Abstract
Background: Artificial intelligence (AI) has significantly impacted medical imaging, particularly in gastrointestinal endoscopy. Computer-aided detection and diagnosis systems (CADe and CADx) are thought to enhance the quality of colonoscopy procedures. Summary: Colonoscopy is essential for colorectal cancer screening but often misses a significant percentage of adenomas. AI-assisted systems employing deep learning offer improved detection and differentiation of colorectal polyps, potentially increasing adenoma detection rates by 8%–10%. The main benefit of CADe is in detecting small adenomas, whereas it has a limited impact on advanced neoplasm detection. Recent advancements include real-time CADe systems and CADx for histopathological predictions, aiding in the differentiation of neoplastic and nonneoplastic lesions. Biases such as the Hawthorne effect and potential overdiagnosis necessitate large-scale clinical trials to validate the long-term benefits of AI. Additionally, novel concepts such as computer-aided quality improvement systems are emerging to address limitations facing current CADe systems. Key Messages: Despite the potential of AI for enhancing colonoscopy outcomes, its effectiveness in reducing colorectal cancer incidence and mortality remains unproven. Further prospective studies are essential to establish the overall utility and clinical benefits of AI in colonoscopy.
Introduction
Artificial intelligence (AI) has heralded a new era of innovation across various fields, particularly in healthcare. A key area of interest is the application of AI in medical imaging. Computer-aided detection and diagnosis systems (CADe and CADx, respectively) have shown significant potential in gastrointestinal endoscopy; these AI-powered systems assist endoscopists in spotting abnormal regions and distinguishing between detected abnormalities, which could lead to better patient outcomes.
One particular procedure in which these systems could be of particular benefit is colonoscopy, which is essential for colorectal cancer screening but has well-documented limitations; approximately one-fourth of colorectal adenomas are missed during a single colonoscopy [1], and differentiation between adenomas and non-adenomas frequently falls below expected thresholds [2]. Several factors can influence the quality of colonoscopies, including physician experience and the quality of bowel preparation. AI-assisted systems employing deep learning may enhance the quality of colonoscopies by facilitating the detection and diagnosis of colorectal polyps.
Although these systems appear to offer many advantages, a thorough evaluation of their performance in real-world environments is crucial. In this review, we aimed to provide an overview of the latest research trends in AI applications for colonoscopy. In addition, we will examine how these systems have improved clinical outcomes and explore emerging concepts such as computer-aided quality improvement (CAQ). We hope to add to the growing knowledge base on the use of AI in colonoscopy and suggest possible future directions for research.
Research Trends in CADe for Colonoscopy
In the realm of colonoscopy, adenoma detection rate (ADR) is ragarded as one of the most important quality indicator because it inversely associates with colorectal cancer death [3]. In addition, missing lesions (such as polyps) has been a longstanding challenge [1]. Sixty percent of post-colonoscopy colorectal cancers, which are sometimes fatal, are thought to be caused by missing neoplasms [4]. This challenge has been a driving factor in the development of a CADe system to accurately identify and reduce missed polyps. Since the 2000s, various image feature parameters such as edge detection and texture analysis have been examined for integration into machine learning processes [5], primarily within information engineering for CADe. However, the accuracy of these CADe systems never consistently exceeded 90%, and there were no instances of successful real-time detection support because of computing limitations.
The advent of deep learning in the 2010s changed the situation, with accuracy levels exceeding 90% and the emergence of the potential for real-time detection. In 2018, Misawa [6], Urban [7], and Wang [8] sequentially reported the use of deep learning for real-time functioning CADe systems, underscoring its potential clinical applications in colonoscopy and igniting extensive research and development efforts. As of March 2024, commercial distribution in Japan has begun for products such as EndoBRAIN®-EYE (developed by Showa University, Nagoya University, and Cybernet Systems Corp.), WISE VISION® (developed by NEC Corporation and the National Cancer Center), CAD EYE® (developed by Fujifilm Corporation), and EIRL Colon Polyp (developed by LPIXEL), all of which have received approval from the Japanese regulatory body. Abroad, numerous other software products have been released. Within the scope of our research, several products have been launched on the market as CADe systems, including GI Genius (Medtronic, Dublin, Ireland), DISCOVERY (Pentax, Tokyo, Japan), ENDO-AID (Olympus, Tokyo, Japan), MAGENTIQ-COLO (Magentiq Eye, Haifa, Israel), CADDIE (Odin Vision, London, UK), EndoScreener (Wision AI, Chengdu, China), and SKOUT (Iterative Health, Cambridge, MA, USA). Among these, GI Genius, EndoScreener, and SKOUT have become commercially available following approval from the US Food and Drug Administration.
