When observing the esophageal mucosa under Blue Light Imaging (BLI) [1-4] in intermediate to distant views, differences in color were observed among cases, ranging from brownish (Figure 1) to blue-green (Figure 2). To elucidate the cause of these color differences, a preliminary investigation was conducted using parameters from each case. The results indicated a tendency toward stronger blue-green color in young male patients and suggested a correlation with hemoglobin (Hb) concentration. Therefore, this study aimed to examine in detail the relationship between image color and Hb concentration.

Figure 1: Two Cases in Which the Esophageal Mucosa Observed under BLI Appears Brownish are Presented. The Hemoglobin Level in Case A is 6.1 mg/dL, and in Case B it is 9.7 mg/dL.

Figure 2: Two Cases in Which the Esophageal Mucosa Observed under BLI Appears Bluish-Green are Presented. The Hemoglobin Level in Case A is 15.7 mg/dL, and in Case B it is 15.3 mg/dL.

The primary objective of this research is to clarify whether a significant correlation exists between the color of esophageal BLI images and Hb concentration. Furthermore, if this correlation can be confirmed, it could enable applications such as predicting Hb concentration using endoscopic AI and non-invasive assessment of Hb concentration.

Additionally, in this study, the Green/Red (G/R) ratio and Blue/Red (B/R) ratio were calculated and evaluated from the Red-Green-Blue (RGB) color model of endoscopic images. These methods may also have potential applications in other clinical fields.

This study was conducted as a retrospective observational study. The subjects included cases in which upper gastrointestinal endoscopy and complete blood count were performed simultaneously at Nishikawa Gastrointestinal Clinic. These cases were collected consecutively to avoid selection bias. Additionally, the same endoscope and processor were used, and images were captured under identical settings. All procedures were performed by the one endoscopist to eliminate variability caused by different operators. The endoscopy system utilized a FUJIFILM EG-840N scope and a FUJIFILM EP-8000 processor. Blue Light Imaging (BLI) was performed in BLI-bright mode.

A power analysis was conducted beforehand using G*Power (version 3.1.9.7) [5]. It was estimated that a minimum sample size of 20 cases would be required to meet an alpha error of 0.05 and a power of 0.8, assuming a correlation coefficient of approximately 0.6. To account for potential missing data and variability during the actual collection process, a more conservative sample size of 50 cases was set.

Exclusion criteria included cases with extensive esophageal lesions. Information on the subjects was obtained from medical records and endoscopic images.

Investigation Items and Analysis Methods

The color of the esophageal mucosa was assessed by calculating the average RGB values from images taken during upper gastrointestinal endoscopy. The RGB color model of the hole endoscopic image were calculated using an image color averaging tool (ImgAverager Version 1.0.1). To eliminate the influence of brightness and darkness, the analysis focused on the ratios of green to red (G/R) and blue to red (B/R), rather than absolute RGB values, and their correlation with Hb concentration was examined.

Six images in total were taken during the insertion and removal of the upper, middle, and lower esophagus, respectively. From each image, the G/R ratio and B/R ratio were calculated. The average values were then determined from the four data points obtained by excluding the highest and lowest values among the six, and the correlation with Hb concentration was examined. To examine potential data bias, the frequency of occurrence of the highest and lowest values was tested using the chi-squared goodness-of-fit test.

Furthermore, for the 37 cases from the 13th to the 50th, images were taken in three locations of the upper, middle, and lower esophagus during insertion, using three modes: White Light, Linked Color Imaging (LCI) [6-8], and BLI. The G/R ratio was calculated for each image, and the correlation between the average values for each mode and Hb concentration was examined.

Statistical Analysis

The correlation coefficients (Pearson's correlation coefficient) [9,10] for each item were calculated, and p-values were determined using t-tests. The threshold for statistical significance was set at p < 0.05. Microsoft Excel was used for the analysis. The statistical power of the sample size was evaluated using G*Power.

A total of 50 cases were included, with an average age of 65.3 ± 15.8 years. The gender distribution was 24 males and 26 females. The endoscopic diagnoses were primarily atrophic gastritis (27 cases with prior H. pylori infection and 3 cases with current H. pylori infection), followed by gastric fundic gland polyps (8 cases), reflux esophagitis (3 cases), and one case each of gastric ulcer, duodenal ulcer, erosive gastritis, and gastric cancer. Additionally, 5 cases showed no abnormal findings.

Correlation between Image Analysis and Blood Data

A significant positive correlation was observed between the G/R ratio under BLI and Hb concentration, with a correlation coefficient of 0.628 (p=0.016) (Figure 3). Similarly, the B/R ratio under BLI showed a significant positive correlation with Hb concentration, with a correlation coefficient of 0.612 (p=0.020) (Figure 4).

