Open Access
Add to BibSonomyAbstract
Object visualization is influenced by the spectral distribution of an illuminant impinging upon it. In this paper, we proposed a color entropy evaluation method to provide the optimal illumination that best helps surgeons distinguish tissue features. The target-specific optimal illumination was obtained by maximizing the color entropy value of our sample tissue, whose spectral reflectance was measured using multispectral imaging. Sample images captured under optimal light were compared with that under commercial white light emitting diodes (3000K, 4000K and 5500K). Results showed images under the optimized illuminant had better visual performance such as more subtle details exhibited.
© 2016 Optical Society of America
1. Introduction
For many years, illumination of wounds during surgery has been done by surgical lighting system (SLS) [1]. The luminaire has been designed such that high-intensity light is supplied to the wound while minimizing shadows of heads and hands of the surgical team. The requirements for surgical lighting have remained unchanged over the past several years–illumination level, shadow and glare reduction and reduction of the heat impinging on the illuminated area are the dominant factors in choosing the lighting system [2].
Traditional surgical light typically consists of a metallic reflector with a halogen or high-pressure lamp in its focal point, such as arc lamps embedded in the laparoscopy as endoscopic illumination [3]. They meet the dominant brightness requirements for surgical lighting over the past several years. But conventional light sources have several limitations such as low efficiency, waste of energy and short lifetime. For example, arc-lamp used in the laparoscopy has large power consumption due to its inefficient electrical to light power conversion. As lost energy is converted into heat, the temperature can reach 95°C at the end of the endoscope and 239°C at the fiber optic bundle of light cables [4].
Recently, more attention has been paid with the adoption of cooler Light Emitting Diode (LED) arrays for overhead surgical lighting as well as task lighting. LEDs are solid-state semiconductor devices which convert electrical potential energy to electromagnetic energy in the form of light. Today, LEDs are available in a variety of colors and offer positive characteristics: compactness, adjustable intensity, long lifetime and energy saving potential [5,6]. With the rapid development of solid-lighting industry, LED has been widely used in various new lighting applications [7]. For example, in indoor lighting area, Corell et al. explored an alternative LED-based solution to replace the existing yellow fluorescent light tubes (YFT) used in photolithography rooms [8]. In energy saving applications, Tsuei et al. presented a hybrid method for using sunlight and LED illumination powered by renewable solar energy for indoor lighting [9].
What’s more, many researches have been done on the spectral optimization to enhance tissue visualization when LEDs were applied in the SLS. For example, Wang et al. proposed a color image reconstruction method that enhanced visualization of oral cavity for early inflamed tissue detection [10]. The enhanced visualization comes from the white light color mixing between the red normal tissue and bluish white light from the LEDs. Murai et al. designed the optimal white illuminant by combining different kinds of LEDs to enhance the color difference between healthy and focal tissues [11]. They found that the optimal illuminant can clarify the blood circulation better than the conventional illuminant as keeping its whiteness.
Different from the traditional spectrum-based imaging modalities and technologies mainly focused on estimation and analysis of differences of the spectral reflectance, color entropy was utilized here to achieve optimal SLS illuminant with more tissue details presented. Illuminant with different optimal spectrum was obtained according to different ROI (region of interest) chosen in the sample image. We found that tissue image captured under optimal illuminant got higher tissue contrast than the ones under commercial white light in 3000K, 4000K, 5500K and the traditional halogen light. The advantage of optimal illuminant was confirmed by image frequency analysis. In addition, LEDs with different spectra can be used, providing more choices in the surgical light mixing process.
2. Method
Spectral reflectance is the fingerprint of an object. The object’s appearance is determined by illumination, the spectral reflectance of objects, and the spectral sensitivity of the imaging sensor [12]. We proposed color entropy evaluation to explore the optimal illumination conditions that best display the target tissues with more details offered and clearer texture presented.
2.1 Color entropy
Tissue contrast and the amount of details target tissues offered was quantitated using standard ‘Entropy’ analysis. In information theory, the concept of entropy is used to quantify the amount of information necessary to describe the macrostate of a system [13]. The entropy is related to the concept of Kolmogorov complexity, which reflects the information content of a sequence of symbols independent of any particular probability model [14,15]. More specifically, the Kolmogorov complexity of an object is a measure of the computational resources needed to specify the object. Then, if a system has a high value of entropy, it means that much information is presented along the system.
