Introduction
Barrett’s esophagus (BE) is defined as columnar metaplasia of the distal esophagus.
It involves abnormal cellular changes in the esophageal mucosa and is considered a
precursor of BE adenocarcinoma (BEA) [1], the incidence rate of which has been rapidly increasing in Western countries [2]. In Asian countries, including Japan, the incidence of BEA has also risen owing
to an increase in the prevalence of obesity and gastroesophageal reflux disease as
a consequence of lower Helicobacter pylori infection rates [3]
[4].
The prognosis of BEA is poor when the disease is diagnosed in an advanced stage [5] but is relatively favorable when treated while still in its superficial stage [6]. Therefore, its early detection is critically important; however, identifying superficial
BEA is often difficult. In Western countries, the Seattle protocol for BE surveillance
recommends that four-quadrant biopsy specimens be acquired at intervals of 1 to 2 cm
[7]. However, random biopsies can lead to numerous sampling errors, and only 4 % to
5 % of BEs are discovered using this method [8]
[9]
[10]. Owing to such limitations, various endoscopic imaging techniques such as narrow
band imaging (NBI), acetic acid chromoendoscopy (AAC), and endoscopy-based confocal
laser endomicroscopy were developed to improve the diagnosis of superficial BEA [11]; among these, NBI has been the most widely researched.
To diagnose neoplastic lesions in patients with BE, several groups proposed classifications
based on mucosal and vascular patterns visualized using magnifying endoscopy with
NBI (M-NBI); however, these classifications did not markedly improve lesion detection
owing to the complexity and diversification of vascular patterns [12]
[13]
[14]
[15]. To address these issues, the Japan Esophageal Society Barrett’s esophagus (JES-BE)
working group proposed a simpler M-NBI classification [16] that is based on representative early gastric cancer diagnostic criteria [17]. Although a study to validate this method showed promising diagnostic accuracy and
interobserver agreement, it was conducted only by endoscopists at high-volume academic
centers [18].
In contrast, a recent meta-analysis revealed that AAC was effective [19]. Acetic acid causes reversible acetylation of BE nucleoproteins and vascular congestion,
which leads to highlighting microstructural patterns. However, there have been no
dedicated studies on the usefulness of M-NBI that incorporates acetic acid spraying
(M-AANBI). In addition, we are aware of no study that directly compared the diagnostic
accuracy of M-AANBI with that of M-NBI. Hence, this study aimed to determine the ability
of M-AANBI to detect BEA and compare it to that of M-NBI when used by either expert
or non-expert endoscopists.
Patients and methods
Patients
This retrospective study included consecutive patients with BEA who underwent endoscopic
submucosal dissection (ESD) between July 2005 and November 2021 at Mie University
Hospital. Operable patients with a preoperative diagnosis of early-stage BEA were
eligible for ESD. During the study period, 34 BEA lesions from 33 patients were evaluated
with preoperative M-NBI/M-AANBI. Of these, four samples lacking images of either M-NBI
or M-AANBI and one with low-quality endoscopic images were excluded; hence, 29 BEA
lesions from 28 patients were retrospectively analyzed. The study was approved by
the local ethics committee and was conducted according to the ethical standards laid
out in the Declaration of Helsinki. The requirement for written informed consent was
waived owing to the retrospective nature of the study. Instead, detailed information
about the study was available to the public on our institutional website, and patients
were offered the opportunity to opt out.
Definitions of BE, BEA, and non-neoplastic BE (NNBE)
BE was defined as an esophagus in which any portion of the normal distal squamous
epithelial lining was replaced by metaplastic columnar epithelium as clearly visible
endoscopically ≥ 1 cm above the esophagogastric junction and confirmed histopathologically
[20]. The esophagogastric junction was defined as the end of the lower esophageal palisade
vessels or upper limit of the gastric fold [21]
[22].
BEA was defined as adenocarcinoma that arose from BE as verified endoscopically and
histologically, including non-invasive well-differentiated adenocarcinoma (high grade)
that is equivalent to high-grade dysplasia in Western countries. NNBE was defined
as the mucosa adjacent to the BEA that was resected during the ESD procedure.
