Abstract
Objectives:
Detailed superiority of CAD EYE (Fujifilm, Tokyo, Japan), an artificial intelligence for polyp detection/diagnosis, compared to endoscopists is not well examined. We examined endoscopist's ability using movie sets of colorectal lesions which were detected and diagnosed by CAD EYE accurately.
Methods:
Consecutive lesions of ≤10 mm were examined live by CAD EYE from March-June 2022 in our institution. Short unique movie sets of each lesion with and without CAD EYE were recorded simultaneously using two recorders for detection under white light imaging (WLI) and linked color imaging (LCI) and diagnosis under blue laser/light imaging (BLI). Excluding inappropriate movies, 100 lesions detected and diagnosed with CAD EYE accurately were evaluated. Movies without CAD EYE were evaluated first by three trainees and three experts. Subsequently, movies with CAD EYE were examined. The rates of accurate detection and diagnosis were evaluated for both movie sets.
Results:
Among 100 lesions (mean size: 4.7±2.6 mm; 67 neoplastic/33 hyperplastic), mean accurate detection rates of movies without or with CAD EYE were 78.7%/96.7% under WLI (p<0.01) and 91.3%/97.3% under LCI (p<0.01) for trainees and 85.3%/99.0% under WLI (p<0.01) and 92.6%/99.3% under LCI (p<0.01) for experts. Mean accurate diagnosis rates of movies without or with CAD EYE for BLI were 85.3%/100% for trainees (p<0.01) and 92.3%/100% for experts (p<0.01), respectively. The significant risk factors of not-detected lesions for trainees were right-sided, hyperplastic, not-reddish, in the corner, halation, and inadequate bowel preparation.
Conclusions:
Unique movie sets with and without CAD EYE could suggest it's efficacy for lesion detection/diagnosis.
Keywords: artificial intelligence, LCI, BLI, CAD EYE, polyp
Introduction
Colonoscopy and polyp resection by it enable to reduce the colorectal cancer (CRC) death[1,2]. Thus, polyp resection is performed worldwide for preventing CRC incidence and death[3-5]. However, a recent systematic review revealed adenoma miss rates of 26% (95% confidence interval (CI): 23-30%)[6]. For improving polyp detection, numerous reports have discussed the improvement of lesion visibility and detection under various endoscopic tools including linked color imaging (LCI) and narrow band imaging (NBI)[7-11]. Regarding polyp characterization, blue laser/light imaging (BLI) and NBI is useful for achieving accurate histopathologic prediction[11,12]. However, polyp detection and characterization remain difficult, especially for non-experts.
Regarding colonoscopic field, both lesion detection and diagnosis using artificial intelligence (AI) have been examined, and some of them have been marketed[13-16]. A systematic review of five randomized control trials showed that the adenoma detection rate (ADR) of endoscopists with AI was significantly higher than that of those without AI (36.6% vs. 25.2%, p<0.01)[17]. CAD EYE (Fujifilm, Tokyo, Japan), an endoscopic AI system available in Japan and Europe since 2020, can help endoscopists to detect colorectal polyps under white light imaging (WLI) and LCI. It is also available for colorectal polyp's diagnosis under BLI, with or without magnification.
Several studies have shown the efficacy of lesion detection and diagnosis in CAD EYE using recorded images and videos[18-20]. Recently, some clinical trials also showed the efficacy of CAD EYE for in real cases[21-23]. However, detailed superiority of CAD EYE compared to endoscopists is still unknown. In this study, we examined endoscopist's ability using movie sets of colorectal lesions which were detected and diagnosed by CAD EYE accurately. What kind of lesions are helped by CAD EYE is also examined.
Methods
This was a single-center retrospective study using movies recorded prospectively. We included 130 consecutive lesions of ≤10 mm, which were examined live by CAD EYE in March-June 2022. Either LASER endoscope or LED endoscope made by Fujifilm was used. Short movies of 5-10 seconds for each mode of WLI/LCI for polyp detection and with magnified BLI for polyp diagnosis for each lesion were recorded (Figure 1). The movies of WLI and LCI were recorded under same conditions about angle, insufflation, and withdrawal speed. Additionally, for each mode of WLI/LCI/BLI, two sets of unique movies with and without CAD EYE were recorded simultaneously using two recorders connected to one endoscopic system. Thus, a total of six short movies were prepared for each lesion. Among the 130 lesions, 13 lesions not detected by CAD EYE with either WLI or LCI and eight lesions not diagnosed by CAD EYE with magnified BLI accurately were excluded (Figure 1). Additionally, six lesions with different conditions of WLI and LCI and three lesions with two or more lesions were excluded. Finally, 100 lesions in 29 patients detected by CAD EYE under WLI and LCI and diagnosed with BLI accurately were evaluated in this study. Besides these movies with lesions, additional eight movies without lesions were prepared as a negative control, including two movies for both WLI and LCI of no lesions, with and without CAD EYE.