Figure 3: Scatter Plot Showing the Relationship between Hb Concentration and G/R Ratio under BLI. The Correlation Coefficient (r) = 0.628, p=0.016.

Figure 4: Scatter Plot Showing the Relationship between Hb Concentration and B/R Ratio under BLI. The Correlation Coefficient (r) = 0.612, p=0.020.

Furthermore, when examining relationships with other blood test parameters, the G/R ratio showed a significant positive correlation with hematocrit (Hct) at 0.602 (p=0.023), and the B/R ratio also correlated positively with Hct at 0.585 (p=0.029). However, correlations with red blood cell count did not reach statistical significance, with G/R at 0.487 (p=0.087) and B/R at 0.454 (p=0.117).

Statistical Power and Validity

With a sample size of 50 cases, the statistical power to detect a correlation coefficient of approximately 0.6 (effect size 0.6) was estimated at about 0.85, indicating sufficient reliability and generalizability of the study results. Additionally, since the sample size exceeds the minimum recommended number of 20 cases for an effect size of 0.6, the tests are considered to have adequate detection power.

Analysis of Bias in Maximum and Minimum Values at Measurement Sites

A chi-squared goodness-of-fit test was conducted to examine whether there was any bias in the occurrence frequency of maximum and minimum values at each measurement site (upper, middle, and lower parts during insertion, and upper, middle, and lower parts during removal). The results showed no significant bias (p > 0.05). Additionally, no significant difference was observed in the frequency of maximum and minimum values between the insertion and removal phases (p > 0.05). These findings suggest that biases related to measurement position or timing are minimal.

Correlation Analysis by Observation Mode

For the 37 cases from the 13th to the 50th, the correlation between the G/R ratio and Hb concentration under white light, LCI, and BLI was compared, with the following results:

Under white light, the correlation coefficient was -0.189 (p > 0.05) (Figure 5)

Under LCI, the correlation coefficient was -0.085 (p > 0.05) (Figure 6)

Under BLI, the correlation coefficient was 0.600 (p = 0.024) (Figure 7)

Figure 5: Scatter Plot Showing the Relationship between Hb Concentration and B/R Ratio under White Light. The Correlation Coefficient (r) = -0.189, p > 0.05.

Figure 6: Scatter Plot Showing the Relationship between Hb Concentration and G/R Ratio under LCI. The Correlation Coefficient (r) = -0.085, p > 0.05.

Figure 7: Scatter Plot Showing the Relationship between Hb Concentration and B/R Ratio under BLI. The Correlation Coefficient (r) = 0.600, p=0.024.

These results indicate that only BLI showed a significant positive correlation, suggesting that the relationship between the G/R ratio and Hb values is specific to BLI.

In this study, we examined the correlation between the color of the esophageal mucosa under BLI and the Hb concentration. As a result, a significant positive correlation was observed between the G/R ratio and B/R ratio in BLI images and the Hb concentration. Additionally, we investigated the relationship with red blood cell count and hematocrit value; although the correlation with hematocrit was not as strong as with Hb, a significant positive correlation was still observed. Conversely, no clear correlation was found with red blood cell count. These findings suggest that the color in BLI images is particularly related to the Hb concentration in the blood.

Notably, the positive correlation between the G/R ratio, B/R ratio, and Hb concentration indicates the potential to estimate blood Hb concentration from the images. Non-invasive assessment of Hb has long been desired, and various attempts have been reported [11]. The results of this study suggest that this method could be a promising approach among these non-invasive evaluation techniques.

Furthermore, regarding the measurement of color, no bias was observed in the maximum and minimum values during insertion and removal, indicating a low likelihood of bias based on measurement site or timing. Additionally, compared to white light and Linked Color Imaging (LCI), the correlation between the G/R ratio and Hb concentration appears to be characteristic of BLI, demonstrating its specificity.

Additionally, we attempted a method to measure RGB values by cropping only the regions within the endoscopic images where the mucosa was sufficiently stretched and the capillary vessels were clearly observable. However, this approach resulted in significant variability in the data depending on the cropping location, leading to potential bias from arbitrary selection. Therefore, to avoid such bias, we measured the RGB values across the entire image.

When observing the esophageal mucosa in mid- to distant views under BLI, blood vessels, including capillaries, appear in a blue-green color. Conversely, in cases with high Hb concentration, the background mucosa outside the vessels also takes on a blue-green color, making the capillaries less prominent. In contrast, in cases with low Hb concentration, the background mucosa exhibits a slightly brownish tone, and the capillaries are more distinctly observed. It is presumed that these differences in background mucosal color are the main factors influencing the changes in G/R ratio and B/R ratio.

