Accessible Published by De Gruyter September 22, 2018

Optical imaging methods in medicine: how can we escape the plausibility trap?

Werner Nahm, Christoph Hornberger, Ute Morgenstern and Stephan B. Sobottka

Humans receive nearly 80% of their perceptional input from their visual sense. Therefore, it is not surprising that typically visual inspection is the introduction to any physical examination for medical diagnosis. We intuitively tend to take the things we see for real as long as what we see is not in conflict with what we know. With the development of analog and digital photography and modern video technology, we have learned to transfer this claim into reality on images as well. As a result of this development, we are in a constant conflict between our rational understanding of the fallibility of visual perception on the one hand and on the other hand the ability of our visual perception to continuously seek for analogies and similarities in what we see.

Problems may arise if we tend to interpret images based on what we take as plausible from our personal base of knowledge. If we define plausibility as in congruency with individual expertise or common sense [1] our conclusions may no longer be based on the scientific method, even if they turn out to be true. Two examples for plausible interpretations of images in medical diagnosis can be considered for discussion: the difference in the interpretation of a magnetic resonance image, saying either “I see a tumor in a certain place” or “I see a defined nuclear magnetic resonance contrast in a certain place” or the difference in interpreting a spectrally resolved camera image, saying either “I see hyper-perfusion in a certain area” or “I see a defined spectral signature in a certain area”. In both examples the diagnostic statement can be reproduced by any scientist only if the hypothetical relation between a measured quantity and a physiologically or anatomically interpretable parameter can be understood with the help of validated deterministic models, based on natural laws, or if it can be supported by valid statistics.

From a measurement technology point of view, the optical principles of a human eye and a camera system are very similar. The basic physical measurement parameter is the number of photons per time unit received on a defined area of the detector. Those photons can be filtered or separated according to their energy. Furthermore, by means of wave optics we can analyze the radiation emitted or remitted from the object according to point of origin, wave front direction, state of polarization or phase shift and coherence, respectively. But we cannot make any statement about the history of photons or the path they travelled due to their indistinguishability. Therefore, we do not know how they got to their virtual point of origin on the object’s surface. The optical analysis of tissue is based on the assumption of mainly three different light-tissue interaction mechanisms: scattering, absorption and fluorescence. Hence, the photons received on the detector originate either from an external light source or from internal fluorophores and they have been either reflected on the surface or multiply scattered within the object, without being absorbed. In fact, that is all we can say about the physics of image generation. Beyond that, all analytical, morphological, functional or diagnostic conclusions require models based on assumptions regarding the history of the photons and regarding the relation between the optical properties of the tissue with physiological and anatomical parameters.

The expert committee “Optical Methods in Medicine” in the German Society of Biomedical Engineering (DGBMT) aims at bridging the gap between what engineers can measure and what physicians need to see. The first goal is to generate a mutual understanding for both, technical and diagnostic parameters. Starting from that, a next step would be the demand analysis for further studies needed to bring optical imaging methods onto a next level of confirmability and validity. The present special issue of Biomedical Engineering contains a selection of contributions that have been discussed in recent joint workshops amongst physicists, scientists, engineers and manufacturers in the field of optical imaging.

The key focus of the first four papers of this special issue is mainly on technical aspects of different optical measurement methods. Grosenick et al. [2] emphasize the time-domain approach for diffuse near infrared (NIR) imaging. The authors review the state of the art of optical tomography based on the time-of-flight (TOF) measurement of photons through deep tissue structures. The clinical applications mentioned are optical mammography and cortical functional imaging. Using those examples, the future diagnostic potential of combining temporal, spectral and spatial information is discussed for broader clinical application. The authors conclude that time-resolved methods provide very useful information on light propagation in tissue, even though system costs are still relatively high.

The technical aspects for the application of a camera set-up for hyperspectral imaging (HSI) in the field of wound diagnostics and therapy are described in the paper by Kulcke et al. [3]. Here, the working principle of an HSI camera based on the pushbroom approach is described in detail corresponding with algorithms for calculating characteristic tissue perfusion parameters from relative spectral intensities. The authors propose HSI as a method for generating evidence in clinical studies on wound treatment.

Heimpold et al. [4] propose an alternative approach for HSI using a multiplexed array of light-emitting diodes (LEDs) of different wavelengths in combination with a monochrome camera instead of a spectral camera together with a white light source. The technical focus of this paper is on the optimization of the LED combination to achieve a targeted spectral distribution.

