2.1.

Specimen

Porcine heart tissues were cleaned and fixed on the lab table under LED illumination. Anticoagulant heparin was dropped into fresh arterial blood of adult sheep to prevent condensation. The ethics committee of Zhejiang University approved the study.

2.2.

Apparatus

The light source used here is the LED ceiling system constructed by Zhejiang University, which is an LED panel with a size of $3\mathrm{m}\times 1.5\mathrm{m}$ integrated with 1700 LEDs (11 kinds of LEDs, specifications shown in Table 1). The 11 high power LEDs consisted of eight color LEDs and three white LEDs with different correlated color temperatures (CCTs), whose spectral distributions (SPD) are all given in Fig. 1. The luminance level of each LED can be controlled via the LED control circuit and software was developed to generate the light with the target SPD based on a light matching algorithm and the feedback signal from a spectrometer.

Table 1

Light-emitting diode (LED) specifications.

Fig. 1

Measured spectra of light-emittig diodes (LEDs) used. Intensity is in arbitrary units.

Images of specimens were obtained using a charge-coupled device (CCD) camera. A commercially available 8-bit CCD camera DFK 31BU03 with a pixel resolution of $1024\times 768\text{pixels}$ is used here. It was made by the Imaging Source, Germany. It is interfaced using a Universal Serial Bus (USB2.0). Output RAW image files were analyzed by using self-developed software based on MATLAB. Reflection spectra were acquired using a spectroradiometer. The spectrometer adopted is PR655 SpectraScan produced by Photo Research Inc. For PR655, the measurement range covers 380 to 780 nm with an average bandwidth of 4 nm. The spectral data also can be transported through the USB port. Before the experiments, the camera and the spectrometer were all calibrated with the D65 standard light source.

A multispectral imaging system was used in the acquisition of spectral reflectance of porcine fat tissues. It was self-developed by the Department of Information Science and Electronic Engineering of Zhejiang University.²⁹ As illustrated in Fig. 2, the multispectral imaging system consists 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 their system, they developed an improved reflectance reconstruction method by adaptively selecting training samples for the autocorrelation matrix calculation in the Wiener estimation. 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 a 10-nm interval.

Fig. 2

Sketch of the multispectral imaging system.

2.3.

Spectral Reflectance Comparison Model for Wavelength Selection

Using different colored LEDs in surgical lighting requires a method to select the proper wavelengths. The traditional wavelength selection method was focused on the color difference of tissues in multipictures, which required a lot of time to complete the complicated calculation in the data process. Here, the spectral reflectance comparison (SRC) model was developed to find the wavelengths, which made the biggest difference in reflectance comparison between blood vessels and background tissues. Tissues under these wavelengths will have a strong gray contrast in the display image, which will help the surgeon easily identify them. Only the spectral reflectance of target tissues was necessary for the whole data processing, which took much less time and was easily applied. Therefore, the spectral reflectance is vital to the selection of the wavelengths used to mix white light. We measured spectral reflectance using spectroradiometer PR655 and then performed multispectral analysis to prove that the measurement was correct.

The spectral reflectance of fat measured with a spectroradiometer is shown in Fig. 3(a). For multispectral image analysis, we selected two regions of interest (ROIs) in the specimen, each containing fat tissue [Fig. 3(c)] and obtained their spectral reflectance [Fig. 3(b)]. Comparison of the results of the two methods showed that they matched. However, the spectral resolution of the spectroradiometer chosen was 4 nm while that of multispectral imaging system was 10 nm. Thus, we adopted direct measurement with the spectroradiometer as it had better resolution.

Fig. 3

(a) Spectral reflectance of fat in regions of interest (ROI) [circled in (c)] scanned by a spectroradiometer directly. (b) Spectral reflectance of two sampling points [marked in (c)] obtained by multispectral image analysis. (c) Fresh porcine specimen.

We obtained the reflectance spectra of blood and background tissue separately using the same incident light. Equation (1) demonstrates the comparison of two tissues’ spectral reflectances by the ratio of their reflectivity:

2.4.

Image Evaluation

The standard image evaluation method focuses on gray level contrast. We added entropy analysis to the evaluation process to evaluate the image quality with respect to tissue texture and the details that were shown. Contrast was quantified using our evaluation function in Eq. (2), which combines entropy analysis and brightness contrast:

$A$ is the entropy of the ROIs in the input image. Image entropy is used in the evaluation of the texture of images $P_i$ is the probability of gray value $i$ occurring in the image and changes from 0 to 1, where 1 means that the image is a solid color and entropy is 0. A higher entropy value indicates that the image has more detail, which means the texture is refined. Thus, $A$ expresses the distribution of gray values in the ROIs and the amount of information in them. When the image is a solid color, there is no information and $A$ is 0.

While $A$ characterizes the texture of the tissue, $B$ reflects the gray level contrast of two different tissues in an image. For example, our experiment focused on the contrast between blood vessels and background tissue. An image with these two parts could be acquired by a CCD camera and then changed into a grayscale image. The gray values of the pixels in the blood vessels and background parts of the image are calculated separately. $m$ and $n$ are the number of pixels selected in the images of the blood vessels and the background tissue, respectively. The average gray value for the two tissues is calculated and then the value for blood vessels is divided by that for the background to form $B$. A higher $B$ value means a higher brightness contrast.

