Open Access
Add to BibSonomy
Abstract
As endoscopic imaging technology advances, there is a growing clinical demand for enhanced imaging capabilities. Although conventional white light imaging (WLI) endoscopy offers realistic images, it often cannot reveal detailed characteristics of the mucosa. On the other hand, optical staining endoscopy, such as Compound Band Imaging (CBI), can discern subtle structures, serving to some extent as an optical biopsy. However, its image brightness is low, and the colors can be abrupt. These two techniques, commonly used in clinical settings, have complementary advantages. Nonetheless, they require different lighting conditions, which makes it challenging to combine their imaging strengths on living tissues. In this study, we introduce a novel endoscopic imaging technique that effectively combines the advantages of both WLI and CBI. Doctors don’t need to manually switch between these two observation modes, as they can obtain the image information of both modes in one image. We calibrated an appropriate proportion for simultaneous illumination with the light required for WLI and CBI. We designed a new illumination spectrum tailored for gastrointestinal examination, achieving their fusion at the optical level. Using a new algorithm that focuses on enhancing specific hemoglobin tissue features, we restored narrow-band image characteristics lost due to the introduction of white light. Our hardware and software innovations not only boost the illumination brightness of the endoscope but also ensure the narrow-band feature details of the image. To evaluate the reliability and safety of the new endoscopic system, we conducted a series of tests in line with relevant international standards and validated the design parameters. For clinical trials, we collected a total of 256 sets of images, each set comprising images of the same lesion location captured using WLI, CBI, and our proposed method. We recruited four experienced clinicians to conduct subjective evaluations of the collected images. The results affirmed the significant advantages of our method. We believe that the novel endoscopic system we introduced has vast potential for clinical application in the future.
© 2024 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
Cancer, a global public health challenge, has consistently driven the medical community to advance treatments, offering patients more options and hope [1]. Specifically, Endoscopic Submucosal Dissection (ESD) often results in nearly 100% five-year survival rates for early-stage cancer patients [2], underscoring the pivotal role of endoscopy in early cancer diagnosis. Currently, both researchers and clinicians aspire to harness endoscopic technology for more precise detection of early mucosal tissue changes to aid clinical treatments [3]. Routine endoscopic examinations typically employ White Light Imaging (WLI) illumination systems for tissue observation [4]. However, this method faces challenges, as early lesions, often minuscule and with indistinct color and shape, may be difficult to detect, potentially impacting the accuracy of clinical examinations [5]. To address this challenge, endoscopic staining technology was developed. Introduced in the 1970s, chromoendoscopy enhanced the precision with which doctors could determine the dimensions and boundaries of early-stage cancers [6]. Early-stage cancer detection is critical yet challenging, as limitations with WLI and image-enhanced endoscopy (IEE) significantly impact detection rates and outcomes [7]. Studies indicate over 70% of cancer mortalities occur at sites lacking early detection methods, with early-stage detection's low sensitivity and high false positives leading to unnecessary morbidity or mortality [8]. For gastric cancer, misdiagnosis is common as many early cases mimic gastritis under WLI. Reports demonstrate IEE techniques’ effectiveness in early gastric cancer detection [9]. One study noted non-magnified NBI images were too dim for lesion detection [10], while another cited magnified NBI's diagnostic challenges for early gastric cancer [11].
Endoscopic virtual staining technology, renowned for its simplicity and real-time capabilities, is highly regarded by clinicians. Broadly, this technology branches into two categories: optical staining via light source modification, exemplified by Narrow Band Imaging (NBI, Olympus, Tokyo, Japan) [12], Compound Band Imaging (CBI; Aohua, Shanghai, China) [13], Red Dichromatic Imaging (RDI, Olympus) [14], Linked Color Imaging (LCI, Fuji, Tokyo, Japan) [15], Blue Light Imaging (BLI, Fuji) [16], and i-SCAN Optical Enhancement (i-SCAN OE, Pentax, Tokyo, Japan) [17]; and algorithm-driven coloring methods like i-SCAN (Pentax) [18], Flexible Spectral Imaging Color Enhancement (FICE, Fuji) [19], Hemoglobin Enhancement (HbE, Olympus) [20], and Storz Professional Imaging Enhancement System (SPIES, Karl Storz, Tuttlingen, Germany) [21], etc. The former is prevalent in clinical settings, contributing significantly to optical biopsies [22]. Light's interaction with biological tissue varies based on its wavelength, with different tissue components absorbing distinct wavelengths [23]. As light traverses these tissues, scattering occurs, with longer wavelengths exhibiting more extensive and deeper diffusion. Notably, the gastrointestinal tract's coloration largely hinges on hemoglobin in blood, predominantly absorbing blue and green light [24]. Leveraging this knowledge, optical-based staining technologies have carried out special treatment on the light source design. For instance, CBI employs blue-green narrow-band composite light, aligning with hemoglobin's absorption peaks. By narrowing the spectrum, the vascular contrast reduction due to scattering is effectively mitigated. This enhancement, amplifying vascular contrast in mucosal tissue, proves invaluable for early lesion detection since such anomalies typically initiate from unusual mucosal surface proliferation.