Hornberger and Wabnitz [5] address the problem of the quantification of physiological parameters based on optical measurements. The authors present the state of the art of the different methods for calibration and validation in pulse oximetry for arterial oxygenation monitoring and tissue oximetry for monitoring of cerebral or muscle oxygenation. They also discuss the limitations and the potential of digital and physical calibration phantoms. The main conclusion of this paper is that the current calibration gold standard via in vivo tests on humans is still unsatisfying. New efforts should be undertaken to develop physical or digital calibration phantoms.

Besides hardware and measuring technology, algorithms for data analysis are important for system performance in optical imaging methods. The paper by Holmer et al. [6] addresses the problem of extracting clinical relevant parameters from three-dimensional (3D) data cubes of hyperspectral images. Parameters that are related to the supply of tissue with blood and oxygen are significant for every clinician. The paper describes some calibration steps necessary after image acquisition and presents algorithms to extract clinical relevant data which are based on the second derivative of the spectra and the known spectra of hemoglobin and other tissue chromophores. Parameters of interest are, besides others, tissue oxygenation and a tissue water index. Finally, the results of a volunteer occlusion study are presented where the calculated parameters are compared to the parameters determined by a commercial, non-imaging NIR oxygenation sensor.

Markgraf et al. [7] describe algorithms to compute tissue oxygen saturation (StO2) of ex vivo kidneys during normothermic machine perfusion (NMP) from HSI data. After thorough tests of three different algorithms on HSI data collected in a porcine blood flow model in a selected wavelength range from 590 to 810 nm, they found a discrete wavelength algorithm and a partial least squares regression (PLSR) to be best suited for the analysis of the quantitative change of the tissue oxygen status in organ perfusion experiments. The evaluation of the algorithms was performed in a reoxygenation experiment with porcine kidneys where an increase in StO2 was expected.

Clinicians often have to cope with the problem of combining the diagnostic information from different medical imaging methods. To discriminate tumor tissue from eloquent brain areas during surgery, besides other techniques, electrical stimulation combined with thermal imaging is used. Hoffmann et al. [8] introduce a framework to detect weak thermal signals of focal neural activity and separate it from background signals such as thermal waves induced by thermoregulation. For the filtering they use a semiparametric regression model. The intraoperative procedure is tested and the possibility of integration into the operation procedure is demonstrated whereas the reduction of the induced delay for the surgical intervention is proposed for further studies.

A problem with imaging methods in neurosurgery like infrared thermography is the disturbance by motion, e.g. pulsatile brain motion or motion induced by the patient or surgery devices which may hinder the accurate identification of cortical areas. Chen et al. [9] compare four motion correction algorithms, two intensity-based optical flow methods and two feature-based approaches, with regard to the accuracy of motion correction. To evaluate the correction algorithms, intraoperative images are used and the target registration error on manually selected marking points and the intensity temporal standard deviation are calculated. Depending on the type of motion and the extent of displacement, the advantages and disadvantages of the different algorithms differ. The authors propose a registration of thermal and white-light images to improve the motion correction in future work.

The third part of this special issue includes statements about application of optical imaging methods in clinical practice. It is a long way from optical raw data acquisition to interpretable images on the clinician’s monitor, at most attended by objectively established facts and arguments for decision support. Besides technical equipment for measuring optical parameters, software algorithms for recording, data processing and visualization and an intelligent protocol for optimal workflow fitting to clinical scenario, the focus is on practical boundary conditions, specified by clinical surroundings.

In the special field of neurosurgery, often new and promising technologies attract interest because of the complex and cross-linked problems to solve. Optical signal and image data acquisition in combination with traditional physico-chemical, electrical or electromagnetic measuring methods deliver information for characterizing biological tissue structures and functional properties. In research, all technically applicable methods are tested to get information in maximal spatial, temporal and spectral resolution. Galli et al. [10] refer to optical molecular imaging in human brain tissue obtained from pharmacoresistant epilepsy or brain tumor surgery. Multimodal multiphoton images from second harmonic generation (SHG) microscopy and Coherent anti-Stokes Raman scattering spectroscopy (CARS) are compared to Fourier-transform infrared (FT-IR) spectroscopic imaging to describe the density and distribution of corpora amylacea (CoA) in the pathological brain. The practical goal could be to find additional biomarkers for intraoperative assessment of pathological nervous structures.

