Figures
Explore figures and images from publications
Figure - uploaded by Peter Lasch
Content may be subject to copyright.
Abb. 4.1.3. Obige Abbildung wurde über eine Faktoranalyse berechnet (Daten: Dünnschnitt M7915/96, Fläche: 14 · 6 mm). Die euklidischen Abstände der Faktorenladungen 2 und 3 eines jeden Spektrums (vektornormierte 1. Ableitungen von 1450-900 cm-1 ) wurden gegen entsprechenden Koordinaten von Referenzspektren ermittelt. Die Distanzen im Faktorenraum wurden nach einer Normierung auf Werte zwischen 0 und 1 (0: Identität mit dem Referenzspektrum, 1: größtmögliche Distanz) farbcodiert und mit den Ortsparametern kombiniert. In diesem Beispiel wurden insgesamt 5 Einzelbilder übereinandergelegt. Hohe Farbintensitäten signalisieren hohe Ähnlichkeiten mit den jeweils verwendeten Referenzspektren. Ein Vergleich mit dem sequentiellem HE-Schnitt (Abb. 4.1.2.) ergibt eine gute Übereinstimmung zwischen der konventionell mikroskopischen und mikrospektrometrischen Technik.

Abb. 4.1.3. Obige Abbildung wurde über eine Faktoranalyse berechnet (Daten: Dünnschnitt M7915/96, Fläche: 14 · 6 mm). Die euklidischen Abstände der Faktorenladungen 2 und 3 eines jeden Spektrums (vektornormierte 1. Ableitungen von 1450-900 cm-1 ) wurden gegen entsprechenden Koordinaten von Referenzspektren ermittelt. Die Distanzen im Faktorenraum wurden nach einer Normierung auf Werte zwischen 0 und 1 (0: Identität mit dem Referenzspektrum, 1: größtmögliche Distanz) farbcodiert und mit den Ortsparametern kombiniert. In diesem Beispiel wurden insgesamt 5 Einzelbilder übereinandergelegt. Hohe Farbintensitäten signalisieren hohe Ähnlichkeiten mit den jeweils verwendeten Referenzspektren. Ein Vergleich mit dem sequentiellem HE-Schnitt (Abb. 4.1.2.) ergibt eine gute Übereinstimmung zwischen der konventionell mikroskopischen und mikrospektrometrischen Technik.

Source publication
Thesis
Full-text available
FT-IR Microspectroscopic maps of unstained thin sections from human melanoma and colon carcinoma tissues were obtained on a conventional infrared microscope equipped with an automatic xy stage. Mapped infrared data were analyzed by different image re-assembling techniques, namely functional group mapping ("chemical mapping", CM) and, for the first...

Citations

... If PCA plots show signicant spectral differences between spectral classes, it is likely that a supervised algorithm can be trained to analyze or diagnose unknown data. Supervised algorithms, such as Articial Neural Networks (ANNs) [22][23][24] have been used to diagnose spectra from test datasets with high accuracy. For example, an ANN was trained to differentiate clinical samples from the tongue, the anatomical region for which we had the largest dataset, from an equal number of normal tongue spectra. ...
Article
This paper presents a short review on the improvements in data processing for spectral cytopathology, the diagnostic method developed for large scale diagnostic analysis of spectral data of individual dried and fixed cells. This review is followed by the analysis of the confounding effects introduced by utilizing reflecting "low-emissivity" (low-e) slides as sample substrates in infrared micro-spectroscopy of biological samples such as individual dried cells or tissue sections. The artifact introduced by these substrates, referred to as the "standing electromagnetic wave" artifact, indeed, distorts the spectra noticeably, as postulated recently by several research groups. An analysis of the standing wave effect reveals that careful data pre-processing can reduce the spurious effects to a level where they are not creating a major problem for spectral cytopathology and spectral histopathology.
... In addition, it can be seen as a major advantage of infrared microspectroscopy as compared to MS based methods that it works on for- malin-fixed paraffinized samples, since such samples are available in large numbers and keep stable even over long periods of time. In agreement with the early work by Lasch [19,22] we observe, that pa- tient-to-patient variations in spectral patterns are smaller than those caused by disease, or even due to different tissue types. Thus, these pioneering studies realized and implemented the possibility of training supervised machine-learning algorithms for the diag- nosis of colon cancer [23]. ...