Is CADe Effective for Colonoscopy?
It is still unknown to what extent CADe systems enhance clinical outcomes. We searched Medline, using the following search words: (“colonoscopy” OR “colonoscopies”) AND (“artificial intelligence” OR “AI” OR “machine learning” OR “deep learning” OR “computer-aided”) AND (“detection”) AND (“prospective” OR “Randomized” OR “Randomised”). The literature search resulted in 180 articles; of these, 165 were excluded because they were not related to the search aim. In this analysis, back-to-back (tandem) colonoscopy designs were excluded. The ADR observed in tandem trials is inherently different from that of non-tandem trial designs due to the sequential nature of tandem procedures, which may artificially enhance detection rates. In this review, we aimed to assess the effectiveness of CADe under real-world conditions; therefore, tandem trials were excluded to ensure that the reported ADR more accurately reflects standard clinical practice. Table 1 summarizes the reported randomized control trials (RCTs) of CADe [9‒23]. In all but three trials, the use of CADe significantly increased the ADR. Recent meta-analyses of CADe have demonstrated that its utilization can enhance the ADR by approximately 8%–10% [24, 25]. However, these meta-analyses also showed that the increased detection rates of polyps using CADe is mainly related to diminutive adenomas (<5 mm in size), and there is no increase in advanced adenomas. A Japanese large-scale RCT that evaluated the efficacy of colonoscopies reported that even after a two-round baseline colonoscopy, advanced neoplasia, mainly composed of non-polypoid lesions (especially laterally spreading nongranular tumors), was detected at surveillance colonoscopy [26]. Since laterally spreading nongranular tumors and depressed-type neoplasms, which have high malignancy potential, are categorized as advanced neoplasms, such lesions may be missed even using CADe [27]. The latest meta-analysis that included 44 RCTs with 36,201 cases showed that the CADe did not increase the mean number of advanced colorectal neoplasia (0.16 vs. 0.15), but advanced colorectal neoplasia detection rate was higher in CADe group (12.7% vs. 11.5%) [28]. Based on the current evidence, there is insufficient strong evidence to support its significant benefit in improving the detection of advanced neoplasia.
Reported randomized control trials on computer-aided detection (CADe)
Author . | Year . | Patients, n . | ADR without CADe, % . | ADR with CADe, % . | p value . | AADR without CADe, % . | AADR with CADe, % . | p value . | CADe system . |
---|---|---|---|---|---|---|---|---|---|
Thiruvengadam et al. [23] | 2024 | 1,100 | 34.4 | 42.5 | 0.005 | 4.4 | 4.7 | 0.88 | GI Genius |
Maas et al. [22] | 2,024 | 916 | 30 | 37 | 0.014 | - | - | - | MAGENTIQ |
Lau et al. [21] | 2023 | 766 | 44.5 | 57.5 | <0.001 | 10.0 | 8.3 | 0.397 | ENDO-AID |
Mangas-Sanjuan et al. [18] | 2023 | 3,213 | 64.2 | 62.0 | 0.23 | 30.5 | 31.3 | 0.60 | GI Genius |
Karsenti et al. [17] | 2023 | 2,592 | 33.7 | 37.5 | 0.051 | 7.6 | 9.3 | 0.18 | GI Genius |
Nakashima et al. [19] | 2023 | 415 | 47.6 | 59.4 | 0.018 | 7.2 | 7.7 | 1 | CAD EYE |
Ahmad et al. [15] | 2023 | 614 | 65 | 71.4 | 0.09 | - | - | - | CAD EYE |
