Endoscope is a general term for medical imaging devices used to examine internal organs via orifices of the human body to conduct diagnostic, therapeutic, and observational tasks [1]. Different types of endoscopes have been designed to gain access to the ear (otoscope), nose (rhinoscope), throat (nasopharyngoscope), larynx (laryngoscope), tracheobronchial tree (bronchoscope), gastrointestinal tract including esophagus, stomach and small bowel (gastroscope), duodenum (duedenoscope), bile duct (cholesdoscope), urinary track (cystroscope, ureteroscope), cervix/uterurs (hysteroscope), colon/rectum (colonoscope), etc.

A video endoscope delivers the light and imaging components into the human body near the region of interest (RoI) for image acquisition and then transmits the image data from RoI for image display. A rigid endoscope transmits the light and image through optical paths such as fiber optics. A flexible endoscope uses a flexible tube to manually control the scope’s position, shape, and angle while conveying the image data through electrical conduits. Both rigid and flexible endoscopes allow the endoscopist to control the device to view the image in real time, so the RoI can be revisited from different angles to get a better view, if needed. In contrast, a capsule endoscope is passively propelled by the gastrointestinal tract and stores the images wirelessly in an external recorder. The stored images will be processed and viewed offline after the device is excreted to conclude the session [2–4]. The endoscopist cannot control the view unless the device can be remotely controlled [5] as in the case of a magnetically maneuvered capsule endoscope. Unlike most flexible endoscopes, capsule endoscopes do not have the water/air channel for rinsing or the instrument channel for biopsy. Due to the limited form factor and single-use nature, capsule endoscopes usually do not have the capability to deliver imaging quality equivalent to flexible endoscopes.

The visualization capability of an endoscope is the foundation for safely conducting diagnostic and therapeutic tasks, which may pose risks to the patient depending on the intended use. A video endoscope can be considered as a complete color imaging system that is usually divided into three cascaded components as shown in Figure 1: the scope that includes optical components (e.g., imaging sensor, lens system, and light source/guide) for image acquisition, the video processor for processing the image data, and the display for reproducing the image for the human user. The interface between the scope and the video processor is usually a well-defined proprietary interface such that different scopes can operate with the same video processor. Single-use scopes provide a convenience option that eliminates the burden of sterilization or reprocessing. However, single-use scopes’ disposable imaging components do not deliver the same image quality as re-usable flexible scopes. Depending on the design of a flexible endoscope, the light source can be either built in the distal end of the scope, or housed inside the video processor. The display can be built in the video processor as a closed system, or it can be an external medical-grade monitor in an open system. Some endoscope systems use a desktop or tablet computer to implement the hardware and software functions of a video processor.

Most endoscopes use white light to provide the basic white light imaging (WLI) mode. Some endoscopes use specific light spectra to enhance color contrast for certain features such as blood vessels, called optical chromoendoscopy. The narrow band imaging (NBI) mode uses narrow-band color filters and a broad-band white light source to generate the specific spectra. When individually controlled red, green, and blue light-emitting diodes (LED) are used to mix white light in modern endoscopes, new illumination modes can be easily populated [6, 7]. Other endoscopes achieve similar effects by digitally processing white light-illuminated images, called digital chromoendoscopy. Such digital chromoendoscopy modes can be implemented in software without additional hardware and multiplied quickly by changing the mix ratio between different color channels.

Color performance is an important safety and performance attribute of endoscopes. An endoscope that exhibits poor color performance inadequately reproduces color images and may make it difficult for the endoscopist to visualize features needed for diagnostic or therapeutic tasks. Unfortunately, there is not a standard, well-adopted color performance testing method dedicated to endoscopes yet. Endoscopes are frequently tested with methods that measure optical characteristics with monochrome test patterns. The drawback is that such grayscale-based testing results do not represent the actual color performance. For example, a black/white test pattern is suitable for measuring the grayscale contrast of an endoscope. However, in colonoscopy, the image is rendered not in gray shades but in pink shades. A colonoscope with higher grayscale contrast may have lower contrast between pink shades. Using grayscale contrast to estimate the color performance may be misleading.

Since the first commercial video endoscope marketed in 1983 [8, 9], color performance has been an important characteristic to be tested [10–19]. The first colorimetry-based, quantitative color performance testing was reported in 1987 [12]. To assess accurate color reproduction of mucosal and biological fluids, Knyrim et al. used six color patches to test four pioneering video endoscopes. The device output was measured from the display with a colorimeter to obtain the CIEXYZ values. The measurement results were plotted on the CIE chromaticity diagram to compare their color gamut areas against the ground truth. Test results showed that all endoscopes exhibited color shift and reduced color gamut in addition to non-uniformity issues. Although the devices were quantitatively measured, the inter-device comparison was qualitative and limited to chromaticity (i.e., lightness excluded) evaluation only.

Three decades later, the same testing framework is still being used in recent endoscope studies, and more colorimetry models and tools have been developed and utilized to evaluate endoscopes. For example, the perceptually uniform CIELAB color space was used to predict the perceived color [7]. The CIE 1976 color difference model (ΔE₇₆) was used to evaluate color difference between color shades [6]. The improved CIEDE2000 color difference model (ΔE₀₀) was used to predicate the perceived color difference [18, 19]. Fig. 1 depicts such a testing framework, which is described as follows.

