicone. Silicone breast prostheses have been implanted into over 2 million patients in the USA for reconstructive surgery after mastectomies or for cosmetic augmentation. Silicone gel-filled implants were banned in the US in 1992 because of concerns over a possible connection between silicone and autoimmune diseases (7). Gel-filled implants
samples. Additionally, preliminary results indicate that spectral signatures in the data set are also potentially sensitive to structural changes in the surrounding tissue that may be induced by the foreign body. Conclusions: NIR chemical imaging is a powerful, nondestructive tool for localization and identifying foreign contaminants in biological tissue. Preliminary results indicate that NIR imaging is also sensitive enough to differentiate tissue types (perhaps based on collagen structural differences), and provide data on the spatial localization of these components. q 2006 International Society for Analytical Cytology Key terms: spectral imaging; near infrared imaging; forensic visualization; silicone gel inclusions; collagen structure; chemical imaging
were known to leak or rupture and deposit silicone fluid in a patient’s body. At the time the ban went into effect, implants constructed of a silicone elastomer shell and filled with normal saline solution were still available. Recently, however, a semisolid high-strength silicone gel implant has been introduced as an alternative to the saline implants (8). However, as a result of continued research into the possible connection between silicone implants and disease, the FDA has considered revoking the ban on gel-filled implants (9). As part of the foreign body response, it is common for foreign inclusions to be surrounded by a fibrous encapsulation that serves to isolate the foreign body from local tissues. Studies in which polymeric objects have been implanted into laboratory animals have shown that the collagen content within the capsular matrix changes with time. This composition change is specifically exhibited by the preferential formation of type I collagen when com*Correspondence to: E. Neil Lewis. E-mail:
[email protected] Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/cyto.a.20277
FORENSIC VISUALIZATION OF FOREIGN MATTER IN HUMAN TISSUE BY NIR IMAGING
pared with type III collagen within the capsular matrix (10,11). Collagen is one of the most abundant proteins in the human body, where it acts as the structural backbone for most tissue types, providing rigidity that keeps tissue from stretching. Its basic structure is a triple helix. At least 16 different types of collagen have been identified, although the body is dominated by Types I, II, and III. The difference between the types of collagen is largely structural, based on how the triple helical strands pack together and the structure of segments that are interspersed between the triple helical segments. Many chemical species can be identified based on the unique fingerprint of their molecular vibrations, which mid-infrared (MIR) [2,500–25,000 nm, more commonly 4,000–400 cm21 for Fourier transform infrared (FT-IR) spectrometers], near-infrared (NIR) [700–2,500 nm], and Raman spectroscopy (over the MIR range) all collect. Because vibrational spectroscopy is sensitive to the localized environment of CH, NH, and OH moieties, the different collagen types can be identified by subtle changes in their spectra. Depending on the type of inclusion present in the tissue, these spectroscopic techniques can be used to both identify the inclusion and potentially study changes in the structure of the tissue surrounding the foreign body. Each technique has attributes that make it particularly applicable for certain sample types. Absorptions arising from overtones and combination bands of O H, N H, and C H stretching and bending fundamentals are found in the analytical near infrared (NIR) region. The technique is particularly useful for performing rapid and reproducible analyses of easily differentiable materials (12). The absorptions of the fundamental vibrational bands in the MIR are one or two orders of magnitude more intense and are more easily interpreted than the overtone and combination bands of the NIR region (13). Because of the strength of the fundamental absorptions, rigorous sampling techniques are required to limit the amount of material interacting with the incoming radiation, and acquiring quality data is strongly operator dependent. However, where exquisite molecular sensitivity is required, MIR spectroscopy can distinguish subtle changes in biochemistry and structure more readily than its NIR counterpart. Raman spectroscopy works on the principal of inelastic scattering instead of direct absorption, and provides information in both the NIR and MIR spectral ranges, depending on experimental implementation (14). It provides complementary information to NIR and MIR, and works well with hydrated samples. Requiring single wavelength illumination, most often with lasers, can contribute to experimental complexity, although technology advances in recent years have made this a robust and user-friendly technique. Traditional approaches for MIR, NIR, and Raman collect and average vibrational spectral information from a large area of sample. The ability to detect and identify contaminants is not easily solved in this way, as the spectral signature of contaminants is diluted and dominated by the sample signature. Microspectroscopy, though in which a small spatially localized area of a sample is probed, can be useful Cytometry Part A DOI 10.1002/cyto.a
889