Article
During the past years, many studies have shown that infrared spectral histopathology (SHP) can distinguish different tissue types and disease types independently of morphological criteria. In this manuscript, we report a comparison of immunohistochemical (IHC), histopathological and spectral histopathological results for colon cancer tissue sections. A supervised algorithm, based on the "random forest" methodology, was trained using classical histopathology, and used to automatically identify colon tissue types, and areas of colon adenocarcinoma. The SHP images subsequently were compared to IHC-based images. This comparison revealed excellent agreement between the methods, and demonstrated that label-free SHP detects compositional changes in tissue that are the basis of the sensitivity of IHC. (© 2012 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim).
... The algorithm to construct spectra of individual cells from imaging datasets has been reported [20] and submitted for IP protection. Following earlier arguments [5,21], all data analysis was carried out on second-derivative spectra. ...
... The similarity of SHP and H&E images allows a detailed "annotation," that is, the correlation with tissue and cell morphological feature with corresponding spectral features, which, in turn, permits the training of diagnostic algorithms. The course to be taken for successful SHP studies was first outlined in a series of pioneering papers by the group at the Robert Koch Institut, Berlin [21,[65][66][67], and involves the following key steps: acquisition of very high S/N spectral data (the spatial resolution in the original studies were restricted by instrument performance), preprocessing including computation of 1st or 2nd derivatives and normalization to minimize instrumental and background artifacts, data presegmentation by unsupervised methods such as hierarchical cluster analysis (HCA), very careful annotation of diseased areas by a pathologist, and sufficiently large training datasets to construct a robust diagnostic algorithm. The diagnostic algorithm used in these initial studies was an artificial neural network (ANN) trained on thousands of spectra [67]. ...
... The diagnostic algorithm used in these initial studies was an artificial neural network (ANN) trained on thousands of spectra [67]. This work laid the ground rules in SHP and demonstrated that the patient-topatient variations of the observed spectra were smaller than those due to disease classification or tissue type [21]. ...
Article
Full-text available
This paper summarizes the progress achieved over the past fifteen years in applying vibrational (Raman and IR) spectroscopy to problems of medical diagnostics and cellular biology. During this time, a number of research groups have verified the enormous information content of vibrational spectra; in fact, genomic, proteomic, and metabolomic information can be deduced by decoding the observed vibrational spectra. This decoding process is aided enormously by the availability of high-power computer workstations and advanced algorithms for data analysis. Furthermore, commercial instrumentation for the fast collection of both Raman and infrared microspectral data has rendered practical the collection of images based solely on spectral data. The progress in the field has been manifested by a steady increase in the number and quality of publications submitted by established and new research groups in vibrational biological and biomedical arenas.
... The possibility for achieving SHP-based diagnostics was first demonstrated in the PhD dissertation of Lasch. 40 Owing to some unexpected experimental difficulties, computational restrictions and the lack of theoretical foundations at the time, it took another 10 years to refine experimental and computational methods for wide-spread applications of SHP in medicine. ...
Article
Full-text available
We report results of a study utilizing a recently developed tissue diagnostic method, based on label-free spectral techniques, for the classification of lung cancer histopathological samples from a tissue microarray. The spectral diagnostic method allows reproducible and objective diagnosis of unstained tissue sections. This is accomplished by acquiring infrared hyperspectral data sets containing thousands of spectra, each collected from tissue pixels about 6 μm on edge; these pixel spectra contain an encoded snapshot of the entire biochemical composition of the pixel area. The hyperspectral data sets are subsequently decoded by methods of multivariate analysis, which reveal changes in the biochemical composition between tissue types, and between various stages and states of disease. In this study, a detailed comparison between classical and spectral histopathology (SHP) is presented, which suggests SHP can achieve levels of diagnostic accuracy that is comparable to that of multi-panel immunohistochemistry.
... 512 for RES = 1 cm -1 ) was applied. Absorbance spectra were transformed by Savitzky-Golay derivation with 9 smoothing points (adapted from Ref. [17]). ...
