Visibility Evaluation of Fundic Gland Polyp Associated With Proton Pump Inhibitor in Texture and Color Enhancement Imaging

2.1 Study Design and Patients

In a single-center prospective clinical study, we evaluated whether GCS captured during endoscopy was more readily detected by TXI-1, TXI-2, and NBI compared to WLI. We took biopsies of GCS areas within the registered lesions to examine GCS histology. We enrolled patients who underwent esophagogastroduodenoscopy (EGD) with WLI, TXI-1, TXI-2, and NBI from April 2021 to October 2022 at our hospital, and who received a histopathological diagnosis of PPI-FGP. The inclusion criteria for enrolment were: patients over 20 years old with a history of taking PPIs/PCABs for at least 6 months and were continued to take it to the current time. Endoscopies were performed for a variety of reasons, including the presence of abdominal pain, medical check-ups, gastroesophageal reflux disease symptoms, and follow-up appointments for all of these issues. We excluded patients with a history of gastrectomy or advanced gastric cancer, and those ineligible for endoscopy due to serious primary diseases such as heart, hepatic, or respiratory failure [25].

2.2 EGD Procedure

We used GIF-H290Z, GIF-XZ1200 endoscopes, the EVIS X1, and CV-1500 systems (Olympus Medical Systems Corporation) [25]. All endoscopic examinations were performed by expert endoscopists (Yoichi Akazawa, Tsutomu Takeda, and Hiroya Ueyama). Imaging was first performed using WLI, which was followed by TXI-1, TXI-2, and NBI. The WLI image enhancement level was set to B6, and the TXI image enhancement level was set to moderate. Three expert endoscopists (Yoichi Akazawa, Tsutomu Takeda, and Hiroya Ueyama) assessed and discussed WLI, TXI, and NBI images until all were in agreement for each image. JPEG images were downloaded from a server in a lossless format; each image file was about 100 kB, with a pixel array of 640 × 510 and 24-bit color [25].

2.3 Definition and Histopathological Analysis of PPI-FGP, GCS, and Black Spots

A PPI-FGP diagnosis was based on the presence of lesions with at least one of the following histopathological findings following biopsy: parietal cell protrusion, foveolar cell hyperplasia, or enlargement of fundic gland cysts [26-31]. GCS was defined as a mottled or circular gray translucent finding in PPI-FGP (Figure 1A–F) [24]. Black spots (BSs) were defined as black pigmentation in gastric mucosa by conventional endoscopic WLI (Figure 2A) [3]. BSs have been known to be composed of brownish substances and eosinophilic materials in fundic gland cysts, which are thought to result from the accumulation of secretion from the lining cells of fundic gland cysts [3, 24, 32].

Following the isolation of biopsies from the GCS region, we analyzed the following characteristics: the presence of parietal cell protrusion, foveolar cell hyperplasia, histopathological classification of enlarged fundic gland cysts just below the surface (fundic gland type, foveolar type, and mixture type), and presence of brownish substances and eosinophilic materials in fundic gland cysts.

2.4 Evaluation of GCS Visibility

Ten endoscopists (five experts and five trainees) compared GCS visibility between TXI-1, TXI-2, NBI, and WLI. Images were presented randomly against a dark background in PowerPoint. WLI, TXI-1, TXI-2, and NBI images were shown next to each other. Endoscopists were blinded with respect to clinical data and image capture dates. All endoscopists were not familiar with the characteristics of TXI and GCS and had not received any prior guidance. Combined score classifications were as follows: 40 points, improved visibility; 21–39 points, comparable to WLI; <20 points, decreased visibility. The intra-class correlation coefficient (ICC) was used to indicate inter- and intra-rater reliability [25].

2.5 Color Analysis

After processing in Adobe Photoshop CC 2019, images were assessed using L* a* b* (L* = light/dark; a* = red/green; b* = yellow/ blue) color scores in a Commission Internationale de l'éclairage (CIE) LAB color space system [33]; this method was described previously [34]. A region of interest (ROI; 20 × 20 pixels) was for both gray and non-GCS areas in PPI-FGP images. Color values (L, a, b) within the ROI and their mean [A1] were quantified from a histogram analysis. L, a, and b represent the lightness, green-red axis, and blue-yellow axis, respectively, in the CIELAB color space. The L, a, and b values were converted to the CIELAB color space using the following transformations: L* = L/256 × 100; a* = a – 128; b* = b – 128 [35, 36]. Color differences (ΔE*) between pixel values within the ROI were calculated using the Euclidean distance formula in the CIELAB color space to evaluate the perceptual difference between color images. The formula used was: ΔE = [(ΔL) 2+(Δa*)2+(Δ b*) 2] 1/2).