Depending on the various application, entropy can be specifically defined in different ways [16,17]. Here we take the concept of entropy in the sense of image information theory, where the Entropy of an image is a statistical measure of randomness that can be used to characterize the texture of the input image. Traditional Entropy analysis was aimed at the grayscale image evaluation, which focused on the gray value contrast while ignored the color information [18]. But to present surgery, either for operation observed by eye directly or through image display, color information is both necessary. Therefore, color entropy was specifically proposed here to evaluate the tissue color texture. Entropy is defined as:
where p refers to the distribution of gray level in the image or the intensity of different color component. pi is obtained by counting the probability of pixel i with a given color intensity ((red (R), green (G) or blue (B)), where i can vary from 0 to 255 and pi can vary from 0 to 1. Entropy is related to the information provided by the image. In this case, if the gray values or intensity of color components are exactly same to all pixels in one image, the image will present minimal entropy value. Conversely, if all pixels of the image present totally different gray values or the color intensity, the image will get maximum entropy value [19].As discussed above, color information and texture are vital to the operation. Since different textures result in different distribution of gray level or color intensity, it means that the entropy described above can be used for texture characterization. Color entropy here is redefined as:
where EC represents the entropy value in given color component Red, Green and Blue, E represents the color entropy value including all color component information. In this paper, color entropy E was used as the evaluation value to optimize the object illuminant. Higher E means better illuminant performance.2.2 Illuminant spectrum optimization
The light emitted from an illumination source is reflected from a scene element and then enters an imaging device. Thus, the color of a scene element is obtained by integrating the spectral power distribution of the illumination source, the spectral reflectance of the object, and the spectral sensitivity of the digital imaging device.
Let RC(λ), S(x, y, λ) and I(λ) represent the spectral response of a camera in color channel C the spectral reflectance of a scene point with the position coordinate (x, y), and the spectral power distribution of an illumination at the same scene element, respectively. Then the color intensity value in channel C measured at a pixel is given by
where ROI refers to the chosen area of the specimen illuminated and analyzed.Color intensity distribution in each color channel can be obtained by
The optimal illumination spectrum that provides highest color entropy value (Eq. (3)) can be derived through the following maximization formula:
Because it is unrealistic to create an arbitrary spectrum, and if the controllable LED light sources with spectra I1, I2,…, In are given, the Eq. (6) can be rewritten as the problem of finding the best linear combination of illumination sources as follows:
where L is a matrix consisting of source vector I1, I2,…, In and x represents a weight vector for LED light sources.Because Eq. (7) is nonlinear, particle swarm optimization (PSO) was used in this paper to find the optimal weight vector x. PSO is a popular search algorithm with its high efficiency and few adjusting parameters. It can be applied to solve most optimization problems and problems that can be converted to optimization deriving [20].
3. Experiment and results
3.1 Optimal spectral distribution acquisition
In order to obtain an optimal illuminant by Eq. (7), four steps are essential: (1) choosing the sample illuminated for analysis, (2) the spectral reflectance data of each point in the ROI should be measured, (3) specifications of the imaging device and different kinds of LED sources were given for light mixing and spectrum optimization, (4) the optimized SPD should be calculated based on the information from step (2) and (3).
3.1.1 Specimen
Cleaned porcine meat including both fat and lean tissues was used. Porcine heart tissues were adopted. The ethics committee of Zhejiang University approved the study.
3.1.2 Spectral reflectance obtainment
In this study, we used a multispectral imaging system to measure the spectral reflectance of the sample. It was self-developed by Department of Information Science and Electronic Engineering of Zhejiang University [21]. As illustrated in Fig. 1, the multispectral imaging system is made up of a monochrome camera and a filter wheel. The filter wheel contains 11 filters and is installed between the camera and the lens. With this system, the spectral reflectance of a sample can be obtained at pixel-level resolution and with a spectral resolution ranging from 400 to 700 nm with 10-nm intervals. In this system, they developed an improved Wiener estimation method for the reconstruction of spectral reflectance without a prior knowledge of the samples being imaged, by the adaptive selection of training samples. The 11 filters were all specifically designed. With adaptive wiener estimation method, only 11 image channels were used in the system and it can output 30 reflectance data with 10-nm interval. Since spectral reflectance is vital to the acquisition of the optimal spectrum, repeated measurements were conducted here and noise processing was carried out in the spectral reflectance modeling.