Figure 1.
Study flow.
Inclusion criteria were as follows: patients undergoing (1) follow-up of polyps; (2) surveillance after polypectomy or surgery for polyps or CRC; (3) abdominal symptoms; (4) positive fecal occult blood; and (5) CRC screening. All polyps, including recurrent cases, were analyzed. However, for hyperplastic polyps in the sigmoid colon and rectum, only a maximum of two polyps per patient were included. We excluded urgent colonoscopy with severe hematochezia and colonoscopy without bowel preparation by lavage. Patients were enrolled to the study for four months. However, the colonoscopic examination with two recorders could be held only on Monday (once a week) because of an expert's working schedule (N.Y. and K.I.) and a system setting. All colonoscopy procedures were performed by these two experts and three to five patients were enrolled every week. All polyps were histopathologically examined by either endoscopic resection or biopsy.
Regarding the study outcomes, all movies without CAD EYE including negative control (102 movies for WLI, 102 movies LCI, and 100 movies for BLI) were evaluated in the test phase by three trainees and three experts (Figure 1). Then, movies with CAD EYE (102 for WLI, 102 for LCI, and 100 for BLI) were evaluated for the confirmation phase. The confirmation phase was performed just after the test phase to confirm movie with the accurate detection and diagnosis by CAD EYE. In the confirmation phase, we examined whether evaluator could detect and diagnose a lesion accurately, supported by the detection and diagnosis by CAD EYE. Each movie could be seen only once both in the test phase and confirmation phase. We examined the accurate rates for detection about WLI and LCI and the accurate rates for diagnosis about BLI in the both test phase and confirmation phase. We also analyzed the various clinical characteristics of both lesions not detected by endoscopists (not-detected lesions) and lesions not diagnosed by endoscopists (not-diagnosed lesions). We defined not-detected lesions or not-diagnosed lesions that two to three trainees or two to three experts could not detect with WLI or diagnosed with BLI. Regarding lesion characteristics, lesion size, morphology, histopathology, color, monitor location, brightness, and bowel preparation were examined for not-detected lesions. Lesion size, morphology, histopathology, color, and bowel preparation were examined for not-diagnosed lesions. Two expert doctors evaluated the existence of these endoscopic factors for all 100 lesions (Figure 2). Regarding monitor location, the monitor was divided into nine pieces, and polyps away from the center during the entire movie were defined as lesions in the corner (Figure 3). Bowel preparation was performed according to the Aronchick Bowel Preparation Scale, and a score of 0/1 was defined as inadequate bowel preparation[24]. Regarding brightness, halation was defined as a situation in which a whitish and shiny area due to endoscopic lights was seen in more than one-third of the monitor (Figure 3). For lesion diagnosis with BLI, we performed the analyses of inter-observer agreement in the test phase for both three trainees (trainee A, B, C) and three experts (expert A, B, C) to examine the degree of diagnostic agreement by calculating the kappa coefficient.
Figure 2.
The evaluation of movies of WLI, LCI and BLI with or without CAD EYE.
2a. A polypoid lesion, 2 mm, neoplastic (adenoma), cecum. WLI without CAD EYE. 2b. WLI with CAD EYE. CAD EYE enabled to detect the lesion accurately with annotation box. A false positive detection by CAD EYE was also seen on a left side of the monitor. 2c. LCI without CAD EYE. 2d. LCI with CAD EYE. CAD EYE enabled to detect the lesion accurately with annotation box. 2e. BLI without CAD EYE. 2f. BLI with CAD EYE. CAD EYE enabled to diagnose the lesion accurately with the sign “neoplastic” on the bottom of the monitor. Yellow-color sign was seen in the monitor.
Figure 3.
The factors examined for risk factors of not-detected polyps.
We divided the morphology into non-polypoid and polypoid lesions according to the Paris classification[25]. Polyp location was divided into two parts: right-sided colon (cecum to descending colon) and left-sided colon (sigmoid colon to rectum). Expert endoscopists were defined as those who had performed ≥5,000 colonoscopies with LCI and BLI observation according to a previous publication[11]. Trainees were defined as those who had performed less than 500 colonoscopies including LCI and BLI observations. They received a 30-minute lecture on LCI and BLI observations. For histopathological evaluation, the WHO classification was used. CAD EYE has two diagnostic categories, “neoplastic” and “hyperplastic”. Thus, sessile serrated lesions were defined as hyperplastic lesions in this study[26].
