Red dichromatic imaging improves visibility of bleeding during gastric endoscopic submucosal dissection – Scientific Reports

Study design

In this retrospective study, we reviewed the data of 31 consecutive patients who underwent gastric ESD at Fukuchiyama City Hospital from September 2020 to January 2021. The inclusion criterion consisted of the presence of coagulation for intraoperative bleeding with a shift from white light imaging (WLI) to RDI. The visibility of bleeding during ESD was compared between WLI and RDI. For this purpose, the visibility score of the bleeding point as a subjective measure was defined by the operators and the color value and difference between the bleeding points and surrounding pooled blood were calculated as objective measures. In addition, the visibility score was analyzed between experienced (≥100 cases of ESD) and non-experienced (<100 cases of ESD) operators. Furthermore, an analysis of the color difference according to whether RDI could improve the visibility score or not, was performed to elucidate the association between the visibility score and color difference. Finally, a sub-analysis of the visibility score to evaluate the beneficial effect of RDI in terms of the bleeding characteristics, was performed. The study was conducted in accordance with the Declaration of Helsinki, and the protocol was approved by the Institutional Review Board of Fukuchiyama City Hospital (No. 2–47). The opt-out method was used to obtain informed consent owing to its retrospective design.

Visibility score analysis

The operators evaluated visibility score immediately after ESD. We defined the visibility score as four numeric values according to previous studies where 1 corresponds to poor visibility and almost impossible to detect and 4 corresponds to excellent visibility and easy to detect. Moreover, numeric value 2 refers to fair visibility and slightly possible to detect with careful observation, whereas numeric value 3 refers to good visibility and possibility of detection with careful observation^14,15.

Color difference analysis

We calculated the color difference according to the procedure given in the CIE 1976 L^*a^*b color space (CIELAB). The CIELAB defines a three-dimensional color system that represents human color perception where L^* represents lightness, a^* represents the red/green axis, and b^* indicates the yellow/blue axis¹⁶. Three points at 3 × 3 pixels each in the bleeding points and in the surrounding pooled blood were plotted as the regions of interest (ROIs) in the images that were extracted from videos and had similar compositions in WLI and RDI. The flood selection tools of the image editing software Affinity Photo 1.10.4 (Serif. Inc, Nottingham, UK) were applied to select pixels of similar color and CIELAB color values in ROIs were obtained. We compared the averages of L^*, a^*, and b^* color values in ROIs between the bleeding points and surrounding pooled blood and calculated the color value differences between WLI and RDI as ΔL, Δa, and Δb. Finally, we calculated the color difference or ΔE, which represent the distance between the bleeding points and surrounding pooled blood on the three-dimensional color space CIELAB, using the following formula:

Figure 1 shows the representative images of bleeding and the selection of the ROI. Figure 2 shows a representative color difference on the CIELAB.

The CIE 1976 L^*a^*b color space expressed on three-dimensional geometry with the bleeding point plotted. The outlined circle and triangle indicate the color values on white light imaging (WLI). The filled circle and triangle indicate the color values on red dichromatic imaging (RDI).

Evaluation of bleeding characteristics with RDI

We evaluated three categories of bleeding visibility using the captured video images: bleeding type (gushing or oozing), presence of a pulsatile vessel (present or absent), and submergence of the bleeding point (present or absent). The representative images of these categories are shown in Fig. 3. The classifications were performed by a single endoscopist (K.O.). We then assessed the improvement of the bleeding characteristics with RDI using the visibility score.

Representative images of bleeding categories on RDI. (a) Gushing bleeding, (b) oozing bleeding, (c) presence of pulsatile vessel, and (d) submergence of the bleeding spot. RDI, red dichromatic imaging

Endoscopic submucosal dissection

A mucosal incision and submucosal dissection were performed. The bleeding point was evaluated using WLI followed by RDI when hemostasis was required for bleeding. All coagulation procedures were performed only on RDI. The time for hemostasis and the number of coagulations attempted were calculated.

All ESD procedures were performed using the EVIS X1 endoscopic system with a GIF-H290T endoscope, which were installed at our institution in August 2020. We used ITknife2, DualKnife, or a 3.5 mm long Clutch Cutter (Fujifilm Medical, Tokyo, Japan). RDI mode 1 was used that was preferred for hemostasis¹⁷. Submucosal injections with a mixture of 0.6% sodium alginate solution Liftal K, (Kaigen Pharma Co., Tokyo, Japan), physiological saline solution, and indigo carmine were administered. Endoscopic hemostasis was performed using ITknife2, Clutch Cutter, or Coagraspar G with a VIO 3 electrosurgical unit (Erbe Elektromedizin GmbH, Tübingen, Germany). The procedures were recorded using a high-definition medical video recorder UR-4MD (TEAC Europe GmbH, Wiesbaden, Germany). All devices were produced by Olympus Co. Tokyo, Japan unless stated otherwise. The procedures were performed by five operators (K.O., N.I., T.O., T.T., and H.S.) among whom three (N.I., T.O., and T.T.) were experienced operators (≥100 of ESD) and two (K.O. and H.S.) were non-experienced operators (<100 cases).

Statistical analysis

Numerical variables are presented as mean ± standard deviation. Comparisons between the two groups were performed using the Wilcoxon signed-rank test. Multivariate logistic regression analysis was performed to evaluate bleeding characteristics with RDI for variables that reached p < 0.05 in the univariate analysis. Statistical significance was set at p < 0.05. All statistical analyses were conducted using JMP14.3 (SAS Institute Inc., Cary, NC, USA).