Evaluation of endoscopic findings
All endoscopic images were obtained using a magnifying endoscope (GIF-Q240Z, GIF-H260Z,
or GIF-H290Z; Olympus Corp., Tokyo, Japan) and an endoscopic system with NBI (EVIS
LUCERA ELITE or EVIS LUCERA SPECTRUM; Olympus Corp.). A distal attachment (D-201-11804,
D-201-11804, MAJ-1989, or MAJ-1990; Olympus Corp.) was placed on the tip of the endoscope
to maintain a suitable focusing distance during magnification.
Tumor size, surface color, and macroscopic type were evaluated using white-light endoscopy.
M-NBI was performed with optimal foci to evaluate mucosal and surface patterns according
to the JES-BE classification [16]. Subsequently, 1.5 % acetic acid was sprayed onto the lesion with a 20-mL syringe
at low pressure, and M-AANBI images were obtained and evaluated as detailed below.
Non-magnifying NBI was performed before each M-NBI or M-AANBI session; alternating
between magnified and non-magnified images helped identify the lesion section that
was assessed.
Histopathological evaluation
All BEA lesions were resected using the ESD procedure. Each resected specimen was
cut into 2 mm slices after formalin fixation; the histological type, size, depth of
invasion, and margins (horizontal/vertical) were then evaluated. The pathological
diagnosis was performed based on hematoxylin-eosin staining by two expert pathologists,
who were blinded to endoscopic findings, according to the Japanese Classification
of Esophageal Cancer [22]. The depth of tumor invasion was recorded as the T category; T1a and T1b were defined
as tumors confined to the mucosa and submucosa, respectively. The extent of BEA was
pathologically evaluated and compared to the endoscopic demarcation line (DL), which
was confirmed by comparison to the pathological BEA border.
Derivation study for developing BEA/NNBE classifications (phase 1)
In phase 1, two endoscopists (Y.I. and K.T.) evaluated 60 high-quality M-AANBI images
of BEA and NNBE based on the AAC classification for gastric cancer [23]. Disagreements between the raters were resolved through discussion. The images were
independent of those used for validation.
The characteristic surface patterns of BEA/NNBE were classified into five types as
follows: type I, small round pits of uniform size and shape ([Fig. 1a]); type II, slit-like pits ([Fig. 1b]); type III, gyrus and villous patterns ([Fig. 1c, d]); type IV, irregular arrangement and size ([Fig. 1e]); and type V, destructive pattern ([Fig. 1f]).
Fig. 1 Classifications of BEA/NNBE surface patterns using M-AANBI. The surface patterns of
BEA and NNBE obtained using M-AANBI were classified into five types. a Small round pits of uniform size and shape (type I); b Slit-like pits (type II); c Gyrus pattern (type III); d, Villous pattern (type III); e, f Irregular arrangement and size (type IV); g, h Destructive pattern (type V); The characteristic surface patterns of NNBE were types
I-III, while those of BEA were types IV and V. BEA, Barrett’s esophageal adenocarcinoma;
NNBE, non-neoplastic Barrett’s esophagus; M-AANBI, magnifying endoscopy with narrow
band imaging plus acetic acid spraying.
Validation studies (phase 2)
In phase 2, we conducted three validation studies to evaluate the diagnostic performance
of M-AANBI. The images were examined by four expert endoscopists (Y.U., H.Y., Y.H.,
and M.K.) and four non-expert endoscopists (A.H., S.S., I.A., and W.Y.). We defined
experts as operators who had experience with more than 100 M-NBI and M-AANBI procedures
and non-experts as those who had experience with fewer than 20 such procedures. Non-expert
endoscopists were provided a short explanation of M-AANBI-based diagnosis of BEA/NNBE
(according to our developed classification) prior to their participation.
Validation 1 (main validation): Comparing the diagnostic performances of M-NBI and
M-AANBI
An independent validation dataset was prepared to compare the accuracies of M-NBI
and M-AANBI when diagnosing BEA. Based on the sample size (described below), the validation
dataset included 130 each of M-NBI- and M-AANBI-acquired images; 65 BEA and 65 NNBE
images were obtained with each of these two modalities. All images were randomized
and verified, and each endoscopist then made a diagnosis of BEA or NNBE while assigning
a confidence level (high/low) for each image. Agreement among the raters was quantified
using Fleiss’ kappa coefficient.
