Asgeir Bjorgan

Estimation of skin optical parameters for real-time hyperspectral imaging applications

Abstract. Hyperspectral imaging combines high spectral and spatial resolution in one modality. This imaging technique is a promising tool for objective medical diagnostics. However, to be attractive in a clinical setting, the technique needs to be fast and accurate. Hyperspectral imaging can be used to analyze tissue properties using spectroscopic methods, and is thus useful as a general purpose diagnostic tool. We combine an analytic diffusion model for photon transport with real-time analysis of the hyperspectral images. This is achieved by parallelizing the inverse photon transport model on a graphics processing unit to yield optical parameters from diffuse reflectance spectra. The validity of this approach was verified by Monte Carlo simulations. Hyperspectral images of human skin in the wavelength range 400–1000 nm, with a spectral resolution of 3.6 nm and 1600 pixels across the field of view (Hyspex VNIR-1600), were used to develop the presented approach. The implemented algorithm was found to output optical properties at a speed of 3.5 ms per line of image data. The presented method is thus capable of meeting the defined real-time requirement, which was 30 ms per line of data.The algorithm is a proof of principle, which will be further developed.

Real-Time Noise Removal for Line-Scanning Hyperspectral Devices Using a Minimum Noise Fraction-Based Approach

Processing line-by-line and in real-time can be convenient for some applications of line-scanning hyperspectral imaging technology. Some types of processing, like inverse modeling and spectral analysis, can be sensitive to noise. The MNF (minimum noise fraction) transform provides suitable denoising performance, but requires full image availability for the estimation of image and noise statistics. In this work, a modified algorithm is proposed. Incrementally-updated statistics enables the algorithm to denoise the image line-by-line. The denoising performance has been compared to conventional MNF and found to be equal. With a satisfying denoising performance and real-time implementation, the developed algorithm can denoise line-scanned hyperspectral images in real-time. The elimination of waiting time before denoised data are available is an important step towards real-time visualization of processed hyperspectral data. The source code can be found at http://www.github.com/ntnu-bioopt/mnf. This includes an implementation of conventional MNF denoising.

Wavelet based feature extraction and visualization in hyperspectral tissue characterization

Hyperspectral images of tissue contain extensive and complex information relevant for clinical applications. In this work, wavelet decomposition is explored for feature extraction from such data. Wavelet methods are simple and computationally effective, and can be implemented in real-time. The aim of this study was to correlate results from wavelet decomposition in the spectral domain with physical parameters (tissue oxygenation, blood and melanin content). Wavelet decomposition was tested on Monte Carlo simulations, measurements of a tissue phantom and hyperspectral data from a human volunteer during an occlusion experiment. Reflectance spectra were decomposed, and the coefficients were correlated to tissue parameters. This approach was used to identify wavelet components that can be utilized to map levels of blood, melanin and oxygen saturation. The results show a significant correlation (p <0.02) between the chosen tissue parameters and the selected wavelet components. The tissue parameters could be mapped using a subset of the calculated components due to redundancy in spectral information. Vessel structures are well visualized. Wavelet analysis appears as a promising tool for extraction of spectral features in skin. Future studies will aim at developing quantitative mapping of optical properties based on wavelet decomposition.

Mirror Mirror on the Wall... An Unobtrusive Intelligent Multisensory Mirror for Well-Being Status Self-Assessment and Visualization

A persons well-being status is reflected by their face through a combination of facial expressions and physical signs. The SEMEOTICONS project translates the semeiotic code of the human face into measurements and computational descriptors that are automatically extracted from images, videos, and three-dimensional scans of the face. SEMEOTICONS developed a multisensory platform in the form of a smart mirror to identify signs related to cardio-metabolic risk. The aim was to enable users to self-monitor their well-being status over time and guide them to improve their lifestyle. Significant scientific and technological challenges have been addressed to build the multisensory mirror, from touchless data acquisition, to real-time processing and integration of multimodal data.

Can spectral-spatial image segmentation be used to discriminate experimental burn wounds?

Abstract. Hyperspectral imaging (HSI) is a noncontact and noninvasive optical modality emerging the field of medical research. The goal of this study was to determine the ability of HSI and image segmentation to discriminate burn wounds in a preclinical porcine model. A heated brass rod was used to introduce burn wounds of graded severity in a pig model and a sequence of hyperspectral data was recorded up to 8-h postinjury. The hyperspectral images were processed by an unsupervised spectral–spatial segmentation algorithm. Segmentation was validated using results from histology. The proposed algorithm was compared to K-means segmentation and was found superior. The obtained segmentation maps revealed separated zones within the burn sites, indicating a variation in burn severity. The suggested image-processing scheme allowed mapping dynamic changes of spectral properties within the burn wounds over time. The results of this study indicate that unsupervised spectral–spatial segmentation applied on hyperspectral images can discriminate burn injuries of varying severity.