In contrast, intraoperative optical imaging (IOI) is already in clinical use for identification of impaired tumor tissue and functional intact normal brain tissue. Several stimulation methods enable to detect functional areas during brain surgery. Oelschlägel et al. [11] verified metabolic changes of optical tissue parameters after direct cortical electrical stimulation to develop an additional clinical tool for guidance during tumor resection. After processing spatio-temporal image data presenting stimulation answers at 568 nm wavelength in a time course after simulation, a two-dimensional false colored map indicates cortical tissue functionality. Optical imaging during operation means applying tools and technologies under clinical restrictions concerning lighting, cameras, optical filters, computers and additional effort for testing.

Spectral information could be very useful to get additional structural and functional information about biological tissues. Tetschke et al. [12] uses Raman spectroscopy (RS), optical coherence tomography (OCT) and HSI to give an overview of optical and spectroscopic properties of healthy and carious human teeth in vitro. In dentistry, it is necessary to detect the beginning of demineralization in dental hard tissue, microstructural and chemical changes with caries on the surface. Each method shows special characteristics, more detailed than obtained by radiographic imaging. Especially HSI in visible (VIS) and NIR wavelength area supports early identification of demineralization.

Daeschlein et al. [13] and Sicher et al. [14] present HSI in wound diagnostics and therapy. Mechanisms of wound healing should be objectified to optimize tumor and wound healing therapy, for instance, application of cold atmospheric plasma (CAP). HSI enables to visualize microcirculation of large tissue surfaces. Special parameters derived from image data specify perfusion and oxygen saturation in chronic or acute wound areas. Comparing optical properties before and after CAP treatment, the results demonstrate effects on microcirculation. Superficial and also deeper oxygenation and hemoglobin perfusion can be improved [13]. Different hemodynamic disorders in patients suffering from scleroderma, Dupuytren contracture surgery, chronic foot ulcera and skin infections are examined by HSI [14]. Modern HSI camera equipment offers sufficient spectral, spatial and temporal resolution and is manageable also by non-technicians in clinical environment to get reproducible image data. But for calculating objectively comparable, tissue characterizing parameters for clinical interpretation, smart calculation algorithms based on intelligent tissue models are necessary. Every kind of tissue structure and functionality delivers special results dependent on technologies and measuring conditions. Thus, deviated perfusion or oxygenation parameters from brain, liver, wound tissue or teeth are separately defined only in small application fields. As far as camera equipment, set-ups and algorithms are newly introduced for optical imaging in medicine, basic experimental scientific work has to be done to establish powerful technology for challenging clinical tasks.

Minimally or non-invasive measuring is an important goal for medical application of innovative data acquisition technologies. Imaging is much more advantageous than local sensing because 2D maps or even 3D data cubes of parameter sets in space (x, y, z), time (t) and – speaking about optical methods – wavelengths (λ) can be obtained. Artifacts should be eliminated; signal-to-noise ratio should be maximized. Traditional methods, e.g. cardiovascular assessment, are based on local sensing. Non-invasive, contactless imaging of large areas of the patient’s surface allows extraction of cardiovascular parameters also from camera data. Biological activity of the imaged object (patient, living tissue or organs, e.g. for transplantation) provokes motion. In some imaging processes, motion variations need to be corrected from optical data such as an unwanted signal of noise like brain movement from rhythmic heart and respiration activity during neurosurgical operation [[9]. Other applications use these motion signals as mean information about physiological processes. Zaunseder et al. 15] provide an overview of imaging photoplethysmography (iPPG, or remote PPG) techniques for generating cardiovascular parameter characteristics. The most important parameter is the heart rate (HR), calculated from the recorded video stream from several channels, extracted after image and signal processing and analyzing steps. Heart rate variability and respiration related parameters can be generated reliably, if hardware equipment and analyzing algorithms work not only under experimental conditions with healthy test persons, but also in real-world medical applications with patients and in highly disturbed surroundings.

Optical imaging has many facets in technology, data processing and special applications in medical fields. Many specific technological and also implementation and interpretation problems, pros and cons, are known. The main task for biomedical engineers lies in ingenious combining of all available methods for optimizing the multimodal data acquisition process and for finding out how to extract and present essential information to solve the physician’s questions in diagnostics and treatment of patients.