The correction factor $k$ is defined according to the purpose of the evaluation. A higher $k$ means that there is more emphasis on image resolution; otherwise, $k$ highlights the gray level contrast of two specific tissues. In our processing analysis, $k$ is defined as

3.1.

Spectral Reflectance Comparison Model Validation

The spectroradiometer scanning region is a circular zone [as in Fig. 3(c)]. The artery vessel was too small to fill the zone, so we obtained reflectance spectra of blood vessels by scanning the fresh ovine blood (into which heparin was dropped) instead. Figure 4 showed the wavelength selection process. Incident lights of artery vessels and background myocardium tissue were the same and the spectral radiance is shown in Fig. 4(a). Reflectance spectra of two tissues were measured as in Figs. 4(a) and 4(c). The SRC results, $C(\lambda )$, shown in Fig. 4(d), imply that the spectral components between 450 and 550 nm contribute the most to the identification of blood vessels that have a meat background.

Fig. 4

(a) Spectral radiance of incident light. (b) Reflectance spectral radiance by PR655 scanning of fresh ovine blood. (c) Reflectance spectral radiance of ovine myocardium tissue measured directly. (d) Comparison of spectral reflectance, $C(\lambda )$, of blood and background myocardium tissue.

To verify the accuracy and stability of the measured results, an ovine heart was set and imaged (shown in Fig. 5) under eight monochromatic LEDs (data given in Table 1). The gray level contrast ($G$) values of blood and background myocardium tissue in each image were calculated and are shown in Fig. 5. Meanwhile, the gray level contrast of the artery vessel and background myocardium changed with the LED wavelengths and could be extracted by SRC (Fig. 4); the extracted result is plotted in Fig. 5 (bottom right panel). We found that the trends of the two results were similar.

Fig. 5

Ovine heart imaged using eight monochromatic LEDs with wavelengths from 447 to 657 nm. (Lower right) Graph of contrast value obtained by spectral comparison (black curve) and the analysis of the eight images (red curve).

3.2.

Contrast Enhancement

Based on the preliminary reflectance comparison result in Fig. 4(d), light in the 450 to 550 nm range was found to be helpful in identifying blood vessels. Therefore, we chose LEDs with wavelengths of 479, 506, and 522 nm between our 11 kinds of LEDs to mix the light. In addition, we added a 593-nm wavelength light in the selection group to mix with white light. Then warm (3000 K), neutral (4100 K), or cool (6000 K) white light was mixed with the selected LEDs. Specimens (cleaned ovine heart tissues) were imaged under the three compound lights and the three commercial white LEDs separately (two examples are shown in Fig. 6). In Fig. 6(a), region A, which contained both heart artery vessel and background myocardium tissue, was used for brightness contrast analysis using the evaluation function. Region B, which contained mainly artery vessels and background myocardium tissue, was used to analyze the gray level contrast [$B$ in $F$, Eq. (2)]. The result of the analysis is shown in Fig. 6(b). A higher $F$ value means a higher illumination quality. The gray level contrast values of the specimen under mixed white light and different CCTs were higher than those under commercial white LEDs.

Fig. 6

(a) Ovine heart imaged under white light, where region A was used for entropy analysis and region B for gray contrast analysis. (b) Evaluation function value $F$ (normalized) for six LED illuminations; mixed white light (black bars), and commercial white LEDs (red bars) with different correlated color temperatures.

Figures 7(a)–7(c) are images of the ovine heart taken under mixed white light and Figs. 7(d)–7(f) are images taken under commercial warm, neutral, and cool white LEDs. The contrast of blood vessels and background tissues is low in Fig. 7(f), while details were lost in Fig. 7(e). The images in Figs. 7(a)–7(c) have higher contrast and sharpness.

Fig. 7

Images of ovine heart taken in (a) cool (5500 K), (b) neutral (4000 K), and (c) warm (3000 K) mixed white light and in (d) cool, (e) neutral, and (f) warm broadband LED. All images were captured under the same white balance of stubborn 5500 K.

3.3.

Validation Using the New Approach

The traditional image evaluation method focused mainly on gray value contrast, which ignores the value of the details of the tissue in the image. Our evaluation function [Eq. (2)], which includes entropy analysis, can solve this problem because it balances the two parts. Figure 8 shows gray images of the same tissue under two different lighting conditions. The gray value contrast of artery vessels and background myocardium tissue in the two images was the same ($\text{gray contrast}=23$); however, the image captured under the mixed white light had a higher $F$ value (83.27) than the image taken under the commercial white LED (77.04).

Fig. 8

Gray image of ovine heart captured in (a) mixed cool white light (5500 K), with the white balance set at 5650 K and (b) commercial cool white light (5500 K), with the white balance set as a standard bulb.

The two images in Fig. 8 would show the same quality performance with the traditional image evaluation method because only the gray level contrast is considered. However, there is more detail and texture information in the red box region of the ovine heart in Fig. 8(a) than in that of Fig. 8(b). The mixed white light with an optimal spectrum achieved a better performance as diagnostic lighting.