Validation 2: Changes in diagnostic performance owing to combining M-AANBI and M-NBI
To evaluate the effect of combining M-AANBI with M-NBI on diagnosis, 56 pairs of M-NBI
and M-AANBI images acquired from the same lesion areas (28 BEA and 28 NNBE) were prepared.
First, endoscopists performed their diagnoses using only M-NBI images. Subsequently,
they re-diagnosed the samples using both M-NBI and M-AANBI; the difference in diagnostic
performance between these two methods was then evaluated ([Fig. 2]). Agreement among the raters was quantified using Fleiss’ kappa coefficient.
Fig. 2 Evaluating changes in diagnostic performance when using both M-AANBI and M-NBI. The
first evaluation was performed using M-NBI only: a second evaluation was then performed
using both M-NBI and M-AANBI. Abbreviations: M-AANBI, magnifying endoscopy with narrow
band imaging plus acetic acid spraying; M-NBI, magnifying endoscopy with narrow band
imaging.
Validation 3: Evaluating the ease of recognizing the DL using three methods
The DL was defined as the endoscopic border between the background non-cancerous and
cancerous mucosae [24]
[25]
[26]; this has been confirmed to be consistent with pathological findings. DL recognition
in the BEA samples of 28 patients was evaluated using three different methods: magnifying
endoscopy with white-light imaging (M-WLI), M-NBI, and M-AANBI. For each lesion, three
images of the same DL area were arranged on one slide for examination ([Fig. 3]). Each endoscopist was instructed to score the ease of DL recognition when using each
of these methods according to a visual analog scale (VAS), which was graded from 0
to 10 wherein 0 represented “invisible” and 10 represented “perfect visibility”.
Fig. 3 Magnifying endoscopic images of the DL of BEA (same area). Abbreviations: DL, demarcation
line; BEA, Barrett’s esophageal adenocarcinoma; M-NBI, magnifying endoscopy with narrow
band imaging; M-WLI, magnifying endoscopy with white-light imaging; M-AANBI, magnifying
endoscopy with narrow band imaging plus acetic acid spraying; “Marked,” marked image
of the DL superimposed on the magnifying endoscopic image.
Sample size determination and statistical analysis
A pilot study was conducted to determine the appropriate sample size. Four expert
endoscopists (Y.U., H.Y., Y.H., and M.K.) and four non-expert endoscopists (A.H.,
S.S., I.A., and W.Y.) evaluated 40 M-NBI and 40 M-AANBI images; 20 BEA and 20 NNBE
images were examined using each modality. The mean diagnostic accuracies were 72.5 %
for M-NBI and 85.6 % for M-AANBI. According to McNemar’s test, 130 pairs of M-NBI
and M-AANBI images were required to achieve a power of > 90 % (assuming a two-sided
alpha of 0.05) based on eight diagnostic accuracy ratings from as many endoscopists.
Categorical variables are summarized as frequencies and percentages, while quantitative
variables are presented as means and standard deviations. The performance of BEA diagnosis
was determined by calculating the sensitivity, specificity, and accuracy for each
endoscopist and the overall group. The interobserver agreement among endoscopists
was calculated using Fleiss’ kappa, with the strength of each agreement graded using
the kappa value (< 0.20 = poor, 0.21–0.40 = fair, 0.41–0.60 = moderate, 0.61–0.80 = substantial,
and 0.81–1.00 = almost perfect) [27]. Fisher’s exact test was used to compare M-AANBI findings. The differences in the
diagnostic test results between M-NBI and M-AANBI were analyzed using the McNemar’s
test. The average scores reflecting the ease of DL recognition using the three methods
were compared using the Friedman test and Bonferroni’s multiple comparison test. P < 0.05 was considered indicative of statistical significance in all tests. All calculations
were performed using EZR version 1.27 (Saitama Medical Center, Jichi Medical University,
Japan) [28].
Results
Patient characteristics
The clinical characteristics of patients with BEA and of their lesions are shown in
[Table 1]. The mean patient age was 73.6 years (range, 50–92 years), and 79.2 % were men.
Per the Prague criteria, the mean circumferential and maximal extents of BE were 2.2 cm
and 3.4 cm, respectively. The mean tumor size was 20.5 mm; the most frequent macroscopic
type was depressed (51.7 %), and 65.5 % of lesions had a reddish color. The dominant
histopathology in most cases was well-differentiated adenocarcinoma (86.2 %), and
tumor invasion in most cases was confined to the mucosa (79.3 %).
