At present, cancer diagnosis is generally based on conventional histopathology. The pathologist reaches a decision by interpreting the morphological changes at the tissue and cellular levels after staining excised tissue biopsies. However, cancer is a complex disease, and it can be difficult to reach inter- and intra-observer agreement on diagnosis. Furthermore, limited automation capabilities make the tissue-staining method labor-intensive, and this can adversely affect the speed of diagnosis.
In the search for novel cancer diagnostic tools, biophotonic approaches based on vibrational spectroscopy (IR and Raman) are interesting candidates. These techniques probe the biomolecular composition of cells, tissues, and biofluids in a non-destructive and label-free manner. Of particular interest is spectral imaging in the mid-IR region (around 2–12μm), which can rapidly complement histopathology because of its high signal-to-noise ratio in biological materials.1 However, until recently mid-IR imaging in a laboratory setup was limited to low-intensity light sources and smaller detectors, which allowed only for small-scale screening of samples. To resolve this problem, synchrotron facilities provided high-intensity mid-IR radiation, but their availability for continuous usage is limited.
In recent years, the field of mid-IR spectral histopathology has undergone rapid development, especially with respect to high-intensity light sources and focal plane array (FPA) detectors within laboratory setups. For instance, quantum cascade laser (QCL)-based IR spectroscopic imaging is now available. In this approach, multiple QCLs can be tuned to frequencies in the range of 5–11μm (1000–1800cm−1) either in total or in selected discrete frequencies within this range.3Alternatively, advances in materials science have enabled the development of novel broadband supercontinuum light sources in the range of 2–12μm (around 800–4000cm−1). It may be possible to combine these light sources with megapixel detectors that are also being developed.4 The large-array detector enables simultaneous screening of larger sample areas, and the high brightness of the supercontinuum laser means that less time is required to obtain spectral information with sufficient signal-to-noise ratios. Furthermore, rather than using the supercontinuum source as an alternative source for an interferometric approach, it is feasible to employ an acousto-optical tunable filter to enable discrete frequency imaging. By doing so, it may be possible to measure selected wave numbers that are of diagnostic importance, thereby increasing the rapidity of the process.5 Several parties are currently developing novel approaches to large-scale pathology screening for both histopathology and cytopathology through mid-IR spectral imaging.6
In this context, we are undertaking a background study that would optimize various parameters for mid-IR imaging of biological tissues, different aspects of sample preparation protocols, spectral preprocessing, and processing methods that use the current state-of-the-art mid-IR Fourier transform IR (FTIR) imaging technology. We carried out FTIR imaging of paraffinized colon tissue using an Agilent FTIR imaging system (an Agilent 620 FTIR microscope coupled with an Agilent 670 FTIR spectrometer) with a Globar light source and a liquid-nitrogen-cooled 128×128-FPA detector. We used the instrument's standard magnification (5.5×5.5μm2 pixel size) and a high-magnification modality (1.1×1.1μm2) in the spectral range of 1000–1800cm−1 at a spectral resolution of 4cm−1 with 64 scans per FPA tile. The different magnifications enabled study of the biomolecular compositions from larger glandular regions as well as smaller cellular regions. We preprocessed the IR spectra and subjected them to cluster analysis to identify different pathology types in colon tissues (see Figure 1). We considered three types of pathologies, which corresponded to cancerous, non-cancerous, and intermediate groups (adenoma and hyperplastic). We then subjected the spectral signatures from the segmented epithelial regions of the three pathology groups to multivariate statistical analysis, with the aim of discriminating the signatures into respective groups in both magnifications.