Although CBI enhances vascular structure contrast, it struggles with depth differentiation. Its dual narrow bands can't represent features of other wavelengths, leading to a somewhat monotone image due to false color processing. In contrast, WLI's spectral range encompasses nearly the entire visible spectrum, resulting in more authentic colors. Furthermore, while WLI offers greater penetration for a more comprehensive view of tissue structures, it simultaneously faces challenges in terms of contrast resolution [25]. In clinical settings, the complementary nature of WLI mode and CBI mode is evident. Combining CBI's microvascular clarity with WLI's comprehensive tissue imaging could pave the way for a superior imaging mode, crucial for disease detection. Our prior research optimized the endoscope's light source, merging WLI and CBI images to achieve a higher contrast image [13]. Yet, in a clinical context, quick toggling between WLI and CBI light sources escalates hardware and computational expenses. Therefore, in the design of our new endoscopic imaging technology, we innovatively integrate these two technologies at the optical level. Concurrently, we prioritize enhancing image color authenticity and delving into a balance between vascular contrast augmentation and comprehensive tissue information capture. We posit that emerging technologies should aspire to meet these benchmarks, bolstering the efficacy of early disease detection.
Recent advancements in endoscopic imaging technology have enhanced the precision of gastrointestinal examinations [25]. This study ventures to refine this technique by amalgamating two prevalent endoscopy modes optically. We've innovatively devised an illumination spectrum tailored for the digestive tract's unique attributes. Crucially, by integrating advanced hemoglobin feature enhancement algorithms, we've harmonized these modes, aiming to offer clinicians more lucid and comprehensive image data. After ensuring compliance with international safety standards, we conducted clinical trials, amassing images from 256 lesion sites using WLI, CBI, and our novel method. Four seasoned clinicians’ subjective evaluations underscore the distinct benefits of our approach, highlighting its promising prospects for future clinical applications.
2. Materials and methods
Theoretically, merging images from WLI and CBI typically enhances image contrast. However, acquiring these two types of images at the pixel level in a manner that corresponds accurately poses significant challenges due to the inherent motion of living tissues. While obtaining WLI and CBI images with a color sensor in itself does not create motion artifacts, capturing these images sequentially and then fusing them introduces new motion artifacts and also increases the computational load. To address this issue, we first merged the required light for WLI and CBI at the optical level, and then performed simultaneous imaging using a color image sensor. This approach effectively avoids the problem of motion artifacts. Subsequently, advanced image processing algorithms were employed to produce a single image that integrates the advantages of both methods. Traditionally, endoscopes use xenon light sources, and narrow-band illumination techniques like CBI are achieved by filtering white light, inherently limiting the brightness of the narrow band. The advancement in LED technology has overcome this obstacle, and thus we have employed LED and similar light sources for illumination.
2.1 Flexible endoscopic system and its optical features
As depicted in Fig. 1, the flexible endoscope imaging system mainly includes an image processor, a light source, the endoscope body, and a display. The light source emits high-brightness illumination, which is conveyed through the light-guiding fiber in the endoscope and shone onto human tissue via the illuminating window at the distal end. The tissue's absorption and scatter of this light are captured by the image sensor, converted into an electrical signal, and relayed to the image processor. This processor turns the signal into a real-time video stream displayed for clinicians.
To effectively enhance lesion contrast, a prevalent approach focuses on increasing blood vessel visibility, where spectral selection is crucial. As depicted in Fig. 2, based on the absorption curves of deoxygenated and oxygenated hemoglobin [26], hemoglobin strongly absorbs light in the 400-440 nm and 520-590 nm ranges, influencing light penetration depth in the mucous membrane. Factoring in prevalent endoscopic narrow-band illumination techniques and high-brightness LED choices, our research incorporates three high-performance LED modules into the newly designed light source equipment. These modules emit high-rendering white light (with a color rendering index above 90), 415 nm blue-violet light, and 530 nm green light.
Fig. 2. Absorption curves of deoxygenated hemoglobin and oxygenated hemoglobin & the depth of penetration into the mucous membrane of the light
Different wavelengths of light have varying absorption and scattering coefficients in human tissue. The relationship between wavelength and gastric mucosa depth can be determined through experimental simulation and calculation [27]. As shown in Fig. 3, penetration depth into the mucous membrane increases with increasing wavelength. The 415 nm wavelength captures shallow blood vessel details, the 530 nm targets deeper vascular structures, and white LED provides a broad spectrum of tissue data. For vascular observation, clarity and contrast are essential. The 415 nm light visualizes blood vessels up to a depth of 44um [28], whereas the 530-540 nm range reaches depths close to 500um [29].
2.2 Design and indicators of the light source
In clinical settings, endoscopes require high-intensity illumination. Therefore, when designing the light source device, it is essential to choose LEDs with greater power and higher luminous efficiency per unit area. In our study's light source device, the white LED (CFT-90-WDH-X11-SA557, Luminus) emits 11.5W of radiant power with a radiant efficiency of 1.28W/mm2. The blue LED (CBT-90-UV-L11-P410-22, Luminus), on the other hand, emits 17W of radiant power with a radiant efficiency of 1.88W/mm2, while the green LED (CBT-90-G-L11-CM101-R2, Luminus) produces 5W of radiant power with a radiant efficiency of 0.56W/mm2. Figure 4 illustrates the light path design of our light source.
In optical path design, it's crucial to consider optical transmission efficiency, which places a strong emphasis on designing the etendue for each stage. Etendue reflects the geometric properties of a light beam in an optical system. A larger Etendue complicates the optical design and necessitates a bigger structural size. Optimal design requires controlling the Etendue throughout the system, ensuring it is always greater than that of the final output surface to meet the system's efficiency requirements. In a two-dimensional plane, etendue is expressed as: $\textrm{etendue} = \,\mathrm{n\ast a\ast sin}\theta $, where n is the refractive index, a is the spot diameter, and $\theta $ is the divergence angle of the spot [30]. Following these principles, optical path transmission efficiency relies on etendue design, which, in turn, is determined by the spot size and divergence angle of the three main optical paths [31,32].