Table 1
Clinicopathological characteristics of patients with BEA.
|
Patient characteristics (n = 28)
|
|
Age, mean ± SD, years (range)
|
73.6 ± 9.6 (50–92)
|
|
Sex, male/female, n
|
24/4
|
|
Circumferential length of BE, mean cm (range)
|
2.2 ± 3.1 (0–15)
|
|
Maximal length of BE, mean cm (range)
|
3.4 ± 2.9 (1–15)
|
|
Lesion characteristics (n = 29)
|
|
Size
|
|
|
|
20.5 ± 12.3 (6–53)
|
|
|
16 (55.2)
|
|
|
13 (44.8)
|
|
Macroscopic type, n (%)
|
|
|
|
3 (10.3)
|
|
|
7 (24.1)
|
|
|
4 (13.8)
|
|
|
15 (51.7)
|
|
Color, n (%)
|
|
|
|
10 (34.5)
|
|
|
19 (65.5)
|
|
Dominant histopathology, n (%)
|
|
|
|
25 (86.2)
|
|
|
3 (10.3)
|
|
|
1 (3.4)
|
|
Depth of invasion
|
|
|
|
23 (79.3)
|
|
|
6 (20.7)
|
BEA, Barrett’s esophageal adenocarcinoma; SD, standard deviation; BE, Barrett’s esophagus;
T1a, tumor confined to the mucosa; T1b, submucosal invasion.
Derivation study (phase 1)
The assessment of surface patterns on BEA/NNBE using M-AANBI is shown in [Table 2]. The characteristic surface patterns of NNBE were wholly type I, type II, or type
III. In contrast, almost all BEA images had two characteristic surface patterns (type
IV or type V [P < 0.001]). Therefore, type IV and V surface patterns observed on M-AANBI were deemed
as predictive of BEA.
Table 2
Distribution of surface patterns in BEA and NNBE samples.
|
Surface patterns of M-AANBI
|
BEA
|
NNBE
|
P value[1]
|
|
Types I or II or III, n (type I/type II/type III)
|
1 (0/0/1)
|
30 (4/11/15)
|
< 0.001
|
|
Type IV or V, n (type IV/type V)
|
29 (26/3)
|
0 (0/0)
|
BEA, Barrett’s esophageal adenocarcinoma; NNBE, non-neoplastic Barrett’s esophagus.
1 Fisher’s exact test.
Validation studies (phase 2)
Validation 1 (main validation): Comparing the diagnostic performances of M-NBI and
M-AANBI
[Table 3] shows the diagnostic performance, interobserver agreement (κ value), and confidence
levels for predicting BEA according to the validation 1 scheme. For M-NBI, the mean
accuracy, sensitivity, and specificity among all endoscopists were 65.8 %, 61.7 %,
and 69.8 %, respectively; in comparison, the values for M-AANBI were 91.1 %, 86.3 %,
and 95.2 %, respectively. All parameters were significantly higher when using M-AANBI
than with M-NBI among all eight endoscopists (P < 0.05).
Table 3
Diagnostic performance of M-NBI and M-AANBI in terms of detecting BEA.
|
Raters
|
Modalities
|
Accuracy,
% (95 % CI)
|
Sensitivity,
% (95 % CI)
|
Specificity,
% (95 % CI)
|
IO agreement
K value
(95 %CI)
|
High confidence rate,
% (95 % CI)
|
|
All
|
M-NBI
|
65.8
(61.1–70.5)
|
61.7
(50.3–73.1)
|
69.8
(58.9–80.7)
|
0.25
(0.21–0.28)
|
53.3
(41.0–65.5)
|
|
M-AANBI
|
91.1[1]
(85.7–96.4)
|
86.3[1]
(76.6–96.1)
|
95.2[1]
(91.0–99.4)
|
0.65
(0.61–0.68)
|
75.2[1]
(64.9–85.5)
|
|
Experts
|
M-NBI
|
70.0
(66.1–73.9)
|
64.6
(38.6–90.7)
|
75.4
(55.1–95.7)
|
0.38
(0.31–0.45)
|
61.2
(34.9–87.4)
|
|
M-AANBI
|
91.3[1]
(78.3–104.4)
|
85.0[1]
(57.8–112.2)
|
97.7[1]
(95.2–100.1 %)
|
0.64
(0.56–0.72)
|
81.7[1]
(58.4–100.5)
|
|
Non-experts
|
M-NBI
|
61.5
(54.4–68.7)
|
58.5
(39.7–78.0)
|
64.2
(44.6–83.9)
|
0.33
(0.26–0.40)
|
45.4
(32.9–57.9)
|
|
M-AANBI
|
90.8[1]
(82.2–99.3)
|
87.7[2]
(81.1–94.3)
|
92.7[1]
(82.6–102.8)
|
0.66
(0.59–0.73)
|
68.7
|
BEA, Barrett’s esophageal adenocarcinoma; M-NBI, magnifying narrow band imaging; M-AANBI,
magnifying narrow band imaging plus acetic acid spraying; CI, confidence interval;
IO, interobserver.