Gimeno-García et al. [16] | 2023 | 370 | 40.8 | 54.8 | 0.01 | 12.1 | 11.6 | 0.89 | ENDO-AID |
Xu et al. [20] | 2023 | 3,059 | 32.4 | 39.9 | 0.001 | 4.4 | 6.6 | 0.041 | Original |
Shaukat et al. [14] | 2022 | 1,440 | 43.9 | 47.8 | 0.065 | - | - | - | SKOUT |
Repici et al. [12] | 2020 | 685 | 40.4 | 54.8 | 0.04 | 8 | 9.8 | 0.26 | GI Genius |
Gong et al. [10] | 2020 | 704 | 8 | 16 | 0.001 | - | - | - | ENDOANGEL |
Wang et al. [13] | 2020 | 1,046 | 28 | 34 | 0.03 | 4 | 2 | 0.66 | EndoScreener |
Liu et al. [11] | 2020 | 1,026 | 23.89 | 39.1 | <0.001 | 6.45 | 2.88 | 0.821 | Original CADe |
Wang et al. [9] | 2019 | 1,058 | 20.3 | 29.1 | <0.001 | 5.95 | 3.41 | 0.803 | Original CADe |
Author . | Year . | Patients, n . | ADR without CADe, % . | ADR with CADe, % . | p value . | AADR without CADe, % . | AADR with CADe, % . | p value . | CADe system . |
---|---|---|---|---|---|---|---|---|---|
Thiruvengadam et al. [23] | 2024 | 1,100 | 34.4 | 42.5 | 0.005 | 4.4 | 4.7 | 0.88 | GI Genius |
Maas et al. [22] | 2,024 | 916 | 30 | 37 | 0.014 | - | - | - | MAGENTIQ |
Lau et al. [21] | 2023 | 766 | 44.5 | 57.5 | <0.001 | 10.0 | 8.3 | 0.397 | ENDO-AID |
Mangas-Sanjuan et al. [18] | 2023 | 3,213 | 64.2 | 62.0 | 0.23 | 30.5 | 31.3 | 0.60 | GI Genius |
Karsenti et al. [17] | 2023 | 2,592 | 33.7 | 37.5 | 0.051 | 7.6 | 9.3 | 0.18 | GI Genius |
Nakashima et al. [19] | 2023 | 415 | 47.6 | 59.4 | 0.018 | 7.2 | 7.7 | 1 | CAD EYE |
Ahmad et al. [15] | 2023 | 614 | 65 | 71.4 | 0.09 | - | - | - | CAD EYE |
Gimeno-García et al. [16] | 2023 | 370 | 40.8 | 54.8 | 0.01 | 12.1 | 11.6 | 0.89 | ENDO-AID |
Xu et al. [20] | 2023 | 3,059 | 32.4 | 39.9 | 0.001 | 4.4 | 6.6 | 0.041 | Original |
Shaukat et al. [14] | 2022 | 1,440 | 43.9 | 47.8 | 0.065 | - | - | - | SKOUT |
Repici et al. [12] | 2020 | 685 | 40.4 | 54.8 | 0.04 | 8 | 9.8 | 0.26 | GI Genius |
Gong et al. [10] | 2020 | 704 | 8 | 16 | 0.001 | - | - | - | ENDOANGEL |
Wang et al. [13] | 2020 | 1,046 | 28 | 34 | 0.03 | 4 | 2 | 0.66 | EndoScreener |
Liu et al. [11] | 2020 | 1,026 | 23.89 | 39.1 | <0.001 | 6.45 | 2.88 | 0.821 | Original CADe |
Wang et al. [9] | 2019 | 1,058 | 20.3 | 29.1 | <0.001 | 5.95 | 3.41 | 0.803 | Original CADe |
ADR, adenoma detection rate; AADR, advanced adenoma detection rate; CADe, computer-aided detection.
With regard to the adenoma miss rates, there are some reports. Wang et al. [29] reported the adenoma miss rate was significantly lower with CADe colonoscopy (13.89% vs. 40.00%). In Japan, Kamba et al. [30] conducted a similar tandem study. The results showed that the CADe colonoscopy showed lower adenoma miss rate (13.8% vs. 36.7%). The latest meta-analysis which contained six studies showed the adenoma miss rate was significantly lower with CADe (16.1% vs. 35.3%) [28].
There may be potential biases in these studies. Of particular concern is the Hawthorne effect, where endoscopists in the intervention group alter their behavior to improve the outcome. This bias may be present in reports from research groups involved in AI development or those with conflicts of interest with AI development companies. Consequently, large-scale prospective clinical studies, independent of AI developers, are deemed necessary to accurately ascertain the value of CADe.