3.1.3 LED sources and imaging device
The light source used here is the LED lighting box constructed by Zhejiang University, which is an LED panel with the size of 0.7 m × 0.4m integrated with 128 LEDs (11 kinds of LEDs, sketch shown in Fig. 2). The 11 high power LEDs consisted of 8 color LEDs and 3 white LEDs with different CCTs, whose spectral distribution (SPD) were all given by Fig. 3. Luminance level of each LED can be controlled via the LED control circuit and a software was developed to generate the light with the target spectral distribution based on a light matching algorithm and the feedback signal from a spectrometer.
The spectrum of both single original monochromatic LED light and generated mixing LED light was acquired for calibration using a spectroradiometer. The Spectrometer adopted is PR655 SpectraScan produced by Photo Research Inc. For PR655, the measurement range covers 380 nm to 780 nm with an average band width of 4 nm. The spectral data also can be transported through the USB port.
Images of specimens were captured using a CCD camera. A commercially available 8-bit CCD camera DFK 31BU03 made by the Imaging Source with the pixel resolution of 1024 × 768 was adopted here. It is interfaced using Universal Serial Bus (USB2.0). Output image files were processed by self-developed software based on Matlab. The spectral responses of three color channels have been illustrated in Fig. 4.
3.1.4 Illuminant spectral distribution optimization
The first example using porcine heart tissue is shown in Fig. 5. Regions of interest A-E chosen for analysis are indicated covering cardiac muscle, cardiac muscle/artery, fat/artery, cardiac muscle/fat and cardiac muscle/artery/fat separately. Spectral reflectance data of the specimen in five ROIs were measured through step (2). In our work, the optimization was also constrained by the physical constraints of LEDs adopted. To solve this problem and obtain the optimal illuminant with a specific spectrum, maximal flux of every selected LED channel was measured and the upper limit of the final achievable lightness was carried out as Imax. According to the xopt and Imax obtained, optimized spectral distribution was calculated for the five regions and shown in Fig. 6.
The second example using porcine meat is shown in Fig. 7. ROI F was chosen to cover both fat and lean meat while ROI G was filled with fat only. Optimized spectral distributions were also calculated and displayed.
3.2 Contrast enhancement with optimized illuminant
Optimized illuminants for different tissue segments ROI A-G (spectrum shown in Figs. 6 and 7) were used on porcine tissues with ROI a-g, which one on one covered the same kinds of tissues referring to ROI A-G. These ROIs chosen for analysis were specifically indicated in Fig. 8.
Figures 9(a)-9(d) show the specimen captured under commercial 3000K (CRI 80), commercial 4000K (CRI 75), commercial 5500K (CRI 70) and traditional halogen light, and for better comparison, parts (ROI a-e) of which were replaced by the images captured under the optimized illuminant in Fig. 6. Repeated experiments were conducted in the processing of noise. Color entropy value of these segments were also calculated and showed in Fig. 10. Results shown that images of tissues under optimized lights got higher color entropy value with more details presented than the ones under commercial white illuminants and traditional halogen light. Analogously, the same approach was applied on porcine meat sample (shown in Fig. 11) and color entropy values were showed in Fig. 12. The comparison results indicated that illuminants with optimized spectrum offered better lighting performance with clear tissue texture and more subtle details exhibited than the other comparison lights.
Fig. 9 Images of porcine heart captured under (a) commercial 3000K, (b) commercial 4000K, (c) commercial 5500K and (d) traditional halogen light. Selected regions (ROI a-e) were replaced by the images captured under the optimized illuminants in Fig. 6. All the compared images were captured under same luminance.
Fig. 10 Color entropy values of five image segments (ROI a-e shown in Fig. 8) captured under optimal light, commercial 3000K, commercial 4000K, commercial 5500K and halogen light.
Fig. 11 Images of porcine meat captured under (a) commercial 3000K, (b) commercial 4000K, (c) commercial 5500K and (d) traditional halogen light. Selected regions (ROI f-g) were replaced by the images captured under the optimized lights shown in Fig. 7. All the images were captured and compared under same luminance.
Fig. 12 Color entropy values of two image segments (ROI f-g shown in Fig. 8) captured under optimal light, commercial 3000K, commercial 4000K, commercial 5500K and halogen light.
3.3 New approach validation
Color entropy was specifically proposed in our paper to evaluate the image quality under kinds of illuminant. To verify the effectiveness of our method, spatial frequency analysis was adopted here. The ratio of high-frequency components is an important index to evaluate the amount of detailed information in an image [22,23]. Color images captured under different kinds of illuminants were transformed into frequency domain through FFT, in which zero Frequency had been shifted in the center. Annular frequency histogram was introduced to represent frequency distribution feature. Suppose the frequency domain image was split by n concentric circles. The radius difference of two adjacent circles was equivalent. Let Ri (1≤ i ≤n) be the radius of the circle i and | fi | be the total intensity of frequency components in the annulus i. Here high frequency component intensity | Fi | is defined as follow:
These are illustrated in Fig. 13.