Comparison of IR spectral-histopathological characteristics of a non-tumoral colon tissue sample that was measured using different IR imaging setups independently. Top row: Conventional IR imaging. Top row, middle panel: Cluster analysis (using 11 cluster groups) of non-tumoral colon tissue. The groups were measured using the conventional imaging setup (5.5×5.5
)and compared with the reference hematoxylin and eosin (HE)-stained image (top row, left panel). The corresponding dendrogram (top row, right panel) represents the heterogeneity of the clusters. Bottom row: High-magnification IR imaging. Bottom row, middle panel: Cluster analysis of the normal colon tissue (using 11 cluster groups). The groups were measured using the high-magnification imaging setup (1.1×1.1
)and compared with the reference HE-stained image (bottom row, left panel). The corresponding dendrogram (bottom row, right panel) shows the heterogeneity of the clusters. (Figure reproduced with permission.2
The preliminary analysis of the discrimination of tissue pathologies for the two group models (cancerous and non-cancerous) shows discrimination attributes with sensitivity and specificity values of 81 and 86% in standard magnification, and 86 and 84% in high magnification, respectively. However, when we introduced the intermediate group into the model, we saw a drop in these values, particularly the sensitivities, which indicated an overlap between the groups. Taking this into consideration, we will test alternative discrimination models to see whether there is an improvement in the discrimination values. Nonetheless, the high-magnification imaging showed detailed histological aspects of colonic glands, and could resolve smaller cellular features (such as the goblet cells) directly.2, 7
In summary, we have highlighted the potential benefits of high-magnification IR imaging for visualization of the biochemical distribution across colonic tissue sections. There appears to be some improvement in performance for two-group discrimination of disease using higher magnification optics with 1.1μm pixels. However, when introducing intermediate pathologies, there is a significant loss of performance. We are now further investigating the causes of this. We will extend our work, and the protocols established, to test the supercontinuum source and the megapixel detector being developed. This work is part of a multi-organization project to enable large-scale pathology screening and the development of novel diagnostic models based on spectral histopathology and cytology.
Nick Stone, Jayakrupakar Nallala
University of Exeter
Nick Stone is professor of biomedical imaging and biosensing and head of the Physics and Astronomy Department. During almost 20 years in the UK's National Health Service, he has undertaken pioneering work on clinical biophotonics, and has published more than 170 peer-reviewed papers and proceedings.
Jayakrupakar Nallala is a postdoctoral early career researcher. His current research interests include implementation of vibrational spectroscopic methods for next-generation mid-IR spectral imaging and Raman in combination with clinical/pathological practices for the diagnosis of cancer and other molecular diseases.
Gavin R. Lloyd
Biophotonics Research Unit
Gloucestershire Hospitals NHS Foundation Trust
Gavin Rhys Lloyd completed his PhD at the University of Bristol and is currently a senior researcher. His research interests include the analysis of data sets with biochemical origin. Recently, he has focused on pattern recognition for early detection of cancers through the use of vibrational spectroscopy.
1. D. Fernandez, R. Bhargava, S. Hewitt, I. Levin, Infrared spectroscopic imaging for histopathologic recognition, Nat. Biotech. 23, p. 469-474, 2005.
2. J. Nallala, G. Lloyd, N. Shepherd, N. Stone, High-resolution FTIR imaging of colon tissues for elucidation of individual cellular and histopathological features, Analyst 141, p. 630-639, 2016.
3. C. Hughes, G. Clemens, B. Bird, T. Dawson, K. M. Ashton, M. D. Jenkinson, A. Brodbelt, et al., Introducing discrete frequency infrared technology for high-throughput biofluid screening, Sci. Rep. 6, p. 20173, 2016.
4. C. Rosenberg Petersen, U. M⊘ller, I. Kubat, B. Zhou, S. Dupont, J. Ramsay, T. Benson, et al., Mid-infrared supercontinuum covering the 1.4--13.3μm molecular fingerprint region using ultra-high NA chalcogenide step-index fibre, Nat. Photon. 8, p. 830, 2014.
5. G. Lloyd, N. Stone, Method for identification of spectral targets in discrete frequency infrared spectroscopy for clinical diagnostics, Appl. Spectrosc. 69, p. 1066-1073, 2015.
A description of EU-funded research on mid- and near-IR spectroscopy for diagnostics. Accessed 28 April 2016.
7. J. Nallala, G. Lloyd, C. Kendall, H. Barr, N. Shepherd, N. Stone, Identification of GI cancers utilising rapid mid-infrared spectral imaging, Proc. SPIE
9703, p. 970303, 2016. doi:10.1117/12.2209363