In the collimating optical path, when designing the collimating lens group, it's essential to match the etendue of the LED light to that of the collimated light. In the combined light path, the etendue remains unaltered. In the coupling light path, while designing the coupling lens group, it's crucial to align the etendue of the collimated light with that of the converged light, which will be coupled into the light guide module. Since the light guide module limits the etendue of this system, ensuring maximum transmission efficiency requires each stage's optical etendue in the collimation and coupling optical paths to exceed that of the light guide module. Table 1 illustrates the etendue design for this system. Furthermore, when designing the etendue for the collimated beam, consider that the collimated beam will pass through the combined optical path, including a dichroic beam splitter. When incident on the dichroic beam splitter within a divergence angle of ±5°, the light splitting efficiency and spectral ratio remain consistent. However, an increased incident angle can cause deviations from the designed spectral efficiency and spectral ratio.
Table 1. Optical expansion design at each stage of the light path of the light source
When designing the combined light path, spectral distribution is a critical consideration. The white LED has an effective spectral range of 440-700 nm, the blue-violet LED covers 400-430 nm effectively, and the green LED spans 515-545 nm. To address this, the first dichroic beamsplitter is configured as a long-pass beamsplitter. When normal incidence occurs, it achieves over 95% transmittance within the 440-700 nm wavelength range, and for side incidence, it maintains over 95% transmittance within the 400-430 nm range. The second dichroic beamsplitter is designed as a bandstop beamsplitter. Under normal incidence, it attains over 95% transmittance within the 400-500 nm and 550-700 nm wavelength ranges. When subjected to side incidence, it maintains over 95% transmittance within the 515-545 nm range.
In the light source design, we need to accommodate both white light and narrow-band illumination for contrast observation. This necessitates the inclusion of white light mode, narrow-band mode, and mixed mode. Each LED in the light path can be adjusted in brightness through current regulation. In white light mode, the power distribution is as follows: 1W for the blue-violet LED, 11.5W for the white LED, and 2W for the green LED. For narrow-band mode, the power distribution is: 17W for the blue-violet LED and 3.5W for the green LED. In mixed mode, the power distribution remains the same as in the narrow-band mode for both the blue-violet LED and the green LED, with 5W for the white LED.
The light source device directs illumination light into the endoscope's optical fiber for uniform irradiation of the human body's natural cavity mucous membrane. The endoscopic lens houses an imaging module comprising four lenses and an image sensor circuit. With a wide optical design field of view of 145 degrees, it captures reflected light from human tissue comprehensively. The embedded color image sensor boasts a resolution of 1920${\times} $1080 and operates at 60fps. As depicted in Fig. 5, this sensor differentiates light signals across three color bands: red, green, and blue. Light reflecting from different depths enters three separate channels.
Incorporating white light into the 415 nm and 530 nm narrow bands significantly enhances the image's signal-to-noise ratio. The Bayer filter's transmittance on the image sensor surface, the image sensor's response characteristics, and its inherent noise all play crucial roles in determining this ratio. The image sensor's response characteristics primarily rely on factors such as semiconductor quantum efficiency, unit charge, wavelength, Planck constant, and the speed of light in a vacuum [33]. During the design phase, we strive to balance the values of the three channels: R, G, and B. This balance ensures that the later-stage color values remain relatively consistent. Based on practical measurements with a white background, our observed response ratio for the three channels, R, G, and B, is approximately 0.5:1:0.708.
2.3 Software algorithms for feature augmentation
In light of the light sources discussed in Section 2.1, we've opted for a compromise design. Broadband white light tends to suffer from scattering, leading to reduced image contrast at the two peak wavelengths. To address this, we utilize software calculations to restore image information for these critical wavelengths while endeavoring to preserve the white light information as much as possible. This approach aims to specifically compensate for the lost contrast.
Hemoglobin, a widely occurring pigment in gastrointestinal mucosa, can be assessed using traditional white-light endoscopy through image data [34]. The commonly employed metric is the Index of Hemoglobin (IHb), relying on the R and G channel ratios in WLI images [35]. In conventional WLI mode, the blue light component is typically weaker than the green light. Consequently, the B channel signals in the image, influenced by scattering factors, exhibit limited discrimination of superficial fine mucosal vessels. Meanwhile, the G channel signals effectively display the thicker blood vessels in the middle and lower layers of the mucosa. Thus, due to light source limitations, texture features like superficial blood vessels in WLI images may appear relatively weak. As a result, IHb measurement primarily hinges on the R-channel and G-channel ratios in WLI images, without considering the B channel.
Building upon the white light component, our light source intensifies the blue and green narrow bands within specific wavelengths. These two narrow-band peaks are readily absorbed by hemoglobin in blood, resulting in significantly improved vessel contrast. We carefully adjusted the spectral ratios to ensure that the B channel and G channel in the mucosa-illuminated images exhibit similar gray contrast for both superficial and deeper blood vessels. Consequently, we obtain two more precise hemoglobin indicators: the superficial mucosal hemoglobin index (IHb1) and the middle and lower mucosal hemoglobin index (IHb2).