Regarding equipment, LASER light source (LL-7000; Fujifilm), LED light source (BL-7000; Fujifilm), a endoscopic system processor (VP-7000; Fujifilm) were used. EC-760ZP as an LED endoscope, EC-L600ZP7 (Fujifilm), and EC-L600ZP (Fujifilm) as LASER endoscopes were used. The setting of the LASER endoscopes were as follows: WLI, H/+4/+4; BLI (LASER), B8/C2; and LCI, B8/C2. Those of the LED endoscope were as follows: WLI, H/+4/+4; BLI (LED), B8/C2; LCI, B8/C3. LED or LASER endoscopes was generally used for each case sequentially.
CAD EYE
CAD EYE (EX-1, Fujifilm, Tokyo, Japan) is an artificial intelligence to assist the endoscopist in the diagnosis and detection of a lesion. Its shape is a box-like device similar to a light source device[20]. For CADx, CAD EYE diagnoses the two categories, “neoplastic” or “hyperplastic” on BLI. These signs are displayed at the bottom of the monitor when a lesion is detected by CAD EYE. Yellow or green curved lines indicating “neoplastic” and “hyperplastic” can be seen around the endoscopic monitor in the lower right window to indicate that CAD EYE has detected a lesion. For CADe, a detected lesion is surrounded by the light blue annotation box when CAD EYE detects it under WLI or LCI[20].
This study was conducted in accordance with the World Medical Association's Declaration of Helsinki. This study was approved by the Ethics Committee of Kyoto Prefectural University of Medicine (ERB-C-1319). This study was an observational study, and opt-out was performed after IRB approval.
Statistical analyses
Sample size was not estimated in this study. In cases meeting the inclusion criteria, all lesions detected by either LASER or LED endoscopy for one month were included. Chi-squared test with Yates continuity correction and Mann-Whitney U test were used for categorical and continuous values, respectively. SPSS software (version 22.0; IBM Japan, Ltd., Tokyo, Japan) was used for all statistical analyses. Statistical significance was set at p<0.05.
Results
A total of 100 polyps (mean size: 4.7±2.6 mm, range: 2-10 mm) in 29 patients were analyzed, including 48 right-sided and 52 left-sided. Histopathological examination revealed 67 neoplastic lesions and 33 hyperplastic lesions (Table 1).
Table 1.
Clinicopathological Features of All Lesions.
| Lesion number | 100 |
|---|---|
| Patient number | 29 |
| Age, mean±SD (range) | 72.6±6.1 (58-81) |
| Sex, % (n) male/female | 62.1/37.9 (18/11) |
| Tumor size, mm, mean±SD (range) | 4.7±2.6 (2-10) |
| The rate of lesions ≤5mm, % (n) | 72.0 (72) |
| Way of BLI Observation, % (n) Laser/LED | 49.0/51.0 (49/51) |
| Location, % (n), right-sided/left-sided | 48.0/52.0 (48/52) |
| Morphology, % (n), polypoid/non-polypoid | 64.0/36.0 (64/36) |
| Histopathology, % (n), HP/SSL/adenoma/HGD/others | 23.0/9.0/60.0/7.0/1.0 (23/9/60/7/1) |
| Neoplastic/hyperplastic, % (n) | 67.0/33.0 (67/33) |
SD: standard deviation, BLI: blue laser/light imaging, right-sided: cecum-descending colon, left-sided: sigmoid colon-rectum, HP: hyperplastic polyp, SSL: sessile serrated lesions, HGD: high-grade dysplasia
There were significant differences about the accurate rates of detection under WLI between without and with CAD EYE for trainees (78.7% and 96.7%, p<0.01) and experts (85.3% and 99.0%, p<0.01) (Table 2). Additionally, there were significant differences between the rates of LCI without and with CAD EYE for trainees (91.3% and 97.3%, p<0.01) and experts (92.6% and 99.3%, p<0.01).
Table 2.
Ability of CADe to Detect Polyps with WLI and LCI.