Article
Full-text available
The past decade has witnessed substantial progress towards the application of infrared microspectrometry as an useful diagnostic tool for spectroscopic characterization of histological specimens. The combination of IR spectroscopy with microscopy, new technical developments such as sensitive multichannel detectors, and the implementation of modern multivariate concepts of data analysis permit high-quality infrared microspectroscopic imaging of tissue samples. The IR imaging methodology provides spatially resolved structural and compositional information of the histological specimens and opens in combination with computer based multivariate image reassembling techniques wide perspectives for routine use in the clinical environment. After 15 years of research it is now an accepted standard that IR spectra of cells, or tissues, can be considered as complex spectral fingerprints which cannot be always completely understood in an analytical sense. It is therefore advantageous to analyze the spectral fingerprints by pattern recognition techniques, preferentially of the supervised type, such as multilayer perceptron artificial neural networks (MLP-ANN), or support vector machines (SVM). As the concept of supervised classification requires a training, or teaching phase, in which labeled subsets of tissue reference spectra are analyzed, the compilation and validation of the teaching spectra will be the main challenge to render IR microspectroscopic expert systems applicable in practice. Thus, it is believed that the development of a non-subjective IR based tissue characterization technique will be dependent on the collection of databases of teaching spectra ideally containing a representative number of standardized spatially resolved IR microspectra of all relevant normal and pathologic tissue structures. The aforementioned aspect is important because despite all the fascinating perspectives, promising technical developments and exciting research papers, progress towards the translation of the technique into a practical application is less evident. The lack of available spectral reference databases is now identified as a factor which limits the transfer of the technique from the research laboratory to the clinical environment. In our presentation we will consequently discuss a number of topics relevant to the collection of spectral databases: the definition of measurement standards, measurement conditions and of quality control criteria for IR microspectrometry of tissues. The main focus of this presentation, however, is on the presentation of basic principles of data evaluation in IR microspectroscopic imaging of tissues. 2. Concepts of data analysis in mid-IR microspectrometric imaging of tissues Fourier Transform Infrared (FT-IR) microspectrometry provides spatially resolved structural and compositional information of the histological specimens under investigation and shows in
... Es konnte beispielsweise gezeigt werden, daß sich maligne und nichtmaligne Zellen bei Leukämie (Schultz et al., 1996) voneinander unterscheiden lassen. Zudem wurden verschiedene humane Tumoren wie Melanom (Lasch, 1999), Colon-Karzinom und Brustkrebs (Jackson und Mantsch, 1997) anhand ihrer IR-Spektren charakterisiert. ...
... Eine Möglichkeit, in Zukunft mit hoher Ortsauflösung viele Spektren in kurzer Zeit aufzunehmen und so die Vorteile der Mikrospektroskopie zu nutzen, ist die Verwendung moderner array-Detektorsysteme, mit denen innerhalb weniger Minuten ein Gewebeareal mit guter Ortsauflösung untersucht werden kann. Erste Untersuchungen an den Dünnschnitten der Hamsterhirne aus dieser Arbeit, welche freundlicherweise durch die Firma Bruker (einen Hersteller solcher array-Detektoren) ermöglicht wurden, ergaben zwar, daß die Qualität der Spektren, die man damit in akzeptabler Zeit aufnehmen kann, noch nicht ausreicht, um die spektralen Unterschiede zwischen TSE-infiziertem und Kontrollgewebe zu erkennen, anderer-seits wurden andere array-Systeme mit zufriedenstellendem Signal/Rausch-Verhältnis bereits für die Anwendung musterbasierter bildgebender Verfahren verwendet (Lasch, 1999). ...
... Hierbei wurden die Parameter Gewebedicke und Wasserdampf evaluiert. Als Maß für die Gewebedicke wurde die integrale Extinktion zwischen 1770 und 1100 cm -1 verwendet, der Wasserdampfgehalt wurde anhand der Intensitäten der Banden bei 1792,4 cm -1 und 1844,6 cm -1 aus den zweiten Ableitungen bestimmt (siehe auchLasch, 1999). Da nach Aufnahme des Hintergrundspektrums die mapping-Messungen jeweils mehrere Stunden dauerten, während denen sich die Luftfeuchtigkeit in einigen Messungenänderte, war trotz Trockenluftspülung des Mikroskops in einigen Datensätzen einë Anderung der Wasserdampfabsorption in den Spektren im Zeitverlauf beobachtbar. ...