Given these characteristics, we need to enhance the contrast of blood-rich elements in the image. This can be seen as remapping the two hemoglobin indicators and enhancing subtle color changes in the endoscopic mucosal image. This approach enhances the contrast of both superficial and deeper blood vessels. Based on this, we've developed the following image-processing algorithm:
Firstly, before computing the two hemoglobin indexes, we preprocess the original image from the image sensor to adjust the image's color balance using the following formula:
Next, we calculate the two hemoglobin indexes IHb1 and IHb2, using the following formulas:
Then, we perform linear stretching on these two hemoglobin indicators using the following formulas:
Here, $\overline {IHb1} $ and $\overline {IHb2} $ represent the averages of IHb1 and IHb2, respectively, and ${k_1}$ and ${k_2}$ are two stretching coefficients.
Based on formulas (4) and (5), for the same pixel, if the IHb value is greater than its average value, it will be transformed into a higher hemoglobin concentration index. If the hemoglobin index IHb value is less than its average value, it will be transformed into a lower hemoglobin concentration index. In this way, slight tonal changes that are easily overlooked in normal mucosa can be visualized.
Finally, we perform the ultimate image enhancement based on the two stretched hemoglobin indicators, using the following formulas:
Figure 6 illustrates a comparison of images before and after applying software algorithms for feature augmentation. The first row displays the raw images captured by the image sensor, while the second row presents the images post-feature enhancement processing. It is evident that our method substantially enhances the image details while preserving the natural color of the images.
Fig. 6. Comparison of raw (first row) and enhanced (second row) images using feature augmentation algorithm.
By implementing the image post-processing strategies outlined in this section, our system achieves real-time performance. This is facilitated by simultaneously capturing an image illuminated with both WLI and CBI modes, thereby maintaining the endoscope video stream's frame rate at 1080p, 60 fps. This approach showcases the system's capability to deliver high-quality imaging efficiently in real-time clinical settings.
2.4 Clinical trials and clinical data evaluation
In medical imaging, subjective evaluations by doctors, based on their expertise and nuanced interpretation of small details, colors, and contrasts, are often more impactful than objective quality indices in clinical settings, recognizing that such human factors detect intricacies beyond the reach of objective measurements [36]. Acknowledging the significance of these subjective assessments is essential to affirm the clinical utility of medical images [37].
We enhanced the endoscopic system with a technical modification allowing for one-button switching between lighting modes to account for the continuous movement within the digestive tract, eliminating the impracticality of manual adjustments. This innovation permitted doctors to effortlessly transition between WLI, CBI, and our method, capturing images swiftly and precisely from the same lesion point, which is crucial for subsequent comparative analysis.
To evaluate our innovative light source and imaging algorithm, we conducted a robust clinical trial at Minhang Hospital, affiliated with Fudan University. Utilizing the Aohua AQ-200 endoscope system and the FHD-GT200J electronic upper gastrointestinal endoscope, we gathered comprehensive endoscopic image data alongside pathological diagnoses from patients undergoing gastroscopy. Our trial included images obtained using standard WLI, CBI, and our proposed method, ensuring the study adhered to ethical guidelines with all participants providing informed consent.
The reliability of our comparative experiments was ensured by the systematic capture of images at identical lesion locations across the three modes. We further corroborated these images with biopsy operations to provide a precise pathological reference, establishing a robust benchmark for effectiveness assessments and experimental comparisons.
Our study's integrity was safeguarded by stringent inclusion and exclusion criteria. We included only those patients aged 18 and above who had completed a reliable gastroscopy and from whom we could successfully collect comparative images under the three modes at the same digestive tract location. We excluded any cases where image data did not correlate with pathological outcomes or where diagnoses were ambiguous, minimizing potential biases.
For an exhaustive and accurate evaluation of the image quality produced by each mode, we engaged four seasoned endoscopy specialists. To diminish subjective biases, we incorporated sufficient intervals between evaluations, facilitated by bespoke software that provided an intuitive interface for assessments as depicted in Fig. 7.
We conducted our assessments across five criteria—color contrast, signal-to-noise ratio, brightness, detail contrast, and an overall assessment—with each criterion rated on a scale of 1 to 5. The doctors also provided subjective diagnostic opinions on each image, specifically noting inflammatory lesions, intestinal atrophy, and hyperplasia. To maintain consistency and uniformity in assessments, detailed evaluation templates and guidelines were provided to the doctors. After the evaluations were collected, we employed statistical methods to analyze the data, quantitatively demonstrating the differences in image quality among the various modes, a process detailed in Fig. 8.
3. Experiments and results
3.1 Laboratory testing
We have conducted thorough testing of the parameters and safety of the light source equipment to affirm its clinical usability and safety standards. The setup of our testing environment is detailed in Fig. 9. For this purpose, we utilized a fast spectrum analysis system equipped with a 0.3 m integrating sphere (SPEC-2000A, MESUREFINE Corporation) and a high-precision fast spectroradiometer with a 1 m integrating sphere (HAAS-2000, EVERFINER Corporation). These instruments were employed to measure and verify the optical properties of our system comprehensively.
Fig. 9. The test environment utilized for mixed-mode spectroscopy, featuring the fast spectrum analysis system with a 0.3 m integrating sphere as the primary measurement device.