| Mean detection rate 3 Trainees |
Mean detection rate 3 Experts |
P value Trainee vs. Expert |
|
|---|---|---|---|
| WLI: without CAD EYE, % (n) | 78.7 (236/300) | 85.3 (256/300) | 0.04 |
| WLI: with CAD EYE, % (n) | 96.7 (290/300) | 99.0 (297/300) | 0.09 |
| P value Endoscopist vs. CAD EYE | 0.01 | <0.01 | |
| LCI: without CAD EYE, % (n) | 91.3 (274/300) | 92.6 (278/300) | 0.54 |
| LCI: with CAD EYE, % (n) | 97.3 (292/300) | 99.3 (298/300) | 0.10 |
| P value Endoscopist vs. CAD EYE | <0.01 | <0.01 |
WLI: white light imaging, LCI: linked color imaging
Mean accurate rates of diagnosis for three trainees and three experts for BLI without CAD EYE were 85.3% and 92.3%, respectively (p<0.01) (Table 3). The kappa values were 0.41 for trainee A and B, 0.48 for trainee B and C, and 0.50 for trainee C and A. Those were 0.63 for expert A and B, 0.62 for expert B and C, and 0.65 for experts C and A. The rates of BLI with CAD EYE were 100.0% and 100.0%, respectively (p=1.0). There were statistically significant differences between the accuracy rate without CAD EYE and the rate with CAD EYE for trainees (85.3% and 100.0%, p<0.01) and experts (92.3% and 100.0%, p<0.01), respectively.
Table 3.
The Evaluation of Movies of BLI with or without CAD EYE by Endoscopists.
| Mean accuracy rate 3 Trainees |
Mean accuracy rate 3 Experts |
P value Trainee vs. Expert |
|
|---|---|---|---|
| BLI: without CAD EYE, % (n) | 85.3 (256/300) | 92.3 (277/300) | <0.01 |
| BLI: with CAD EYE, % (n) | 100 (300/300) | 100 (300/300) | 1.0 |
| P value Endoscopist vs. CAD EYE | <0.01 | <0.01 |
BLI: blue laser/light imaging
Regarding the characteristics of not-detected lesions, statistically significant factors for trainees were right-sided (p<0.01), hyperplastic (p=0.01), not reddish (p<0.01), in the corner (p<0.01), halation (p=0.01), and inadequate bowel preparation (p<0.01) (Table 4). Those for experts were smaller (p=0.02), right-sided (p<0.01), non-polypoid (p=0.04), not reddish (p<0.01), and in the corner (p<0.01).
Table 4.
The Characteristics of Not Detected Lesions under WLI for Trainees and Experts.
| Trainees Not-detected |
Trainees Detected |
P value | Experts Not-detected |
Experts Detected |
P value | |
|---|---|---|---|---|---|---|
| Lesion number, n (%) | 66 (22.0) | 234 (78.0) | - | 42 (14.0) | 298 (86.0) | - |
| Tumor size, mm, mean±SD (range) | 4.1±2.6 (2-10) | 4.9±2.6 (2-10) | 0.15 | 3.0±2.6 (2-8) | 5.0±2.6 (2-10) | 0.02 |
| Location, % (n), right-sided/left-sided | 63.6/36.4 (42/24) | 43.6/56.4 (102/132) | <0.01 | 71.4/28.6 (30/12) | 44.1/55.9 (114/144) | <0.01 |
| Morphology, % (n), polypoid/non-polypoid | 54.5/45.5 (36/30) | 66.7/33.3 (156/78) | 0.07 | 50.0/50.0 (21/21) | 66.3/33.7 (171/87) | 0.04 |
| Neoplastic/hyperplastic, % (n) | 45.5/54.5 (30/36) | 73.1/26.9 (171/63) | <0.01 | 64.3/33.3 (27/15) | 67.4/32.6 (174/84) | 0.68 |
| Color, % (n), reddish/ not reddish | 4.5/95.5 (3/63) | 39.7/60.3 (183/141) | <0.01 | 7.1/92.9 (3/39) | 36.0/64.0 (93/165) | <0.01 |
| Monitor location, % (n), In the center/ in the corners | 59.1/40.9 (39/27) | 76.9/23.1 (180/54) | <0.01 | 28.6/71.4 (12/30) | 80.2/29.8 (207/51) | <0.01 |
| Brightness, % (n), Halation/ not halation | 31.8/68.2 (21/45) | 10.3/89.7 (24/210) | 0.01 | 21.4/78.6 (9/33) | 14.0/86.0 (36/222) | 0.20 |
| Bowel preparation, % (n), Adequate/inadequate | 59.1/40.9 (39/27) | 88.5/11.5 (207/27) | <0.01 | 71.4/28.6 (30/12) | 83.7/16.3 (216/42) | 0.05 |
SD: standard deviation, right-sided: cecum-descending colon, left-sided: sigmoid colon-rectum
Regarding the characteristics of not-diagnosed lesions, statistically significant factors for trainees were right-sided (p<0.01), non-polypoid (p<0.01), not reddish (p<0.01), and inadequate bowel preparation (p<0.01) (Table 5). There were no significant factors for experts.
Table 5.