Benefits and Harms of CADe in Colonoscopy
The benefits and harms of using CADe in colonoscopy are still unknown. As many studies have shown, CADe increases the ADR, a surrogate marker of colorectal cancer incidence and mortality. However, as mentioned above, the main incremental improvement was only seen with small adenomas. CADe for mammography was launched in the USA in the early 2000s and is widely used in screening; it has detected more early-stage cancers, though it has not improved cancer mortality [31]. This overdiagnosis might also be observed in colonoscopy screening if CADe is widely implemented. Long-term large-scale prospective studies are needed to show the true benefits and harms of using CADe in colonoscopy. In addition, from a short-term perspective, prolonged inspection time and an increase in the resection rate of nonneoplastic polyps are notable shortcomings of CADe use. Hassan et al. highlighted these limitations in a meta-analysis. Their analysis, which included 10 studies, observed significant differences between the CADe and control groups in terms of inspection time (9.22 vs. 8.73 min). With regard to nonneoplastic polyps, more nonneoplastic polyps were removed in the CADe group compared to standard colonoscopy (0.52 vs. 0.34 per colonoscopy) [25].
Trends in CADx Research
In clinical practice, a high degree of accuracy is required to differentiate whether a detected polyp is neoplastic and whether it is benign or malignant, as well as to decide on appropriate treatment. However, achieving an accurate differential diagnosis performance is sometimes difficult [32]. CADx that can output AI-based histopathological prediction may prove valuable in enhancing diagnostic accuracy.
Research on CADx within the field of colonoscopy has been ongoing since the 2010s, and multiple research groups have reported studies targeting vessels and surface patterns obtained from magnifying narrow-band imaging (NBI). Kominami et al. [33] developed a CADx for NBI using a traditional machine learning algorithm that was tested on 118 colorectal polyps in 41 patients, achieving a sensitivity of 93.0% for neoplasms. Subsequently, Chen et al. [34] and Byrne et al. [35] reported the utilization of deep learning in CADx, both achieving over 90% sensitivity in differentiating neoplasms. Mori et al. conducted a prospective study using EndoBRAIN, a CADx system designed for ultra-magnifying endoscopy, reporting a sensitivity of 92.7% and a specificity of 89.8% in differentiating neoplasms. The EndoBRAIN study underscores the potential value of CADx in real-world clinical practice [36] and has also been evaluated in an international multicenter study in Japan, Norway, and the UK. These studies also only included trainee endoscopists, the group expected to benefit most from AI. While the use of CADx did not lead to a significant increase in sensitivity (88.4% without vs. 90.4% with CADx), it significantly increased the proportion of high-confidence diagnoses (74.2% without vs. 92.6% with CADx) [37]. In 2021, Fujifilm Corporation launched CAD EYE, a system capable of differentiating neoplasms. Overall, CADx systems capable of differentiating neoplasms are gradually becoming more integrated into daily clinical practice.
Table 2 lists prospective studies that have used CADx [36‒44]. We conducted a literature search based on the following keywords in Medline: (“Computer-aided” OR “artificial intelligence”) AND (“colonoscopy”) AND (“prospective”) AND (“characterization” OR “differentiation” OR “diagnosis” OR “prediction”) NOT “detection”. As of May 2024, no RCT has been conducted on CADx, leading to a lower level of evidence. Additionally, Hassan et al. [45] recently reported a meta-analysis of CADx studies; surprisingly, there was no difference in the proportion of polyps predicted to be nonneoplastic that would avoid removal (55.4% without vs. 58.4% with CADx) or in the proportion of neoplastic polyps that would be erroneously left in situ (8.2% without vs. 7.5% with CADx). This may indicate that the precise development of CADx is difficult because of limitations in the supervised machine learning process; pathological diagnoses are subject to significant interobserver variability among pathologists, resulting in unreliable labeling of data fed into machine learning processes [46].