1
P < 0.001, vs. M-NBI using McNemar’s test.
2
P < 0.01, vs. M-NBI using McNemar’s test.
The accuracy of M-NBI was significantly higher in the expert group than in the non-expert
group (70.0 % vs. 61.5 %, P < 0.05). However, both groups achieved high accuracy with M-AANBI, with no difference
between them.
The interobserver agreement among endoscopists was fair for M-NBI (κ = 0.25) and substantial
for M-AANBI (κ = 0.65). Moreover, the proportion of high confidence for M-AANBI was
greater than that for M-NBI (75.2 % vs. 53.3 %, P < 0.001); both the expert and non-expert groups had similar tendencies.
Validation 2: Changes in diagnostic performance owing to combining M-AANBI and M-NBI
The results of validation 2-related analyses are presented in [Table 4]. The mean accuracy, sensitivity, and specificity of BEA diagnosis using M-NBI alone
were 68.5 %, 64.6 %, and 73.2%, respectively. When using both M-NBI and M-AANBI, these
values improved to 87.3 %, 85.7 %, and 90.6 %, respectively; the accuracy was significantly
greater than that of M-NBI alone (P < 0.05). The interobserver agreement when using M-NBI alone was fair (κ = 0.31),
whereas that when combining M-NBI and M-AANBI was substantial (κ = 0.69). The high
confidence rate improved from 57.6 % to 76.3 % when using both modalities.
Table 4
Diagnostic performance parameters before/after adding M-AANBI to M-NBI.
|
Modalities
|
Accuracy
mean, %
(95 % CI)
|
Sensitivity
mean, %
(95 % CI)
|
Specificity
mean, %
(95 % CI)
|
IO agreement
K value
(95 %CI)
|
High confidence rate, %
(95 % CI)
|
|
M-NBI
|
68.5
(62.9–74.2)
|
64.6
(56.2–73.2)
|
73.2
(61.3–85.2)
|
0.31
(0.26–0.36)
|
57.6
(45.0–70.2)
|
|
M-NBI +
M-AANBI
|
87.3[1]
(82.7–91.9)
|
85.7
(77.9–93.5)
|
90.6
(85.6–95.7)
|
0.69
(0.64–0.74)
|
76.3[2]
(63.0–89.6)
|
M-NBI, magnifying endoscopy with narrow band imaging; M-AANBI, magnifying endoscopy
with narrow band imaging plus acetic acid spraying; CI, confidence interval; IO, interobserver.
1
P < 0.05, vs. M-NBI using McNemar’s test.
2
P < 0.01, vs. M-NBI using McNemar’s test.
Validation 3: Evaluating the ease of recognizing the DL using three methods
As shown in [Table 5], the mean VAS scores for the ease of DL recognition as evaluated by all endoscopists
were 3.60 ± 1.27 for M-WLI, 6.16 ± 1.36 for M-NBI, and 8.40 ± 0.76 for M-AANBI; the
differences between these scores (compared using the Friedman test) were significant
(P < 0.001). Using Bonferroni’s multiple comparison test, the mean VAS score of M-AANBI
was significantly higher than those of the other two methods, with M-NBI scoring significantly
higher than M-WLI. Similar trends were observed in both the expert and non-expert
groups (P < 0.001).