All of the aforementioned testing methods evaluate the absolute color difference between the ground truth and the device output. For example, absolute color errors measured by CIEDE2000 are used to define “color fidelity” in [18]. Evaluating absolute color difference is suitable for color imaging devices that are intended to accurately reproduce the original color. However, in practice, most endoscopy devices cannot and do not intend to faithfully reproduce the original color. As a medical device, an endoscope is intended to be used to visualize biological features. Many endoscopes are not color-calibrated accurately, and some employ intensive image and color processing algorithms. As a result, the absolute color errors may not represent the actual color performance according to its intended use. When interpreting the test results, small absolute color errors can indicate a well-calibrated endoscopy device. But large absolute color errors do not always indicate an inadequate endoscope. It is not uncommon for clinical endoscopes to generate absolute color errors greater than 20 ΔE₀₀. Thus, comparing two endoscopes that both have excessive absolute color errors will not help determine whether they have equivalent color performance.

In this work, we describe alternative methods for evaluating endoscopes that do not intend to reproduce the original color.

Although an endoscopy device may not intend to reproduce the original color faithfully, the color contrast present in the ground truth should be preserved in the image reproduced by the device. In other words, two visually discernible color shades in the original scene should still be discernible in the reproduced image. The CIE color difference formulas are used to measure how well two color shades can be visually discerned.

Let truth_i and device_i be the CIELAB coordinates of patch i measured from the color target and device output, respectively.

Each truth_i and device_i can be represented as a datapoint (L^∗, a^∗, b^∗) in the three-dimensional CIELAB color space. ΔE is a function calculating the color difference between two color shades based on either the ΔE₀₀, ΔE₉₄, or ΔE₇₆ formulas. The color contrast enhancement (CCE) between two color patches i and j is defined as:(1)CCE =ΔE(devicei,devicej)ΔE(truthi,truthj), where i and j are different numbers iterating all pairs of different color patches on the color target. Figure 2 illustrates the difference between color fidelity and color contrast.

The CCE measures the degree to which the color difference between two patches will be changed by the device. When CCE > 1, the color difference is increased by the device. When CCE < 1, the color difference is reduced by the device. When CCE = 1, the color difference remains the same regardless of whether or not the color transformation is faithful. High-quality endoscopy systems are expected to have CCE > 1 since the color contrast will be preserved or more visible. The CCE can also be used to compare two endoscopy systems when both have CCE < 1.

When the color differences of all n(n−1)2 pairs of n patches are graphed on a scatter plot as shown in Figure 3, the identity line x = y can help judge whether a patch-pair has CCE > 1, CCE < 1, or CCE = 1. The identity line can also help calculate the percentage of patch-pairs that have CCE > 1.

Although an endoscopy device may have specific rendering intent to enhance certain color features, the color order in a perceptual color space should still be preserved. For example, considering the lightness attribute of the patches shown in Figure 4, a lighter patch (#21) should always look brighter than a darker patch (#22) regardless of the devices’ rendering intent. Similarly, the order of the hue attribute should be preserved. For example, an orange patch (#7) should always look more reddish than an orange-yellow patch (#12). Finally, the order of the chroma attribute should be preserved. For example, a red patch (#15) should always look more colorful than a moderate red patch (#9).

In this study, the CIELCH color space is used to analyze the patch order in lightness, chroma, and hue. CIELCH is the cylindrical polar coordinate representation of CIELAB, which is the Cartesian coordinate representation of the same color space. The patches in the CIELCH color space should preserve the patch order in lightness, chroma, and hue defined by the ground truth.

Figure 5 shows the patch order in lightness, chroma, and hue in two-dimensional charts. The linear regression result is included in each chart. Ideally, all datapoints should be on a straight line to exhibit perfect linearity. These charts can be used to identify any out-of-order color shades.

If perfect linearity cannot be achieved, monotonicity should be preserved. Figure 6 shows the one-dimensional views of the patch order in lightness, chroma, and hue. The Kendall Tau-a rank correlation coefficients are calculated to quantitatively represent the concordance of the patch order. These charts can be used to identify any out-of-order color shades that may not be visually discerned clearly.

As shown in Figure 7, the CPR tool provides a user interface of three sections for the user to interact.

In the Input section, the user enters the following data:

In the Convert section, the CPR tool utilizes four quadrants to facilitate the user’s verification of the measurement data. The left and right quadrants present the input data for the reference and device output, respectively. The top quadrants display the original input data, while the bottom quadrants exhibit the converted data in CIELAB. Each quadrant includes a table presenting the measurement data and a chart reconstructing the test target. To ensure the integrity of the input data, users can perform numerical verification using the tables and visual verification using the charts as a sanity check.

In the Output section, the user enters a file name for the CPR tool to generate the analysis report. The user can optionally enter four labels to identify the device under test.

Figure 8 shows a sample test report generated by the CPR tool. The report contains three columns of plots.