Regarding usability, key indicators to consider in white light mode include the color rendering index and color temperature. In mixed spectrum mode, it's essential to test the peak wavelength, full width at half maximum, and assess radiation performance as a measure of lighting capability. Table 2 displays the actual test results. White light mode exhibits nearly identical lighting capability compared to mixed spectrum mode, demonstrating ample brightness for clinical use. Additionally, illumination light must meet three specific requirements. Firstly, it must adhere to infrared cut-off criteria, with the radiant flux to luminous flux ratio within the 300 nm to 1700nm wavelength range not exceeding 6 mW/lm. The measured value stands at 3.79 mW/lm. Secondly, the illumination light needs to satisfy the uniformity of illumination, with the actual measured value for uniformity in the reference window being 0.15. Lastly, the illumination light must satisfy the illuminance exceeding the limit point criteria, with the measured value for the reference window's illuminance exceeding the limit point being 0. In our proposed system, spatial resolution characteristics were also tested. The results showed spatial resolutions of 10.1 line pairs per millimeter (lp/mm) at a distance of 3 mm, 20.16 lp/mm at 10 mm, and 1.12 lp/mm at 100 mm.
Table 2. Important optical performance indexes and test results
In terms of safety, our developed light source adheres to international standards, including IEC 60601-1:1988 for general safety requirements of medical electrical equipment and IEC 60601-2-18:2009 for specific safety and essential performance requirements for endoscopic equipment, encompassing both electrical safety and electromagnetic compatibility. Regarding photobiological safety, the selected illumination light falls within the visible spectrum range, aligning with conventional endoscope illumination spectra and complying with EN/IEC 62471:2008 for photobiological safety of lamps and lamp systems. In practical use, our light source is integrated with the endoscope system, featuring a dimming function. When the endoscope approaches the mucous membrane, it automatically senses increased screen brightness and actively reduces lighting intensity, thereby preventing mucous membrane burns. The measured head-end temperature, even at maximum brightness, remains at 40 °C, which is within the compliance threshold of IEC 60601-1:1988, specifying that the insertion section's temperature, except for the light exit part, should not exceed 41°C.
3.2 Subjective evaluation of clinical images and results
We included images from 256 sets of gastric lesions in the study, following stringent criteria. In Fig. 10, four sets of comparative images are presented, each from a different set of these lesions. Within the figure, each row corresponds to images of the same lesion, and each column represents a different imaging mode. Progressing from left to right, the columns display images captured using WLI, CBI, and our proposed method, respectively. Additionally, we have calculated the Peak Signal-to-Noise Ratio (PSNR) for each image.
Fig. 10. Images depicting stomach lesion locations in four individuals. Each row corresponds to the same lesion location for each individual, while each column represents the same imaging mode: (a) WLI, (b) CBI, and (c) Our proposed method. In these images, we can observe distinct characteristics: WLI offers natural colors and adequate brightness but lacks contrast, which hinders the discernment of detailed features. CBI images, while capturing certain details, tend to have a uniform color palette that lacks the vibrancy of natural hues. Additionally, due to low light source brightness, the image's PSNR is poor. In contrast, our proposed method maintains adequate lighting intensity and preserves natural color tones. Crucially, through suitable software algorithms, we've emphasized two vital digestive tract features: blood vessels and textures—essential for routine lesion examinations.
To maintain the objectivity and reliability of the image quality evaluation, we consolidated the assessments of four doctors. Metrics such as color contrast, signal-to-noise ratio, brightness, detail contrast, and overall evaluation were represented by their median to minimize biases from individual evaluations. For diagnostic outcomes, the mode was selected to diminish subjective disparities. Due to the low instances of abnormal hyperplasia across the three techniques, and as each index leans towards binary variables, both abnormal hyperplasia and intestinal metaplasia were categorized together.
Descriptive statistics were employed for metrics: the median (interquartile range) and frequency (proportion) represented the distributions of criteria like color contrast and overall evaluation. The Friedman method was used to discern differences among methods [38], with Bonferroni correction applied for multiple comparisons [39]. Diagnostic results across the four methods and the “gold standard” were denoted by frequency (proportion), with Cochran's Q test identifying differences and the Bonferroni method applied for corrections [40]. The comprehensive evaluation data, not fitting a normal distribution, was described using the median (interquartile range) and analyzed with the Friedman test. All statistical processes were executed in R software version 4.2.3. A two-sided P value below 0.05 indicated statistical significance. Table 3 displays the image quality assessments from the four doctors for the three modes.
Table 3. Comparison of subjective evaluation indicators and diagnostic results of the three diagnostic techniques [frequency (proportion) or median (interquartile range)]a
Our proposed method demonstrates enhanced performance in aspects like color contrast, brightness, and detail contrast when compared to both WLI and CBI modes. In terms of signal-to-noise ratio and overall evaluation, it surpasses CBI and performs on par with WLI, as indicated in Table 3. The gold standard reveals 66.02% inflammatory lesions and 33.98% intestinal atrophy/abnormal hyperplasia from pathological diagnosis. There are statistical differences between the three models and the gold standard (χ2 = 125.083, P,0.001). While our method slightly overestimates inflammatory lesions compared to the gold standard, it demonstrates a closer alignment with the gold standard in overall diagnostic accuracy than WLI and CBI. This suggests that our method offers a non-invasive yet reliable alternative for patient diagnosis. However, it is important to acknowledge that this technology requires further clinical validation to fully ascertain its efficacy and reliability.
The terms “negative” and “positive” represent diagnoses of inflammatory lesions and dysplasia or intestinal metaplasia, respectively. For the three modes, initial pairwise comparisons were conducted. Using McNemar's method [41], we compared the sensitivities of these diagnostic techniques, where sensitivity is the ability to identify intestinal metaplasia and dysplasia. Subsequent Bonferroni corrections were applied for multiple comparisons to ensure result accuracy. Results are in Table 4. Pairwise comparison reveals our proposed method's sensitivity (20.7%) surpasses both WLI (10.5%) and CBI (6.3%).