Here | Fi | represents the high-frequency component intensity in the region covering annulus and outer annuluses. For the same i, higher | Fi | value means that image contained more high-frequency component and more subtle details are exhibited. Figure 14 shows | Fi | values of ROI (a-g) chosen in the porcine tissue images under optimized illuminant, commercial white light and traditional halogen light. Results showed that image captured under optimal light owned higher | Fi | values than the ones under other comparison light. Illuminant with optimal spectrum offered better lighting performance with HD tissue definition than commercial light and halogen light. Thus, the same result was obtained using the spatial frequency analysis and color entropy methodology.
Fig. 14 | Fi | values of ROI (a-g) chosen in porcine tissue images illustrated in Fig. 8 under optimal light, commercial 3000K, commercial 4000K, commercial 5500K and halogen light.
4. Discussion and conclusions
We have successfully substituted an array of LEDs for traditional surgical lamps to provide bright-field illumination for SLS. The LEDs offered several advantages over traditional light sources, including reduced size which facilitates assembly into arrays, high efficiency and less infrared emission. A significant advantage of LEDs is that their output can be summed to produce a desired spectral distribution, which enhances the specific tissue contrast and features. Here 11 kinds of LEDs were adopted in this study to create the illuminant with optimized spectrum. The increasing availability of relatively inexpensive bright LEDs of various colors can be expected to make them an attractive alternative to halogen lamps as a routine source of illumination.
For spectral optimization, color entropy was proposed in this paper. The illuminant with optimal spectral distribution can provide effective visualization of enhanced specific features. A variety of LEDs were used to show that the proposed optimization with a given set of illumination sources is efficient and practical. To get higher contrast and better image quality of sample tissues, the optimal linear combination of the LEDs was computed using Eq. (7). The optimized spectrums are shown in Figs. 6 and 7. Figures 9–11 show the porcine heart and meat images captured under commercial white light in 3000K, 4000K, 5500K and traditional halogen light, part regions of which are replaced by the image under optimized LED illumination combination. The tissues are distinguished better under the optimized illuminant than the other comparison lights. Spatial frequency analysis was adopted in the later chapter to validate the new approach. The same result was obtained that illuminant optimized using our color entropy method showed better lighting performance.
This method has the potential to provide colored illumination real-time enhancement of tissue contrast in vivo. The rise and decay times of LED emission is rapid enough to change color combinations to enhance contrast in dynamic color scenes in vivo. Another big advantage of this proposed approach is its flexibility. We can easily apply target-specific illumination. But until now we still used ex-vivo tissue in the experiment because it was hard to measure the tissue spectral reflectance real-timely. In the later experiment, we may consider combining the endoscopy and multispectral imaging technology, thus more flexible detecting and rapid lighting methodology would be possible. Meanwhile, useful applications of high temporal resolution of illumination would be to ‘segment’ images by differentially strobing color features, or to enhance contrast in selected features by intermittently illuminating. The limitation of the method in this paper is that CCD is the main visual imaging method, but many operations in the SLS are completed directly through eyes. Thus, more work remains to be done regarding to human visual system.
In future, we will pilot-used this approach in more real SLS cases, and update the current system by building an LED illumination endoscope combined with multispectral imaging system to enhance tissue features and achieve color versatility.
Funding
National Natural Science Foundation of China (NSFC) (61327902)
Acknowledgment
The authors acknowledge funding support from the National Natural Science Foundation of China.
References and links
1. A. J. Knulst, R. Mooijweer, F. W. Jansen, L. P. Stassen, and J. Dankelman, “Indicating shortcomings in surgical lighting systems,” Minim. Invasive Ther. Allied Technol. 20(5), 267–275 (2011). [CrossRef] [PubMed]
2. M. Litorja, S. W. Brown, M. E. Nadal, D. Allen, and A. Gorbach, “Development of surgical lighting for enhanced color contrast,” in Medical Imaging, (International Society for Optics and Photonics, 2007), 65150K–65150K–65111.