Hemoglobin has a particular absorption characteristic near 410 nm [12]. Since BLI strongly illuminates this wavelength range (blue-violet) [1-4], it is possible that variations in color are caused by differences in Hb concentration in the blood. Furthermore, the observation that blood-rich vascular areas appear blue-green under BLI may be related to the fact that esophageal mucosa with high Hb concentration appears blue-green. However, the precise optical mechanisms require interpretation by optics experts, and in this study, we aim only to demonstrate the correlation between G/R ratio, B/R ratio, and Hb concentration.

The significance of this study lies in demonstrating that the differences in esophageal mucosal color observed on BLI are attributable to variations in Hb concentration. Furthermore, the G/R ratio and B/R ratio obtained from BLI images could serve as new, non-invasive, and rapid tools for evaluating Hb concentration. These metrics have the potential to contribute to future clinical applications and automated analysis using endoscopic AI (Artificial Intelligence) [13-15]. Moreover, elucidating the optical mechanisms involved could expand the utility of these color indicators beyond Hb measurement, potentially serving as new markers for other pathological conditions.

On the other hand, there are several limitations to consider. Firstly, the sample size is relatively small, consisting of only 50 cases, so caution is required when generalizing the findings. Additionally, although efforts were made to expand the esophagus by infusing air sufficiently to allow the capillaries to be observed, standardization of imaging conditions was challenging, which may have led to variability in G/R and B/R ratios.

Future challenges include conducting larger-scale, multi-center studies, including patients with other diseases such as malignant diseases, and advancing the automation and standardization of image analysis. Additionally, observation and evaluation using magnifying endoscopy are also considered important.

In conclusion, a significant correlation was observed between RGB-related metrics of the esophageal mucosa using BLI and Hb concentration, indicating that this method could become a useful tool for non-invasive Hb assessment and anemia monitoring in the future. Looking ahead, examining correlations between other clinical indicators and RGB-related metrics is anticipated to contribute to the application of this technique in other fields and to the development of non-invasive diagnostic methods utilizing image analysis with AI.

1. Yamamoto H, et al. Blue Light Imaging (BLI) for the detection of early gastric cancer: a prospective study. Gastrointestinal Endoscopy. 2017; 85: 747-755.
2. Kobayashi G, et al. Comparison of Blue Light Imaging and Narrow Band Imaging in the detection of colorectal neoplasia. Endosc Int Open. 2018; 6: E448-E453.
3. Kudo S, et al. Advances in image-enhanced endoscopy: Blue Light Imaging and related technologies. World J Gastroenterol. 2019; 25: 2974-2984.
4. Fujimoto A, et al. Clinical utility of Blue Light Imaging in gastrointestinal endoscopy. Dig Endosc. 2020; 32: 319-326.
5. Faul F, Erdfelder E, Buchner A, Lang A-G. Statistical power analyses using G*Power 3.1: Tests for correlation and regression analyses. Behav Res Methods. 2019; 41: 1149-1160.
6. Yamada A, et al. Linked Color Imaging (LCI) in Endoscopy: A New Technique for Enhanced Visualization of Mucosal Lesions. astrointest Endosc Clin N Am. 2018; 28: 251-263.
7. Yamamoto H, et al. Clinical Utility of Linked Color Imaging (LCI) in the Detection of Gastric and Colorectal Lesions. World J Gastroenterol. 2019; 25: 2454-2464.
8. Kobayashi G, et al. Advances in Endoscopic Imaging: The Role of Linked Color Imaging (LCI) in Gastroenterology. Digestive Endoscopy. 2020; 32: 456-462.
9. Pearson K. Mathematical contributions to the theory of evolution. III. Regression, correlation, and the law of errors. Philosophical Transactions of the Royal Society of London. Series A. Containing Papers of a Mathematical or Physical Character. 1896; 187: 253-318.
10. Veenstra C, Kruitwagen S, Groener D, Petersen W, Steenbergen W, Bosschaart N. Quantification of total haemoglobin concentrations in human whole blood by spectroscopic visible-light optical coherence tomography. Sci Rep. 2019; 9: 1-8.
11. van Kampen EJ, Zijlstra WG. Spectrophotometric analysis of hemoglobin and its derivatives. Adv Clin Chem. 1983; 23: 199-257.
12. Hirasawa T, et al. Development and Validation of a Deep Learning Model for Real-Time Detection of Esophageal Cancer during Endoscopy. Gastroenterol. 2018; 154: 2027-2035.
13. Liu X, et al. Artificial Intelligence in Gastrointestinal Endoscopy: A Review. World J Gastroenterol. 2020; 26: 3854-3864.
14. Kudo S, et al. Artificial Intelligence in Endoscopy: Current Status and Future Perspectives. Digestive Endoscopy. 2020; 32: 385-392.