Table 4. Comparison of diagnostic sensitivity among the three modes.a
We benchmarked the outcomes from the three modes against the gold standard. The assessment was conducted considering both reliability and authenticity, using the gold standard as the reference. The comparative findings are presented in Table 5.
Table 5. Comparison of assessment metrics across the three modes.a
From Table 5, it's evident that our proposed method boasts the highest reliability with a coincidence rate of 75.00% and a Kappa value nearing the 0.4 threshold at 0.38. While still falling short of the gold standard, it outpaces the other two modes, which display considerably lower rates.
In terms of authenticity, our method shines with a sensitivity of 43.68%, making it superior in detecting dysplasia or intestinal metaplasia. On the other hand, CBI stands out for its specificity, identifying inflammatory lesions with a high rate of 98.22%. Yet, considering the overall diagnostic performance, our proposed method tops the list as evidenced by its leading Youden index, indicating its comprehensive diagnostic capability for various lesions.
The receiver operating characteristic curve (ROC) gauges the efficacy of detection methods by balancing sensitivity against 1-specificity. The area under the curve (AUC) quantifies this [42], with values spanning from 0.5 to 1.0; a perfect detection yields an AUC of 1.0. In our study, we compared the AUC values of the three methods to assess their diagnostic authenticity.
Table 6 presents the AUC values of the three modes. All three exceed 0.5, indicating their diagnostic accuracy. Our proposed method boasts the highest AUC at 0.67 (0.62, 0.73), followed by WLI at 0.59 (0.55, 0.64), and CBI at 0.57 (0.53, 0.60). This highlights the superior diagnostic realism of our proposed method.
Table 6. The area under curve (AUC) of three modesa
In summary, our proposed method surpasses WLI and CBI in reliability, authenticity, and subjective assessment. Its diagnostic precision and overall capability are notably superior. The experimental findings indicate that our approach offers clinicians more precise and vivid endoscopic images.
4. Discussion
In this paper, we designed a new light source mode and introduced advanced image processing algorithms. By integrating tissue data from WLI with vessel features from CBI, we offer more comprehensive image information. Superior image quality not only enhances early lesion detection but also aids in precise lesion assessment. In traditional endoscopy, doctors often need to switch between modes like WLI and CBI to gather diverse information, which can lead to potentially overlooking critical tissue details. Our method streamlines this process by eliminating the need for mode switching, presenting comprehensive data in a single image. This integration of key features from both WLI and CBI in one view enhances the efficiency of the diagnostic procedure, providing a more holistic and detailed tissue assessment.
Our method has diverse applications. Primarily, it's invaluable for early lesion screening, combining the tissue insights of WLI with CBI's vascular features. This enables large-scale observation, facilitating early lesion detection and aiding doctors in precise lesion identification during routine checks. Secondly, in surgical applications such as digestive tract operations and tissue biopsies, our approach offers enhanced image clarity, enabling surgeons to more accurately assess lesion extent and depth in the digestive tract and guide doctors to the optimal biopsy location, thereby minimizing unintended harm to nearby tissues and enhancing surgical accuracy and safety.
Each technology has its specific application scenarios and limitations. For situations requiring detailed observation using magnifying endoscopy, our proposed method might not be optimal. With magnification, the brightness of CBI is adequate and provides good imaging of superficial mucosal structures. While our method introduces a white light component for richer tissue information, it might be unnecessary at extremely high magnifications and poses a risk of image degradation. Our approach is better suited for routine, broad examinations to identify suspicious lesions. To optimize our method's adaptability and effectiveness across diverse inspections, further refinement is essential in real-world applications.
HbE technique improves blood vessel imagery by adjusting the red-green channel ratio in WLI images. Our method incorporates blue channel information, considering this alongside the red-green ratio. Given blue light's shorter wavelength and its scattering and absorption properties in tissues, it better reveals superficial mucosa fine structures. This inclusion enhances the depiction of fine blood vessels and mucosal patterns, offering a fresh perspective. Our method excels in visualizing superficial capillaries and mucosa, an area where HbE falls short.
This study offers significant insights and practical value for the diagnosis of gastrointestinal diseases. While acknowledging the exploratory nature of this innovative technique, it offers valuable support to clinicians in lesion assessment and diagnosis. However, our study encounters certain limitations, including a smaller sample size, a limited range of gastrointestinal conditions, and a lack of objective evaluations. Our future endeavors will focus on expanding collaborations with medical institutions, refining evaluative criteria, and broadening the spectrum of diseases studied. We are committed to integrating advanced imaging technologies with comprehensive clinical research to enhance early detection and improve patient care outcomes. To reinforce the validity of our findings, upcoming studies will concentrate on objective assessments, including image standard deviation, average gradient, spatial frequency, entropy, and mutual information. These metrics are intended to provide a more quantitative comparison of our method's efficacy against traditional WLI and CBI approaches.