Table 5
Ease of demarcation line recognition when using M-WLI, M-NBI, and M-AANBI.
|
Modalities
|
All endoscopists
|
Experts
|
Non-experts
|
|
M-WLI,
mean VAS ± SD
|
3.60 ± 1.27
|
3.53 ± 1.28
|
3.68 ± 1.41
|
|
M-NBI,
mean VAS ± SD
|
6.16 ± 1.36[1]
|
6.16 ± 1.53[1]
|
6.16 ± 1.38[1]
|
|
M-AANBI,
mean VAS ± SD
|
8.40 ± 0.76[1]
[2]
|
8.41 ± 0.82[1]
[2]
|
8.40 ± 0.92[1]
[2]
|
VAS, visual analog scale; M-WLI, magnifying endoscopy with white-light imaging; M-NBI,
magnifying endoscopy with narrow band imaging; M-AANBI, magnifying endoscopy with
narrow band imaging plus acetic acid spraying; SD, standard deviation.
1
P < 0.001, vs. M-WLI using Friedman’s test.
2
P < 0.001, vs. M-NBI using Friedman’s test.
Discussion
This study was the first to demonstrate the usefulness of M-AANBI in BEA diagnosis
through direct comparison with M-NBI. The diagnostic accuracy, sensitivity, specificity,
interobserver agreement, and confidence level of M-AANBI were significantly higher
than those of M-NBI. Furthermore, combining M-NBI with M-AANBI produced an additive
effect in terms of improving diagnostic accuracy; M-AANBI also tended to facilitate
DL recognition.
To date, several NBI classifications for the diagnosis of BEA have been developed;
however, they have proven to be complicated with limited diagnostic abilities [12]
[13]
[14]
[15], rendering them inadequate for use in clinical practice. In 2016, the Barrett’s
International NBI Group (BING) constructed a simpler NBI classification for differentiating
between dysplasia and non-dysplasia and demonstrated high diagnostic performance for
the former (overall accuracy: 85.4 %, sensitivity: 80.4 %, and specificity: 88.4 %)
and substantial interobserver agreement (κ = 0.681) [29]. However, this validation was performed only by experts, and it remained unclear
whether a similar diagnostic performance could be achieved by non-experts.
Subsequently, the JES-BE classification that was based on the typical M-NBI diagnostic
criteria for early gastric cancer was proposed [16]
[17]. BEA is diagnosed by an irregular mucosal or vascular pattern; the criteria were
further modified to include the flat pattern originally regarded to be physiological.
The JES-BE classification has been reported to have a high diagnostic ability (overall
accuracy, 91 %; sensitivity, 87 %; and specificity, 97 %) and high interobserver agreement
(κ = 0.77); however, that study was single-armed and could not be compared with other
representative modalities such as M-AANBI [18]. Although the JES-BE classification was useful even for non-experts with little
experience in diagnosing BEA, they were engaged in high-volume or academic centers
and had substantial experience using magnification endoscopy for detecting early gastric
cancer. Given that the JES-BE classification is based on the criteria used for detecting
early gastric cancer, the existing familiarity of operators with magnification endoscopy
may have affected the results; hence, the versatility of this classification in clinical
practice remains unknown. With respect to this point, when we performed our tests
on non-experts, the accuracy of M-NBI for BEA diagnosis using the JES-BE classification
was significantly lower than it was among experts.
AAC has been reported to be a useful technique for BEA diagnosis [19]. Acetic acid changes the color of the BE epithelial surface to white, making it
easier to recognize microstructural patterns. AAC is easy, safe, inexpensive, and
effective, particularly for lesions with blood oozing on the surface; however, its
superiority with respect to M-AANBI remains unknown as this question was not the focus
of this study. Moreover, inflammatory changes can lead to incorrect BEA diagnoses
when using AAC because of the rapid loss of acetowhitening [30]. Despite limited studies on the usefulness of M-AANBI in diagnosing BEA [31], it is better suited to avoid such misdiagnoses.