Hyperspectral imaging for detection of cholesterol in human skin

Hypercholesterolemia is characterized by high levels of cholesterol in the blood and is associated with an increased risk of atherosclerosis and coronary heart disease. Early detection of hypercholesterolemia is necessary to prevent onset and progress of cardiovascular disease. Optical imaging techniques might have a potential for early diagnosis and monitoring of hypercholesterolemia. In this study, hyperspectral imaging was investigated for this application. The main aim of the study was to identify spectral and spatial characteristics that can aid identification of hypercholesterolemia in facial skin. The first part of the study involved a numerical simulation of human skin affected by hypercholesterolemia. A literature survey was performed to identify characteristic morphological and physiological parameters. Realistic models were prepared and Monte Carlo simulations were performed to obtain hyperspectral images. Based on the simulations optimal wavelength regions for differentiation between normal and cholesterol rich skin were identified. Minimum Noise Fraction transformation (MNF) was used for analysis. In the second part of the study, the simulations were verified by a clinical study involving volunteers with elevated and normal levels of cholesterol. The faces of the volunteers were scanned by a hyperspectral camera covering the spectral range between 400 nm and 720 nm, and characteristic spectral features of the affected skin were identified. Processing of the images was done after conversion to reflectance and masking of the images. The identified features were compared to the known cholesterol levels of the subjects. The results of this study demonstrate that hyperspectral imaging of facial skin can be a promising, rapid modality for detection of hypercholesterolemia.

Identification of inflammation sites in arthritic joints using hyperspectral imaging

Inflammatory arthritic diseases have prevalence between 2 and 3% and may lead to joint destruction and deformation resulting in a loss of function. Patient’s quality of life is often severely affected as the disease attacks hands and finger joints. Pathology involved in arthritis includes angiogenesis, hyper-vascularization, hyper-metabolism and relative hypoxia. We have employed hyperspectral imaging to study the hemodynamics of affected- and non-affected joints and tissue. Two hyperspectral, push-broom cameras were used (VNIR-1600, SWIR-320i, Norsk Elektro Optikk AS, Norway). Optical spectra (400nm – 1700nm) of high spectral resolution were collected from 15 patients with visible symptoms of arthritic rheumatic diseases in at least one joint. The control group consisted of 10 healthy individuals. Concentrations of dominant chromophores were calculated based on analytical calculations of light transport in tissue. Image processing was used to analyze hyperspectral data and retrieve information, e.g. blood concentration and tissue oxygenation maps. The obtained results indicate that hyperspectral imaging can be used to quantify changes within affected joints and surrounding tissue. Further improvement of this method will have positive impact on diagnosis of arthritic joints at an early stage. Moreover it will enable development of fast, noninvasive and noncontact diagnostic tool of arthritic joints

Vessel contrast enhancement in hyperspectral images

Imaging of vessel structures can be useful for investigation of endothelial function, angiogenesis and hyper-vascularization. This can be challenging for hyperspectral tissue imaging due to photon scattering and absorption in other parts of the tissue. Real-time processing techniques for enhancement of vessel contrast in hyperspectral tissue images were investigated. Wavelet processing and an inverse diffusion model were employed, and compared to band ratio metrics and statistical methods. A multiscale vesselness filter was applied for further enhancement. The results show that vessel structures in hyperspectral images can be enhanced and characterized using a combination of statistical, numerical and more physics informed models.

Towards real-time medical diagnostics using hyperspectral imaging technology

Hyperspectral imaging provides non-contact, high resolution spectral images which has a substantial diagnostic potential. This can be used for e.g. diagnosis and early detection of arthritis in finger joints. Processing speed is currently a limitation for clinical use of the technique. A real-time system for analysis and visualization using GPU processing and threaded CPU processing is presented. Images showing blood oxygenation, blood volume fraction and vessel enhanced images are among the data calculated in real-time. This study shows the potential of real-time processing in this context. A combination of the processing modules will be used in detection of arthritic finger joints from hyperspectral reflectance and transmittance data.

Spectral-spatial classification combined with diffusion theory based inverse modeling of hyperspectral images

Hyperspectral imagery opens a new perspective for biomedical diagnostics and tissue characterization. High spectral resolution can give insight into optical properties of the skin tissue. However, at the same time the amount of collected data represents a challenge when it comes to decomposition into clusters and extraction of useful diagnostic information. In this study spectral-spatial classification and inverse diffusion modeling were employed to hyperspectral images obtained from a porcine burn model using a hyperspectral push-broom camera. The implemented method takes advantage of spatial and spectral information simultaneously, and provides information about the average optical properties within each cluster. The implemented algorithm allows mapping spectral and spatial heterogeneity of the burn injury as well as dynamic changes of spectral properties within the burn area. The combination of statistical and physics informed tools allowed for initial separation of different burn wounds and further detailed characterization of the injuries in short post-injury time.