5. Conclusion
Acknowledging both the advantages and limitations of WLI and CBI, our study explores an integrative endoscopic imaging method. This method uniquely combines these techniques through an optimized illumination spectrum tailored for gastrointestinal examinations. A newly developed software algorithm enhances the visualization of hemoglobin, potentially improving the interpretation of tissue features. This integrated approach aims to consolidate the strengths of established modalities, aiming for more nuanced and detailed imaging. Our system has been rigorously evaluated for safety and complies with international standards. Preliminary clinical trials suggest its effectiveness; however, further extensive validation is required to confirm these findings. While current evaluations are based on subjective clinical assessments by doctors, future work will involve a balanced approach, incorporating both objective and subjective analyses to affirm the efficacy of our technology. Clinicians have recognized the promise of our system; nevertheless, we acknowledge that its widespread clinical adoption and the confirmation of its superior performance will require a more extensive evidence base. We are hopeful that this system will make a significant contribution to clinical practices in the future.
Funding
National Natural Science Foundation of China (81930119, 82027807, U22A2051); National Key Research and Development Program of China (2022YFC2405200); Beijing Municipal Natural Science Foundation (7212202); Intelligent Healthcare Tsinghua University (2022ZLB001); Tsinghua-Foshan Innovation Special Fund (2021THFS0104).
Acknowledgments
Qiang Li, Yangming Wang, Xiaoxian Yuan, and Qingxiao Chen of Shanghai Aohua Endoscopy Co., Ltd. provided optical design support and manufacturing assistance, and Jia Gong and Chunxia Xiong made efforts in clinical trials.
Institutional Review Board approval was obtained. Written informed consent was obtained from all subjects (patients) in this study.
Disclosures
The authors declare no conflicts of interest.
Data Availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
References
1. H Ono, “Early gastric cancer: diagnosis, pathology, treatment techniques and treatment outcomes,” Eur. J. Gastroenterol. Hepatol. 18(8), 863–866 (2006). [CrossRef]
2. S Nakamoto, Y Sakai, D Kasanuki, et al., “Indications for the use of endoscopic mucosal resection for early gastric cancer in Japan: a comparative study with endoscopic submucosal dissection,” Endoscopy 41(09), 746–750 (2009). [CrossRef]
3. E Young, H Philpott, and R Singh, “Endoscopic diagnosis and treatment of gastric dysplasia and early cancer: Current evidence and what the future may hold,” World Journal of Gastroenterology. 27(31), 5126–5151 (2021). [CrossRef]
4. S Nagao, Y Tsuji, T Sakaguchi, et al., “Highly accurate artificial intelligence systems to predict the invasion depth of gastric cancer: efficacy of conventional white-light imaging, nonmagnifying narrow-band imaging, and indigo-carmine dye contrast imaging,” Gastrointestinal endoscopy. 92(4), 866–873.e1 (2020). [CrossRef]
5. DJ Waterhouse, CR Fitzpatrick, SE di Pietro, et al., “Emerging optical methods for endoscopic surveillance of Barrett's oesophagus,” The Lancet Gastroenterology & Hepatology. 3(5), 349–362 (2018). [CrossRef]
6. LMWK Song, DG Adler, WM Chand, et al., “Chromoendoscopy,” Gastrointestinal endoscopy. 66(4), 639–649 (2007). [CrossRef]
7. C Patriotis and S Srivastava, “Early Cancer Detection: Challenges and Opportunities,” Circulating Tumor Cells: Advances in Liquid Biopsy Technologies. 1, 619–631 (2023). [CrossRef]
8. J Gao, X Zhang, M Meng, et al., “Linked color imaging can improve detection rate of early gastric cancer in a high-risk population: a multi-center randomized controlled clinical trial,” Dig. Dis. Sci. 66(4), 1212–1219 (2021). [CrossRef]
9. T Horimatsu, Y Sano, H Tanaka, et al., “Next-generation narrow band imaging system for colonic polyp detection: a prospective multicenter randomized trial,” Int. J. Colorectal. Dis. 30(7), 947–954 (2015). [CrossRef]
10. K Yao, H Doyama, M Gotoda, et al., “Diagnostic performance and limitations of magnifying narrow-band imaging in screening endoscopy of early gastric cancer: a prospective multicenter feasibility study,” Gastric Cancer 17(4), 669–679 (2014). [CrossRef]
11. K Matsumoto, H Ueyama, H Yao, et al., “Diagnostic limitations of magnifying endoscopy with narrow-band imaging in early gastric cancer,” Endosc. Int. Open. 8(10), E1233–E1242 (2020). [CrossRef]
12. V Du Le, Q Wang, TJ Gould, et al., “Vascular contrast in narrow-band and white light imaging,” Appl. Opt. 53(18), 4061 (2014). [CrossRef]
13. S Zhang, Y Fu, H Zhang, et al., “A novel endoscopy image fusion system: combine white light imaging and compound band imaging,” Int. J. CARS. 1, 1–14 (2023). [CrossRef]
14. T Uraoka and M Igarashi, “Development and clinical usefulness of a unique red dichromatic imaging technology in gastrointestinal endoscopy: A narrative review,” Therap. Adv. Gastroenterol. 15, 17562848221118302 (2022). [CrossRef]
15. H Kanzaki, R Takenaka, K Kawahara, et al., “Linked color imaging (LCI), a novel image-enhanced endoscopy technology, emphasizes the color of early gastric cancer,” Endoscopy international open. 05(10), E1005–E1013 (2017). [CrossRef]
16. N Yoshida, O Dohi, Y Inoue, et al., “Blue laser imaging, blue light imaging, and linked color imaging for the detection and characterization of colorectal tumors,” Gut and liver. 13(2), 140–148 (2019). [CrossRef]
17. E Klenske, S Zopf, F Neufert, et al., “I-scan optical enhancement for the in vivo prediction of diminutive colorectal polyp histology: results from a prospective three-phased multicentre trial,” PLoS One 13(5), e0197520 (2018). [CrossRef]
18. S Kodashima and M Fujishiro, “Novel image-enhanced endoscopy with i-scan technology,” World journal of gastroenterology: WJG. 16(9), 1043 (2010). [CrossRef]
19. R Coriat, A Chryssostalis, S Zeitoun, et al., “Computed virtual chromoendoscopy system (FICE): a new tool for upper endoscopy?” Gastroenterol. Clin. Biol. 32(4), 363–369 (2008). [CrossRef]
20. K Yao, M Kato, and J Fujisaki, “Techniques using the hemoglobin index of the gastric mucosa,” Endoscopy 37(05), 479–486 (2005). [CrossRef]
21. G Kamphuis, D de Bruin, T Fallert, et al., “Storz professional image enhancement system: a new technique to improve endoscopic bladder imaging,” J. Cancer Sci. Ther. 08(03), 71 (2016). [CrossRef]
22. MA Manfredi, BKA Dayyeh, JT Bhat, et al., “Electronic chromoendoscopy,” Gastrointestinal endoscopy. 81(2), 249–261 (2015). [CrossRef]
23. SL Jacques, “Optical properties of biological tissues: a review,” Phys. Med. Biol. 58(11), R37–R61 (2013). [CrossRef]