The Characteristics of Not Diagnosed Lesions under BLI for Trainees and Experts.
| Trainees Not-diagnosed |
Trainees Diagnosed |
P value | Experts Not-diagnosed |
Experts Diagnosed |
P value | |
|---|---|---|---|---|---|---|
| Lesion number, (%) | 33 (11.0) | 267 (89.0) | - | 18 (6.0%) | 282 (94.0) | - |
| Tumor size, mm, mean±SD (range) | 4.3±2.3 (2-10) | 4.7±2.6 (2-10) | 0.61 | 4.3±3.3 (2-10) | 4.7±2.5 (2-10) | 0.69 |
| Location, % (n), right-sided/left-sided | 72.7/27.3 (24/9) | 44.9/55.1 (120/147) | <0.01 | 16.7/83.3 (1/5) | 50.0/50.0 (47/47) | 0.11 |
| Morphology, % (n), polypoid/non-polypoid | 54.5/45.5 (12/21) | 67.4/33.6 (180/87) | <0.01 | 33.3/66.7 (2/4) | 66.0/34.0 (62/32) | 0.10 |
| Neoplastic/hyperplastic, % (n) | 36.4/63.6 (21/12) | 67.4/33.6 (180/87) | 0.66 | 66.7/33.3 (4/2) | 67.0/33.0 (63/31) | 0.98 |
| Color, % (n), reddish/ not reddish | 9.1/90.9 (3/30) | 34.8/65.2 (93/174) | <0.01 | 16.7/83.3 (1/5) | 33.0/67.0 (31/63) | 0.40 |
| Bowel preparation, % (n), Adequate/inadequate | 36.4/63.6 (12/21) | 87.6/12.4 (234/33) | <0.01 | 100.0/0.0 (6/0) | 80.9/19.1 (76/18) | 0.52 |
Discussion
In the current study, by using two unique movies with or without CAD EYE for the same lesion, we successfully demonstrated the efficacy of CAD EYE detecting and diagnosing polyps that were not detected or not diagnosed by experts or trainees. Additionally, risk factors for not-detected or not-diagnosed lesions in those endoscopists were clarified.
Regarding lesion detection, a previous study using 234 endoscopic pictures showed detection rates for CAD EYE of 91.7% for WLI and 94.7% for LCI[19]. Compared to these rates, those for experts were 94.6% for WLI (p<0.05) and 95.7% for LCI (not significant) and those for trainees were 85.4% for WLI (p<0.05) and 87.8% for LCI (p<0.05). Thus, CAD EYE could help trainees to detect lesions using both WLI and LCI. Another study also showed good results of lesion detection by means of CAD EYE regarding the sensitivity (94.5% for WLI and 96.0 for LCI) using 579 WLI images and 605 LCI images[18]. A study also examined LCI movies of 240 lesions, with CAD EYE being able to detect all of polyps[27]. Our previous study showed that CAD EYE could detect 85.0% for WLI and 89.0% for LCI among 100 polyps ≤10 mm in size detected by experts, indicating that expert's vision was better than CAD EYE[20]. In the current study, unlike previous studies, we uniquely used movies in which CAD EYE accurately detected lesions. Movies without CAD EYE were first reviewed by each endoscopist in the test phase, and then movies with CAD EYE were reviewed in the confirmation phase. In this setting, the endoscopists could actually know the efficacy for lesion detection with CAD EYE if they did not find any lesions in the test phase. The efficacy was higher in trainees, because the rates between two phases were higher in trainees than in experts. In addition, we considered another possible efficacy of this study was learning effects for lesion detection using these movies, especially for trainees though clinical efficacy of this effect should be examined. With respect to lesions size, our previous study about CAD EYE and the current study examined lesions of ≤10 mm. Our previous multicenter study using short movies of polyps under WLI and LCI to examine polyp visibility showed that the polyp visibility score (excellent visibility: 4-poor visibility: 1) by endoscopists was lower for polyps of <10 mm than for polyps of ≥10 mm[9]. Therefore, to efficiently investigate the efficacy of CAD, only lesions of ≤10 mm were examined in this study. However, further study should be expected for lesions >10 mm such as laterally spreading tumors.