We previously classified surface patterns when using AAC in the stomach into five
types, as described above [23]. When planning our current study, this classification was applied to differentiate
BEAs from NNBEs ([Fig. 1]); its use with M-AANBI showed a high diagnostic ability (overall accuracy, 91.1 %;
sensitivity, 86.3 %; and specificity, 95.2 %). All these parameters were superior
to those of M-NBI. Although we provided a short tutorial for the diagnostic classification
of M-AANBI to non-experts who had little experience with M-NBI and M-AANBI, the accuracy
of BEA diagnosis by non-experts was high when using M-AANBI (equivalent to that of
experts). Furthermore, the diagnostic performance and certainty were significantly
improved by performing both M-AANBI and M-NBI. M-NBI-based diagnosis normally requires
observation at high magnification to evaluate not only the structure but also the
vascular pattern. However, since M-AANBI only examines structural patterns, it is
possible to establish diagnoses with low magnification that emulates non-magnification;
hence, this method is easy to use for endoscopists accustomed only to NBI in Western
countries. Furthermore, M-AANBI may be utilized without magnification when using high-resolution
endoscopes, which are likely to become more common in the future. Notably, the interobserver
agreement was fair for M-NBI (κ = 0.25) whereas the value for M-AANBI was substantial
(κ = 0.65). This indicates that M-NBI is subjective and difficult to interpret, while
M-AANBI is an objective diagnostic method for all endoscopists including non-experts.
Identifying the DL between BEA and NNBE is generally considered difficult [32]. Since BE exhibits heterogeneous mucosal patterns of intestinal metaplasia, false
DLs can be designated in NNBEs; this is also one of the reasons it is difficult to
recognize the precise DL between BEA and NNBE. The ease of DL recognition using M-WLI
was also low in the current study, whereas M-AANBI recognized the DL much more readily
than M-NBI and M-WLI ([Table 5]).
Artificial intelligence (AI) has recently allowed for remarkable progress in image
recognition of gastrointestinal lesions [33]. A Previous study found that AI-based techniques are highly reliable for diagnosing
gastric cancer using M-NBI [34]; furthermore, Ling et al. reported a unique AI system for identifying the DL in
gastric cancer [35]. As such, AI may also be useful for the qualitative diagnosis of BEA and identification
of the DL using M-AANBI.
This study had certain limitations. First, it was a retrospective study that used
selected endoscopic images, and selection bias toward high-quality images existed.
However, considering that physicians can observe all lesions in real-time in clinical
practice, using high-quality images ought to be acceptable. Second, since this was
a detailed comparative study of M-AANBI and M-NBI, validation was only performed on
a cohort from our own institution. Third, the results were based on a small sample
size, and several images were extracted from the same lesion. Despite our best efforts
to avoid it, some images may have been of overlapping sections of a given lesion.
Hence, more data derived from a larger number of cases (including from multicenter
prospective studies) are required to improve the diagnostic yield of such endoscopic
findings. Fourth, the validation 3 scheme, which measured the confidence of recognizing
the DL with each of three modalities, may have a subjective aspect. However, an endoscopic
diagnosis is essentially a series of subjective evaluations; furthermore, the evaluators
were blinded to the greatest extent possible during testing. As such, subjective bias
ought to have been minimized. Fifth, M-AANBI is not a standard practice technique
at present, with concerns that the essential system with magnifying endoscope for
M-NBI is more expensive than the conventional endoscopy system. Furthermore, the M-AANBI
technique may be complicated for endoscopists who are not familiar with the use of
magnifying endoscopes. However, the NBI instruments have increased in popularity and
acetic acid is not expensive due to its use as a threefold vinegar dilution. Moreover,
similar to M-AANBI diagnosis, a high-resolution non-magnifying endoscope with a combination
of AAC and NBI may be sufficient for diagnosis of BEA. Sixth, this study evaluated
images of exposed BEA areas; however, certain BEA areas can sometimes be covered with
normal squamous epithelium in patients treated with proton pump inhibitors. Therefore,
the usefulness of M-AANBI for qualitative diagnosis and DL detection in areas covered
with squamous epithelium is unknown.
Conclusions
In conclusion, we demonstrated the usefulness of M-AANBI in diagnosing BEA and its
superiority to M-NBI. Although the diagnostic ability of M-NBI was low among non-experts,
that of M-AANBI was high, with good interobserver agreement among all endoscopists
and non-experts alike. The results of this study suggest that M-AANBI is useful for
BEA diagnosis by endoscopists of all experience levels, including non-experts who
have little experience with magnifying endoscopy. In clinical practice, the use of
M-AANBI in addition to M-NBI might be beneficial for BEA detection during surveillance
endoscopy for BE and for DL recognition during ESD for BEA.