24. K. Yao, Principles of Magnifying Endoscopy with Narrow-Band Imaging. Zoom Gastroscopy. (Springer, 2014). p. 49–56.
25. Z He, P Wang, X Liang, et al., “Clinically available optical imaging technologies in endoscopic lesion detection: Current status and future perspective,” Journal of Healthcare Engineering. 2021, 1 (2021). [CrossRef]
26. S Prahl, “Optical absorption of hemoglobin,” Oregon Medical Center News. (1999).
27. AN Bashkatov, EA Genina, VV Kochubey, et al., eds. “Optical properties of human stomach mucosa in the spectral range from 400 to 2000nm,” International Conference on Lasers, Applications, and Technologies 2007: Laser Technologies for Medicine; 2007: SPIE.
28. T Ueda, K Morita, Y Koyama, et al., “A detailed comparison between the endoscopic images using blue laser imaging and three-dimensional reconstructed pathological images of colonic lesions,” PLoS One 15(6), e0235279 (2020). [CrossRef]
29. K Gono, T Obi, T Yamaguchi, et al., “Appearance of enhanced tissue features in narrow-band endoscopic imaging,” J. Biomed. Opt. 9(3), 568 (2004). [CrossRef]
30. WM Mellette, GM Schuster, and JE Ford, “Planar waveguide LED illuminator with controlled directionality and divergence,” Opt. Express 22(S3), A742–A58 (2014). [CrossRef]
31. JC Miñano, P Benítez, A Blen, et al., “High-efficiency free-form condenser overcoming rotational symmetry limitations,” Opt. Express 16(25), 20193 (2008). [CrossRef]
32. T Kari, J Gadegaard, K Søndergaard, et al., “Reliability of point source approximations in compact LED lens designs,” Opt. Express 19(S6), A1190–A5 (2011). [CrossRef]
33. RE Fischer, B Tadic-Galeb, J BGA Yoder, et al., “Optical system design,” Citeseer (2000).
34. S Tsuji, N Sato, T Kawano, et al., “Functional imaging for the analysis of the mucosal blood hemoglobin distribution using electronic endoscopy,” Gastrointestinal endoscopy. 34(4), 332–336 (1988). [CrossRef]
35. K Nakamura, “DEVELOPMENT OF REAL-TIME ENDOSCOPIC IMAGE PROCESSING TECHNOLOGY: ADAPTIVE INDEX OF HEMOGLOBIN COLOR ENHANCEMENT PROCESSING,” Digestive Endoscopy. 14(s1), S40–S7 (2002). [CrossRef]
36. L Lévêque, M Outtas, L Liu, et al., “Comparative study of the methodologies used for subjective medical image quality assessment,” Phys. Med. Biol. 66(15), 15TR02 (2021). [CrossRef]
37. SE Ghrare, MAM Ali, K Ismail, et al., eds. “Diagnostic quality of compressed medical images: Objective and subjective evaluation,” 2008 Second Asia International Conference on Modelling & Simulation (AMS); 2008: IEEE.
38. DG Pereira, A Afonso, and FM Medeiros, “Overview of Friedman’s test and post-hoc analysis,” Communications in Statistics-Simulation and Computation. 44(10), 2636–2653 (2015). [CrossRef]
39. RJ Cabin and RJ Mitchell, “To Bonferroni or not to Bonferroni: when and how are the questions,” Bulletin of the ecological society of America. 81(3), 246 (2000). [CrossRef]
40. DC Hoaglin, “Misunderstandings about Q and ‘Cochran's Q test'in meta-analysis,” Statistics in Medicine 35(4), 485–495 (2016). [CrossRef]
41. OA Adedokun and WD Burgess, “Analysis of paired dichotomous data: A gentle introduction to the McNemar test in SPSS,” J Multidiscip Eval 8(17), 125–131 (2011). [CrossRef]
42. JM Lobo, A Jiménez-Valverde, and R. Real, “AUC: a misleading measure of the performance of predictive distribution models,” Global ecology and Biogeography. 17(2), 145–151 (2008). [CrossRef]