Regarding lesion diagnosis, a previous study showed sensitivity, specificity, and accuracy values of 85.0%, 79.4%, and 83.6%, respectively, for non-magnified BLI with CAD EYE; 79.0%, 80.4%, and 79.3% for experts (not significant to CAD EYE); and 86.0% (not significant), 62.7% (p<0.05), and 80.1% (not significant) for non-experts, respectively[18]. Another study showed good results of the sensitivity, specificity, accuracy for magnified BLI with CAD EYE were 96.3%, 88.7%, and 94.9%, respectively[19]. Our previous study with CAD EYE found that those rates were 90.9%, 85.2%, and 87.8% for magnified BLI, and 91.7%, 86.8%, and 88.8% for non-magnified BLI, respectively[20]. Compared to those with magnified BLI with CAD EYE, those rates for trainees were significantly different about the accuracy of magnified BLI (82.2%, 76.4%, 79.0%, p=0.04 for accuracy) but not for experts (93.3%, 90.9%, 92.0%, p=0.17 for accuracy). Further analyses are required to confirm this non-inferiority. In the current study, the enrolled subjects were selected only for lesions that could be accurately diagnosed using CAD EYE. However, only 85.3% and 92.3% of the enrolled lesions could be diagnosed accurately in the test phase by trainees and experts, respectively. Most of these lesions could be diagnosed in the confirmation phase with movies with CAD EYE. Thus, we suggest that CAD EYE can potentially help to diagnose difficult lesions with a low confidence for both trainees and experts. Additionally, we considered these movies also can be used for learning of polyp diagnosis for trainees though the efficacy of learning should be examined.
In the current study, we examined the characteristics of non-detected lesions, identifying risk factors. Right-sided, not reddish, and in the corner were significant risk factors for both trainees and experts. Hyperplastic, halation, and inadequate bowel preparation were significant risk factors only for trainees. Experts could solve these factors with enough experiences. According to increase of experience, we consider endoscopists can know how hyperplastic polyps, polyps with halation, and polyps in inadequate preparation look like. A previous systematic review showed that miss rates of polyps were higher for proximal advanced adenomas (14%: 95% CI 5%-26%), serrated polyps (27%: 95% CI 16%-40%), and flat adenomas (34%: 95% CI 24%-45%), respectively[6]. Compared to this study, we could find some new risk factors in the current study. However, CAD EYE enabled the detection of the lesions with these various risk factors. Additionally, a study performed in Australia showed a mean ADR of 36.8% in the morning compared with a mean ADR of 30.5% in the afternoon (p<0.0001)[28]. Moreover, for every one-hour delay in commencing the procedure, they reported a reduction in mean ADR by 3.4%. The reason for this was partially due to the endoscopists' fatigue and narrow vision. We suggest that CAD EYE can help resolve this problem. Additionally, we clarified various risk factors for lesion diagnosis for trainees. Our results suggested that these factors also could be solved by CAD EYE.
The efficacy of CAD EYE in recent clinical trials were controversial. A study for examined of endoscopists for diminutive rectosigmoid polyps of ≤5 mm[21]. The accuracy and negative predictive values (NPV) of AI-assisted group optical diagnosis for experts and non-experts were 91.9%/82.3% for experts and 82.3%/88.6% for non-experts though NPV≥90% was expected by both the United States Multi-Society Task Force and European Society of Gastrointestinal Endoscopy guideline[29,30]. However, after about 200 experiences of cases, these rates for the last 50 lesions increased 88.0%/95.2% for non-experts. Another study examined the efficacy of CAD EYE for lesion detection by 6 trainees with experience of ≤20 colonoscopies[22]. The ADR between observation with CAD EYE group and standard observation group were not significant (58.4% vs. 61.0%, p=0.690). However, adenoma miss rates were significant (25.6% and 38.6%, p=0.033). These studies suggested CAD EYE in clinical setting was not enough but should be improved for both lesion detection and diagnosis though our results using movies showed the efficacy of CAD EYE for them.
This study presented some limitations. First, this was a single-center study with a small sample size. The cases were collected only one day a week due to the working schedule of experts (N.Y. and K.I.), although consecutive cases were enrolled. There may have been selection bias. Evaluations were performed by only three trainees and three experts. Evaluations of some risk factors and disturbance of false positive detection were performed subjectively by endoscopists. All movies were evaluated by each endoscopist using a desktop computer with a 16-inch monitor at a distance of 50 cm from the monitor. It was different from an endoscopic monitor and had a possibility to affect the results. All movies were recorded by expert endoscopists who could examine entire lumen with minimum blind areas. Since our study was a diagnostic investigation using short unique movie clips that had been recorded in advance, it was unlike an actual endoscopy.
In conclusion, this study suggested CAD EYE could help detect and diagnose lesions, regardless of endoscopist's experience. CAD EYE can be useful for standardization of endoscopic quality about lesion detection and diagnosis in various areas and countries.
Conflicts of Interest
The LED and LASER endoscopes, CAD EYE, and endoscopic LED systems were lent by Fujifilm for this study. Osamu Dohi and Naohisa Yoshida received the grants for research from Fujifilm Yoshida. Other authors did not have any conflict of interest. No funding was received for this study.