2.6 Statistical Analysis

Differences in visibility scores rated by trainees, experts, ΔE* and L* a* b* color values between images were evaluated for statistical significance (p < 0.01) using the Wilcoxon rank sum test. Inter- and intra-rater reliability was tested using the ICC with 95% confidence intervals (CIs). Inter-rater reliability for interval, ordinal, and ratio variables was assessed using the ICC where multiple coders were involved [37]. Reliability was classified as follows: “perfect” when ICC was 1.0, “excellent” when >0.81, “substantial” when 0.80–0.61, “moderate” when 0.60–0.41, “fair” when 0.40–0.21, and “slight” when <0.20 [38, 39]. The use of 10 assessors ensured that the sample size was sufficient to generate reliable estimates. A minimum acceptable reliability coefficient (ρ₀) of 0.7 and an expected reliability coefficient (ρ₁) of 0.8 were established as benchmarks. A difference at α = 0.05 and β = 0.2 was detected using a sample size of 23 [40]. All statistical analyses were performed using SAS v. 9.4 (SAS Institute, Cary, NC, USA).

3.1 Baseline Characteristics

We enrolled 19 patients, from whom a total of 23 PPI-FGP lesions were evaluated (Table 1). The mean age was 65.5 years (range: 51–84); 11 patients were male and eight were female. All patients were taking PPIs or PCABs, with esomeprazole the most commonly taken PPI (eight cases) followed by PCABs taken by four patients. The administration period of PPIs or PCABs was at least 15 months for all patients, with a median of 74 (±45.7) months. The most common reason for taking PPIs or PCABs was reflux esophagitis (eight cases), although they were also taken to prevent ulcers induced by other agents such as anti-thrombotic drugs. The median diameter of PPI-FGPs was 9 mm (range: 5–20). Atrophic gastritis was rare in this patient group and almost all patients were Helicobacter pylori- negative.

3.2 Histopathological Features of PPI-FGP and GCS

The histopathological features of PPI-FGP and GCS are shown in Table 2. Both parietal cell protrusion and foveolar cell hyperplasia were found in almost all lesions (91.3% [21/23] and 95.7% [22/23], respectively). Enlargement of fundic gland cysts and fundic gland cysts just below the surface were observed in all lesions. The gland subtypes were fundic (91.3%; 21/23), foveolar (65.2%; 15/23), and mixed (78.2%; 18/23). No lesions were exclusive of the fundic gland type, and all lesions contained either foveolar type, mixed type, or both. No brownish substances or eosinophilic materials were found in any of the fundic gland cysts.

3.3 Evaluation and Scoring of GCS Visibility

Comparisons of the performance of TXI-1, TXI-2, or NBI and WLI with regard to GCS visibility were carried out by both trainee and expert endoscopists (Table 3). Compared to WLI, all endoscopists reported that GCS visibility was improved by TXI-1 (82.6%; 19/23) and TXI-2 (86.9%; 20/23). Trainees reported improved visibility for TXI-1 (95.6%; 22/23) and TXI-2 (86.9%; 20/23), and experts reported improved visibility for TXI-1 (82.6%; 19/23) and TXI-2 (86.9%; 20/23). None of the observers reported an improvement in GCS visibility when NBI was used.

The visibility scores reported by experts, trainees, and all endoscopists for TXI and NBI compared with WLI are shown in Table 4. Total visibility scores were as follows: TXI-1 (44.9), TXI-2 (42.9), and NBI (17.4) for all endoscopists; TXI-1 (22.7), TXI-2 (21.8), and NBI (8.5) for trainee endoscopists; TXI-1 (22.2), TXI-2 (21.1), and NBI (8.8) for expert endoscopists. Total visibility scores for all endoscopists were significantly higher using TXI-1 and TXI-2 compared with NBI (p < 0.01). The TXI-1 and TXI-2 visibility scores reported by trainees were not significantly different from those reported by experts.

3.4 Inter-Rater Reliability

Table 5 shows the comparison of inter-rater reliability scores (reported as ICC values) for each rater subgroup and for all endoscopists. The comparison of TXI-1 and TXI-2 to WLI by trainees gave ICC scores of 0.858 and 0.889, respectively. For experts, the corresponding scores were 0.864 and 0.817, while for endoscopists the scores were 0.931 and 0.927. Since all scores were >0.81, we conclude that the inter-rater reliability with regard to comparisons of TXI-1 and TXI-2 to WLI was “excellent” for all rater subgroups [25].

3.5 Objective Evaluations

Representative endoscopic images using WLI, TXI-1, and TXI-2 with ROIs are shown in Figure 3. L* a* b* color values for non-GCS (yellow box) and GCS (black box) were calculated. Table 6 shows the comparison of L* a* b* values for GCS and non-GCS images within the PPI-FGP. There was a significant difference in the a* b* values (p < 0.01) for GCS when TXI-1 was compared to WLI. L* values were significantly lower when TXI-2 was compared to WLI. ΔE* values were as follows: WLI, 12.0 ± 4.3; TXI-1, 22.3 ± 7.8; TXI-2, 25.9 ± 7.0. Significant differences in ΔE* were observed for both TXI-1 and TXI-2 compared to WLI (p < 0.01). There was no significant difference in ΔE* between TXI-1 and TXI-2.