[1] Bohnert M, Reszke P. Linguistisch-philosophische Untersuchungen zu Plausibilitat: uber kommunikative Grundmuster bei der Entstehung von wissenschaftlichen Tatsachen. In: Engelschalt J (Ed.); Maibaum A (Ed.): Auf der Suche nach den Tatsachen: Proceedings der 1. Tagung des Nachwuchsnetzwerks “INSIST”, 22.-23. Oktober 2014, Berlin. 2015.40-67. Search in Google Scholar

[2] Grosenick D, Wabnitz H, Macdonald R. Diffuse near-infrared imaging of tissue with picosecond time resolution. Biomed Eng-Biomed Tech 2018;63:511–8. Search in Google Scholar

[3] Kulcke A, Holmer A, Wahl P, Siemers F, Wild T, Daeschlein G. A compact hyperspectral camera for measurement of perfusion parameters in medicine. Biomed Eng-Biomed Tech 2018;63:519–27. Search in Google Scholar

[4] Heimpold T, Reifegerste F, Drechsel S, Lienig J. LED for hyperspectral imaging – a new selection method. Biomed Eng-Biomed Tech 2018;63:529–35. Search in Google Scholar

[5] Hornberger C, Wabnitz H. Approaches for calibration and validation of near-infrared optical methods for oxygenation monitoring. Biomed Eng-Biomed Tech 2018;63:537–46. Search in Google Scholar

[6] Holmer A, Marotz J, Wahl P, Dau M, Kämmerer PW. Hyperspectral imaging in perfusion and wound diagnostics – methods and algorithms for the determination of tissue parameters. Biomed Eng-Biomed Tech 2018;63:547–56. Search in Google Scholar

[7] Markgraf W, Feistel P, Thiele C, Malberg H. Algorithms for mapping kidney tissue oxygenation during normothermic machine perfusion using hyperspectral imaging. Biomed Eng-Biomed Tech 2018;63:557–66. Search in Google Scholar

[8] Hoffmann N, Radev Y, Koch E, Petersohn U, Steiner G, Kirsch M. Intraoperative mapping of the sensory cortex by time-resolved thermal imaging. Biomed Eng-Biomed Tech 2018;63:567–72. Search in Google Scholar

[9] Chen F, Müller J, Müller J, Müller J, Böhl E, Kirsch M, et al. Intraoperative motion correction in neurosurgery: a comparison of intensity- and feature-based methods. Biomed Eng-Biomed Tech 2018;63:573–8. Search in Google Scholar

[10] Galli R, Meinhardt M, Koch E, Schackert G, Steiner G, Kirsch M, et al. Optical molecular imaging of corpora amylacea in human brain tissue. Biomed Eng-Biomed Tech 2018;63:579–85. Search in Google Scholar

[11] Oelschlägel M, Meyer T, Schackert G, Kirsch M, Sobottka SB, Morgenstern U. Intraoperative optical imaging of metabolic changes after direct cortical stimulation – a clinical tool for guidance during tumor resection? Biomed Eng-Biomed Tech 2018;63:587–94. Search in Google Scholar

[12] Tetschke F, Kirsten L, Golde J, Walther J, Galli R, Koch E, et al. Application of optical and spectroscopic technologies for the characterization of carious lesions in vitro. Biomed Eng-Biomed Tech 2018;63:595–602. Search in Google Scholar

[13] Daeschlein G, Rutkowski R, Lutze S, von Podewils S, Sicher C, Wild T, et al. Hyperspectral imaging: innovative diagnostics to visualize hemodynamic effects of cold plasma in wound therapy. Biomed Eng-Biomed Tech 2018;63:603–8. Search in Google Scholar

[14] Sicher C, Rutkowski R, Lutze S, von Podewils S, Wild T, Kretching M, et al. Hyperspectral imaging as possible tool for visualization of changes in hemoglobin oxygenation in patients with deficient hemodynamics – proof of concept. Biomed Eng-Biomed Tech 2018;63:609–16. Search in Google Scholar

[15] Zaunseder S, Trumpp A, Wedekind D, Malberg H. Cardiovascular assessment by imaging photoplethysmography – a review. Biomed Eng-Biomed Tech 2018;63:617–34. Search in Google Scholar

Published Online: 2018-09-22
Published in Print: 2018-10-25

©2018 Walter de Gruyter GmbH, Berlin/Boston