Article
This paper reports the results of a collaborative lung cancer study between City of Hope Cancer Center (Duarte, California) and CIRECA, LLC (Cambridge, Massachusetts), comprising 328 samples from 249 patients, that used an optical technique known as spectral histopathology (SHP) for tissue classification. Because SHP is based on a physical measurement, it renders diagnoses on a more objective and reproducible basis than methods based on assessing cell morphology and tissue architecture. This report demonstrates that SHP provides distinction of adenocarcinomas from squamous cell carcinomas of the lung with an accuracy comparable to that of immunohistochemistry and highly reliable classification of adenosquamous carcinoma. Furthermore, this report shows that SHP can be used to resolve interobserver differences in lung pathology. Spectral histopathology is based on the detection of changes in biochemical composition, rather than morphologic features, and is therefore more akin to methods such as matrix-assisted laser desorption ionization time-of-flight mass spectrometry imaging. Both matrix-assisted laser desorption ionization time-of-flight mass spectrometry and SHP imaging modalities demonstrate that changes in tissue morphologic features observed in classical pathology are accompanied by, and may be correlated to, changes in the biochemical composition at the cellular level. Thus, these imaging methods provide novel insight into biochemical changes due to disease.
Article
Full-text available
The feasibility of characterizing human colorectal adenocarcinoma by IR microspectrometry is described. Carcinoma thin sections were analyzed by spatially resolved mid-IR FT microspectrometry, and for comparative purposes by conventional histological staining techniques. More than 2300 high quality FTIR reference spectra of 27 patient samples form 11 defined morphological structures such as crypts, tunica muscularis, submucosa and adenocarcinoma were recorded. The analysis of the spectral data included four steps: an initial test for spectral quality, data preprocessing, data reduction and classification of the tissue spectra by pattern recognition techniques. The overall classification accuracy attainted by an optimized ANN of about 95 percent highlights the great potential of FTIR microspectrometry as a diagnostic tool for the determination of a variety of tissue structures. Further improvements are necessary to make the new method applicable to routine and experimental clinical analysis in the future.
Article
Full-text available
FT-IR microspectrometry, particularly in combination with digital imaging techniques shows great promise for in-vivo and ex-vivo medical diagnosis. The statement is based on the knowledge that this method delivers information of the chemical structure and composition of a sample and the fact that any disease is linked to changes in the molecular and structural composition of cells and tissues. Typically, these changes are highly specific for a given tissue structure and are therefore potentially detectable by FT-IR microspectrometry. In this paper we present several approaches for the representation of mid-infrared microspectroscopic data acquired with high spatial resolution by the use of a MCT focal plane array detector. The applicability of image reassembling methodologies like functional group analysis, image reconstruction based on factor analysis and artificial neural network analysis to the IR data is discussed.
Article
Die Dissertation hat sich mit der Aufgabe befasst, durch Kombination von IR-Spektroskopie und chemometrischen Auswertungsalgorithmen eine Differenzierung und Klassifizierung von Hirnmetastasen-Dünnschnitten zu erreichen. Die Untersuchungen konzentrieren sich dabei auf jene fünf Primärtumoren, die besonders oft Metastasen im Gehirn bilden. Das sind kolorektale Karzinome, Mammakarzinome, maligne Melanome, Nierenzellkarzinome und Bronchialkarzinome. Metastasen tragen die molekularen Informationen der Gewebezellen des Primärtumors in sich. Die Anwendung von IR-spektroskopischen Methoden stellt deshalb einen innovativen Ansatz zur Identifikation des Primärtumors von Hirnmetastasen dar, da die Spektren einem molekularen Fingerabdruck entsprechen. Als Klassifizierungsalgorithmen wurden SIMCA (soft independent modeling of class analogies) und ANN (artificial neural networks) herangezogen. Die Entwicklung der Klassifizierungsverfahren gliederte sich in drei Teile. Im ersten Teil wurden Trainingsmodelle mit den ausgewählten homogenen Bereichen der Metastasengewebeschnitte erstellt und an unabhängigen Daten weiterer Proben bekannter und unbekannter Organherkunft getestet. Im zweiten Teil wurden die Modelle mit Hilfe homogener Tumorzelllinien optimiert und auf die Zuordnung der Hirnmetastasen zu den Primärtumoren angewandt. Eine zweistufige Klassifizierungsstrategie gewährleistet damit eine Genauigkeit der Klassifizierung von über 80%.