Author Contributions
Yoshida N organized the study, performed data collection and analysis and prepared the manuscript; Kobayashi R collected data and performed statistical analysis, Tomita Y, Hashimoto H, Inoue K, Hirose R, Dohi O, Inada Y, Murakami T, Morimoto Y arranged the study and performed colonoscopy; Zhu X helped the manuscript about the analysis of CAD EYE; and Itoh Y arranged the study plan and reviewed the manuscript.
Approval by Institutional Review Board (IRB)
ERB-C-1319 in the Ethics Committee of Kyoto Prefectural University of Medicine
Supplementary Files
Three lesions with or without CAD EYE recorded in a same situation under WLI and LCI for lesion detection and BLI for lesion characterization.
Acknowledgements
We thank Fujifilm and all members of the Department of Molecular Gastroenterology and Hepatology, Kyoto Prefectural University of Medicine for their help with this study.
References
- 1.Zauber AG, Winawer SJ, O'Brien MJ, et al. Colonoscopic polypectomy and long-term prevention of colorectal-cancer deaths. N Engl J Med. 2012 Feb; 366(8): 687-96. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Nishihara R, Wu K, Lochhead P, et al. Long-term colorectal-cancer incidence and mortality after lower endoscopy. N Engl J Med. 2013 Sep; 369(12): 1095-105. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Shaukat A, Kaltenbach T, Dominitz JA, et al. Endoscopic Recognition and Management Strategies for Malignant Colorectal Polyps: Recommendations of the US Multi-Society Task Force on Colorectal Cancer. Gastroenterology. 2020 Nov; 159(5): 1916-34. [DOI] [PubMed] [Google Scholar]
- 4.Tanaka S, Saitoh Y, Matsuda T, et al. Evidence-based clinical practice guidelines for management of colorectal polyps. J Gastroenterol. 2021 Apr; 56(4): 323-35. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Ferlitsch M, Moss A, Hassan C, et al. Colorectal polypectomy and endoscopic mucosal resection (EMR): European Society of Gastrointestinal Endoscopy (ESGE) Clinical Guideline. Endoscopy. 2017 Mar; 49(3): 270-97. [DOI] [PubMed] [Google Scholar]
- 6.Zhao S, Wang S, Pan P, et al. Magnitude, Risk Factors, and Factors Associated With Adenoma Miss Rate of Tandem Colonoscopy: A Systematic Review and Meta-analysis. Gastroenterology. 2019 May; 156(6): 1661-74.e11. [DOI] [PubMed] [Google Scholar]
- 7.Suzuki S, Aniwan S, Chiu HM, et al. Linked-Color Imaging Detects More Colorectal Adenoma and Serrated Lesions: An International Randomized Controlled Trial. Clin Gastroenterol Hepatol. 2023 Jun; 21(6): 1493-502. [DOI] [PubMed] [Google Scholar]
- 8.Shinozaki S, Kobayashi Y, Hayashi Y, et al. Colon polyp detection using linked color imaging compared to white light imaging: Systematic review and meta-analysis. Dig Endosc. 2020 Sep; 32(6): 874-81. [DOI] [PubMed] [Google Scholar]
- 9.Yoshida N, Hisabe T, Ikematsu H, et al. Comparison between linked color imaging and blue laser imaging for improving the visibility of flat colorectal polyps-A multicenter pilot study-. Dig Dis Sci. 2020 Jul; 65(7): 2054-62. [DOI] [PubMed] [Google Scholar]
- 10.Atkinson NSS, Ket S, Bassett P, et al. Narrow-Band Imaging for Detection of Neoplasia at Colonoscopy: A meta-analysis of data from Individual Patients in Randomized Controlled Trials. Gastroenterology. 2019 Aug; 157(2): 462-71. [DOI] [PubMed] [Google Scholar]
- 11.Yoshida N, Dohi O, Inoue K, et al. Blue laser imaging, blue light imaging, and linked color imaging for the detection and characterization of colorectal tumors. Gut Liver. 2019 Mar; 13(2): 140-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Iwatate M, Sano Y, Tanaka S, et al. Validation study for development of the Japan NBI Expert Team classification of colorectal lesions. Dig Endosc. 2018 Sep; 30(5): 642-51. [DOI] [PubMed] [Google Scholar]
- 13.Misawa M, Kudo SE, Mori Y, et al. Development of a computer-aided detection system for colonoscopy and a publicly accessible large colonoscopy video database (with video). Gastrointest Endosc. 2021 Apr; 93(4): 960-7. [DOI] [PubMed] [Google Scholar]