in overcoming the limitations and identifying localized sample components. A typical vibrational microspectroscopy experiment proceeds by identifying an anomalous region of the sample, using a light microscope and then guiding the spectrometer to obtain an infrared or Raman spectrum of that microscopic location (15–17). The single spectrum recorded enables chemical identification of that specific region. Clearly, for this approach to work and be useful, the anomalous region must be visible under the light microscope with its various modes of illumination. It is not a particularly good approach for rapid screening, since it requires supervision to determine the area to be probed, and it is also only useful for contaminants visible to a light microscope. Despite these potential drawbacks, however, microspectroscopic approaches are readily available and have been applied to this problem with success (18–21). The ideal tool for screening contaminants is one that couples the chemical identification capabilities of vibrational spectroscopy with optical imaging for increased speed and automation. Although developed later than the optical imaging techniques in the visible regions because of technological hurdles having to do with appropriate detectors, chemical imaging approaches that couple vibrational spectroscopic techniques (NIR, FT-IR, and Raman) with global imaging have been commercialized in the past decade. There are a variety of experimental implementations available, and specifics can be found in several publications (22–24). For all chemical imaging approaches, spectra are collected in a spatially resolved manner across a sample area. Figure 1 shows the resulting data structure, called a hypercube. A hypercube consists of a series of wavelength resolved images, each image corresponding to a different wavelength of light and with image contrast that depends on the molecular absorptions of the chemical components that make up the sample. If the response of a single pixel is plotted across these wavelengths, the result is a localized spectrum. As with all spectroscopic techniques, raw spectra are a convolution of physical and chemical properties of the sample, and appropriate and well-established preprocessing algorithms are applied to separate these effects. A cursory, nonautomated analysis of a hypercube involves examining individual image planes for chemically derived contrast and looking at the spectral differences of identifiable sample components. This type of analysis is useful for relatively simple samples, but by looking only at a few image planes or spectra, the vast amount of information contained within the hypercube is not used. Automated data mining techniques are common though, and often, chemical imaging data sets are processed using statistical and multivariate algorithms that encompass all of the wavelengths and spectra. As with their bulk counterparts, there are particular attributes ascribed to NIR, MIR, and Raman chemical imaging that differentiate the capabilities and determine which technique is best suited for a particular application. Specificity, desired sampling area, image fidelity, and speed of acquisition are important factors that influence the choice of technique. Given that molecular specificity
890
LEE ET AL.
FIG. 1. Typical data construct for chemical imaging, the hypercube. (a) shows the data as a series of wavelength resolved images and spatially resolved spectra. An image at a single wavelength (b) has contrast, based on the strength of the spectral features at that wavelength. Areas of the sample that have a relatively weak spectral signature will appear dark, and areas that have a strong spectral feature at that wavelength will appear bright. The spectra in (c) correspond to individual spatial regions of the sample.
for chemical imaging directly parallels bulk spectroscopy, FT-IR based MIR imaging has the best (22,23). However, because of the long wavelengths of light that are employed (constraints imposed by the diffraction limit of light) and the small format (64 3 64 pixels) of detectors that are available, it often exhibits the lowest image fidelity of the three techniques. Because of a combination of how the sample is illuminated and the type of optics (reflective) that are required by the technique, sample areas are normally no larger than a few millimeter on a side. Acquisition times can also be somewhat lengthy, because it is an interferometric technique, the entire spectral range must be sampled each time a measurement is made, even if only one wavelength is needed to make the chemical identification. Additionally, the sample preparation re-
quirements demanded for the bulk spectroscopic technique remain. Samples must be prepared to limit the amount of material that interacts with the probing radiation (e.g. microtomed thin sections of ~10 lm or KBr dilutions for pure components). Previously, this technique has been used to differentiate silicone from human tissue (24). Raman chemical imaging also has good molecular specificity and, because of the availability of standard scientific CCDs, can also have high image fidelity. A significant draw back of the technique is, however, the extremely small sample areas that can be effectively imaged. As mentioned previously, single frequency illumination is required to observe the effect of Raman-scattered radiation. Global illumination over a large area is not easily obtainable with Cytometry Part A DOI 10.1002/cyto.a
FORENSIC VISUALIZATION OF FOREIGN MATTER IN HUMAN TISSUE BY NIR IMAGING
a laser, and the inherent weakness in the scattered radiation dictates that strong illumination is needed to provide worthwhile spectral signatures. As a result, a 100 3 100 lm sample area is considered to be quite large for Raman imaging. An additional drawback associated with the relative weakness of the Raman signal is that optics for Raman imaging applications demand high light gathering efficiencies and short working distances. This makes it difficult to work with samples that are not perfectly flat. Despite these limitations, Raman imaging has been used to identify inclusions in tissue (25). Although NIR chemical imaging does not have the chemical sensitivity of FT-IR or Raman imaging, many sample applications can be accomplished with the sensitivity that is available. The technique has specific advantages in many other areas. Because of the large format (320 3 256) arrays, easily configurable sample illumination, and excellent signal-to-noise characterisitcs, image fidelity is in practice superior for NIR imaging relative to FT-IR and Raman. Working in the NIR permits the use of relatively simple refractive achromatic optics with long working distances. Curved and nonflat samples can be accommodated without significantly sacrificing image or spectral quality. A standard laboratory system can be quickly configured for microscopic applications (
