Hyperspectral imaging is an optical technique that recently started being used in medical field. The correct extraction of spectral and spatial information from hyperspectral images depends on preprocessing, processing and analysis methods applied for an accurate diagnosis and monitoring medical treatments. A fundamental task in preprocessing hyperspectral images is the elimination of various types of noise generated by the hyperspectral systems. One of the major causes for the noise in a hyperspectral system is dark current noise. This type of noise arises from the temperature difference between environment and charge-coupled device of the hyperspectral camera. Electrons are generated over time and they are independent of the light falling on the detector. These electrons are captured by the potential wells of the charge-coupled device and counted as signal. The dark current noise removal can lead to an improvement in the performance of classification, target detection, anomaly detection and mapping methods, thus contributing to a better and more accurate diagnosis. Two denoising techniques - principal component analysis and minimum noise fractions were used until now in medical hyperspectral imaging applications. In this paper, the wavelet transform was proposed as a denoising technique for medical applications. The study was performed in both laboratory and clinical conditions. Two hyperspectral systems were used for the hyperspectral images acquisition of rabbit liver and a burn wound located on the posterior side of the patient left leg respectively using the same pushbroom hyperspectral camera but with two different scanning components (translation table and scanning mirror). The pushbroom hyperspectral camera acquires the image collecting the x-axis and λ information completely at the same time for a line on the y-axis. The two scanning components are used to move the sample (liver or patient leg) across the field of view of the hyperspectral camera so that the images are acquired line by line. The experimental results showed that the proposed denoising technique achieves better performance when applied to hyperspectral images acquired under laboratory conditions than in clinical situations. In conclusion, the wavelet transform could be considered a successful approach to denoising in laboratory hyperspectral measurements.
Accurate diagnosis of burns, mainly in terms of depth and healing potential, has still remained an unsolved clinical problem. Hyperspectral imaging, with its unique capabilities to simultaneously provide both spatial and spectral information, can be considered as a particularly useful tool in early diagnosis of burns by providing accurate and valuable information about injured biological tissues. In this study, the potential of hyperspectral imaging to generate burn characteristics maps was evaluated. Two supervised classification methods (spectral angle mapper and support vector machine) of the hyperspectral data were investigated and their classification accuracy was compared. The study was performed on a 24 hours old burn of the hand (superficial-partial and deep-partial thickness burn wound). A pushbroom hyperspectral imaging system was used to acquire the hyperspectral image of the burn wound within the wavelength range from 400 nm to 800 nm. The hyperspectral image was calibrated with respect to the white and dark reference images in order to minimize the influences of light intensity variations across the spatial scanning lines and the dark current in the hyperspectral system. Minimum noise fraction transform was used to determine the inherent dimensionality of hyperspectral data and to separate the information from noise before the calibrated hyperspectral image being analyzed using the spectral angle mapper and support vector machine classifiers. The accuracy of these two classification methods in mapping the skin burn characteristics was evaluated based on the classification accuracy assessment of the resulted skin burns characteristic maps. The results revealed that the overall classification accuracy of support vector machine classifier exceeded (overall accuracy = 91.94 % and Kappa coefficient = 0.902) that of the spectral angle mapper classifier (overall accuracy = 84.13 % and Kappa coefficient = 0.808). In conclusion, these preliminary data suggest that hyperspectral imaging combined with support vector machine classifier could play an important role in burn characterization and mapping.
Hyperspectral imaging is a technology that is beginning to occupy an important place in medical research with good prospects in future clinical applications. We evaluated the role of hyperspectral imaging in association with a mixture-tuned matched filtering method in the characterization of open wounds. The methodology and the processing steps of the hyperspectral image that have been performed in order to obtain the most useful information about the wound are described in detail. Correlations between the hyperspectral image and clinical examination are described, leading to a pattern that permits relative evaluation of the square area of the wound and its different components in comparison with the surrounding normal skin. Our results showed that the described method can identify different types of tissues that are present in the wounded area and can objectively measure their respective abundance, which proves its value in wound characterization. In conclusion, the method that was described in this preliminary case presentation shows promising results, but needs further evaluation in order to become a reliable and useful tool.