- 14.Repici A, Badalamenti M, Maselli R, et al. Efficacy of Real-Time Computer-Aided Detection of Colorectal Neoplasia in a Randomized Trial. Gastroenterology. 2020 Aug; 159(2): 512-20. [DOI] [PubMed] [Google Scholar]
- 15.Nemoto D, Guo Z, Katsuki S, et al. Computer-Aided Diagnosis of Early-Stage Colorectal Cancer Using Non-Magnified Endoscopic White Light Images. Gastrointest Endosc. 2023 Feb; S0016-5107(23)00089-5. doi: 10.1016/j.gie.2023.01.050. Online ahead of print. [DOI] [PubMed] [Google Scholar]
- 16.Barua I, Vinsard DG, Jodal HC, et al. Artificial intelligence for polyp detection during colonoscopy: a systematic review and meta-analysis. Endoscopy. 2021 Mar; 53(3): 277-84. [DOI] [PubMed] [Google Scholar]
- 17.Hassan C, Spadaccini M, Iannone A, et al. Performance of artificial intelligence in colonoscopy for adenoma and polyp detection: a systematic review and meta-analysis. Gastrointest Endosc. 2021 Jan; 93(1): 77-85. [DOI] [PubMed] [Google Scholar]
- 18.Sakamoto T, Nakashima H, Nakamura K, et al. Performance of Computer-Aided Detection and Diagnosis of Colorectal Polyps Compares to That of Experienced Endoscopists. Dig Dis Sci. 2022 Aug; 67(8): 3976-83. [DOI] [PubMed] [Google Scholar]
- 19.Weigt J, Repici A, Antonelli G, et al. Performance of a new integrated CADe/CADx system for detection and characterization of colorectal neoplasia. Endoscopy. 2022 Feb; 54(2): 180-4. [DOI] [PubMed] [Google Scholar]
- 20.Yoshida N, Inoue K, Tomita Y, et al. An analysis of the function of a new artificial intelligence, CAD EYE with the lesion recognition and diagnosis for colorectal polyps in clinical practice. Int J Colorectal Dis. 2021 Oct; 36(10): 2237-45. [DOI] [PubMed] [Google Scholar]
- 21.Rondonotti E, Hassan C, Tamanini G, et al. Artificial intelligence-assisted optical diagnosis for the resect-and-discard strategy in clinical practice: the Artificial intelligence BLI Characterization (ABC) study. Endoscopy. 2023 Jan; 55(1): 14-22. [DOI] [PubMed] [Google Scholar]
- 22.Yamaguchi D, Shimoda R, Miyahara K, et al. Impact of an artificial intelligence-aided endoscopic diagnosis system on improving endoscopy quality for trainees in colonoscopy: Prospective, randomized, multicenter study. Dig Endosc. 2023 Apr. doi: 10.1111/den.14573. Online ahead of print. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Dos Santos CEO, Malaman D, Sanmartin IDA, et al. Performance of artificial intelligence in the characterization of colorectal lesions. Saudi J Gastroenterol. 2023 Jul-Aug; 29(4): 219-24. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Aronchick CA. Bowel preparation scale. Gastrointest Endosc. 2004 Dec; 60(6): 1037-8. [DOI] [PubMed] [Google Scholar]
- 25.Participants in the Paris workshop. The Paris endoscopic classification of superficial neoplastic lesions: Esophagus, stomach, and colon-November 30 to December 1. 2002. Gastrointest Endosc. 2003 Dec; 58(6 Suppl): S3-43. [DOI] [PubMed] [Google Scholar]
- 26.WHO Classification of Tumours Editorial Board. WHO Classification of Tumours. Digestive system tumours. 5th ed. Lyon: International Agency for Research on Cancer; 2019. p. 532-4. [Google Scholar]
- 27.Neumann H, Kreft A, Scanathan V, et al. Evaluation of novel LCI CAD EYE system for real time detection of colon polyps. PLoS One. 2021 Aug; 16(8): e0255955. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Zorron Cheng Tao Pu L, Lu K, Ovenden A, et al. Effect of time of day and specialty on polyp detection rates in Australia. J Gastroenterol Hepatol. 2019 May; 34(5): 899-906. [DOI] [PubMed] [Google Scholar]
- 29.Shaukat A, Kaltenbach T, Dominitz JA, et al. Endoscopic recognition and management strategies for malignant colorectal polyps: Recommendations of the US Multi-Society Task Force on Colorectal Cancer. Am J Gastroenterol. 2020 Nov; 115(11): 1751-67. [DOI] [PubMed] [Google Scholar]
- 30.Hassan C, Antonelli G, Dumonceau JM, et al. Post-polypectomy colonoscopy surveillance: European Society of Gastrointestinal Endoscopy (ESGE) Guideline-Update 2020. Endoscopy. 2020 Aug; 52(8): 687-700. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Three lesions with or without CAD EYE recorded in a same situation under WLI and LCI for lesion detection and BLI for lesion characterization.