3. S. Hadrath and N. Morgenbrod, “A design approach for an innovative LED surgical light,” Proceedings of SPIE - The International Society for Optical Engineering 9293(2014). [CrossRef]
4. C. Hensman, G. B. Hanna, T. Drew, H. Moseley, and A. Cuschieri, “Total radiated power, infrared output, and heat generation by cold light sources at the distal end of endoscopes and fiber optic bundle of light cables,” Surg. Endosc. 12(4), 335–337 (1998). [CrossRef] [PubMed]
5. F. Viénot, E. Mahler, J. J. Ezrati, C. Boust, A. Rambaud, and A. Bricoune, “Color appearance under LED illumination: The Visual Judgment of Observers,” J. Light Visual Environment 32(2), 208–213 (2008). [CrossRef]
6. F. Nguyen, “Realizing LED illumination lighting applications,” Proceedings of SPIE - The International Society for Optical Engineering 5941, 594105–594105–594106 (2005).
7. H. C. Wang and Y. T. Chen, “Optimal lighting of RGB LEDs for oral cavity detection,” Opt. Express 20(9), 10186–10199 (2012). [CrossRef] [PubMed]
8. D. Corell, H. Ou, C. Dam-Hansen, P. M. Petersen, and D. Friis, “Light emitting diodes as an alternative ambient illumination source in photolithography environment,” Opt. Express 17(20), 17293–17302 (2009). [CrossRef] [PubMed]
9. C. H. Tsuei, W. S. Sun, and C. C. Kuo, “Hybrid sunlight/LED illumination and renewable solar energy saving concepts for indoor lighting,” Opt. Express 18(S4Suppl 4), A640–A653 (2010). [CrossRef] [PubMed]
10. H.-C. Wang, Y.-T. Chen, J.-T. Lin, C.-P. Chiang, and F.-H. Cheng, “Enhanced visualization of oral cavity for early inflamed tissue detection,” Opt. Express 18(11), 11800–11809 (2010). [CrossRef] [PubMed]
11. K. Murai, H. Kawahira, and H. Haneishi, “Improving color appearance of organ in surgery by optimally designed LED illuminant,” in World Congress on Medical Physics and Biomedical EngineeringMay 26–31, 2012, Beijing, China (Springer, 2013), 1010–1013. [CrossRef]
12. M.-H. Lee, D.-K. Seo, B.-K. Seo, and J.-I. Park, “Optimal illumination for discriminating objects with different spectra,” Opt. Lett. 34(17), 2664–2666 (2009). [CrossRef] [PubMed]
13. T. M. Cover and J. A. Thomas, Entropy, Relative Entropy, and Mutual Information (John Wiley & Sons, Inc., 2005). pp. 13–55.
14. A. N. Kolmogorov, “On tables of random numbers,” Theor. Comput. Sci. 207(2), 387–395 (1998). [CrossRef]
15. C. F. Rothe, “Dynamics of complex systems,” IEEE Eng. Med. Biol. Mag. 16(4), 10–11 (1997). [PubMed]
16. J. V. Neumann, Mathematische Grundlagen der Quantenmechanik (Dover Publications, 1943). pp. 463–730.
17. M. Li and P. Vitányi, An Introduction to Kolmogorov Complexity and Its Applications (Springer, 2009). pp. 1017–1020.
18. K. Burton, R. Zelikowsky, D. Shandling, E. Lindsley, and D. L. Farkas, “Contrast enhancement in biomedical optical imaging using ultrabright color LEDs,” in Biomedical Optics (BiOS)2007, (International Society for Optics and Photonics, 2007), 64411I–64411I–64418.
19. A. L. Barbieri, G. F. D. Arruda, F. A. Rodrigues, O. M. Bruno, and L. D. F. Costa, “An entropy-based approach to automatic image segmentation of satellite images,” Physica A 390(3), 512–518 (2011). [CrossRef]
20. R. C. Eberhart and Y. Shi, “Particle swarm optimization: developments, applications and resources,” in Evolutionary Computation,2001.Proceedings of the 2001 Congress on, 1988), 81–86 vol. 81. [CrossRef]
21. H.-L. Shen, P.-Q. Cai, S.-J. Shao, and J. H. Xin, “Reflectance reconstruction for multispectral imaging by adaptive Wiener estimation,” Opt. Express 15(23), 15545–15554 (2007). [CrossRef] [PubMed]
22. W. C. Sun, Z. K. Yang, and W. Yang, “Fourier transform spectrum characteristics of the image analysis,” computer and information technology (2005).
23. H. Demirel and G. Anbarjafari, “Satellite image resolution enhancement using complex wavelet transform,” IEEE Geosci. Remote Sens. Lett. 7(1), 123–126 (2010). [CrossRef]