Reported prospective studies on computer-aided diagnosis (CADx)
Authors . | Year . | Number of the polyps . | Sensitivity . | Specificity . | Accuracy . | CADx system . | Study setting . |
---|---|---|---|---|---|---|---|
Rex et al. [44] | 2024 | 1,252 | 90.8 | 64.7 | GI Genius | AI-assisted | |
Houwen et al [41] | 2023 | 423 | 89 | 38 | 79 | Original | AI-alone + including SSL |
Hassan et al. [40] | 2023 | 319 | 81.8–86.4 | 92.4–94.0 | - | CAD EYE | AI-assisted |
GI Genius | |||||||
Li et al. [43] | 2023 | 661 | 61.8 | - | 71.6 | CAD EYE | AI-assisted |
Minegishi et al. [39] | 2022 | 465 | 94.4 | 62.5 | 85.5 | EndoBRAIN-X | AI-assisted + including SSL |
Rondonotti et al. [42] | 2023 | 596 | 88.6 | 88.1 | 88.4 | CAD EYE | AI-assisted |
Hassan et al. [38] | 2022 | 544 | 82 | 93.2 | 91.8 | GI Genius | AI-alone |
Barua et al. [37] | 2022 | 892 | 90.4 | 85.9 | - | EndoBRAIN | AI-assisted |
Mori et al. [36] | 2018 | 466 | 92.7 | 89.9 | - | EndoBRAIN | AI-alone |
Authors . | Year . | Number of the polyps . | Sensitivity . | Specificity . | Accuracy . | CADx system . | Study setting . |
---|---|---|---|---|---|---|---|
Rex et al. [44] | 2024 | 1,252 | 90.8 | 64.7 | GI Genius | AI-assisted | |
Houwen et al [41] | 2023 | 423 | 89 | 38 | 79 | Original | AI-alone + including SSL |
Hassan et al. [40] | 2023 | 319 | 81.8–86.4 | 92.4–94.0 | - | CAD EYE | AI-assisted |
GI Genius | |||||||
Li et al. [43] | 2023 | 661 | 61.8 | - | 71.6 | CAD EYE | AI-assisted |
Minegishi et al. [39] | 2022 | 465 | 94.4 | 62.5 | 85.5 | EndoBRAIN-X | AI-assisted + including SSL |
Rondonotti et al. [42] | 2023 | 596 | 88.6 | 88.1 | 88.4 | CAD EYE | AI-assisted |
Hassan et al. [38] | 2022 | 544 | 82 | 93.2 | 91.8 | GI Genius | AI-alone |
Barua et al. [37] | 2022 | 892 | 90.4 | 85.9 | - | EndoBRAIN | AI-assisted |
Mori et al. [36] | 2018 | 466 | 92.7 | 89.9 | - | EndoBRAIN | AI-alone |
AI-alone, the diagnostic performances were measured by standalone AI; AI-assisted, the diagnostic performances were measured according to endoscopists diagnosis with reference to the CADx; CADx, computer-aided diagnosis; SSL, sessile serrated lesion.
Other Target Lesions for CADx
CADx systems were primarily designed to assist in differentiating neoplasms from non-neoplasms; however, as AI technology has advanced, new targets for CADx are being eagerly investigated, including depth prediction of early colorectal cancer and differentiation between sessile serrated lesions (SSLs) and hyperplastic polyps. In the context of clinical practice, SSLs are currently classified as a type of neoplastic polyp; however, differentiating SSLs from hyperplastic polyps can often be challenging, as is diagnosing the depth of a cancer. Consequently, there is an increasing focus on more sophisticated CADx systems, moving beyond simple differentiation of neoplasms.
Tamai et al. [47] developed a CADx for diagnosing invasive cancers using magnifying NBI, reporting a sensitivity of 83.9% and specificity of 82.6%. Takeda et al. [48] also reported an accuracy of 94.1% for invasive cancers using CADx for endocytoscopy. Tokunaga et al. [49] created a more practical CADx system that could predict based on white-light images; this system was retrospectively verified to have an accuracy of 90.3%. Okamoto et al. developed a CADx to identify JNET classification types 1, 2A, 2B, and 3, achieving an accuracy of 94.1% for JNET type 3 [50]. Minegishi et al. [39] conducted a prospective study of a CADx for SSLs; the sensitivity for neoplastic lesions, including SSL, was 94.4%, while the specificity was 62.5%. This CADx, named EndoBRAIN®-X (Cybernet Systems Corp. Tokyo, Japan), obtained approval from the Japanese regulatory body. Houwen et al. [41] developed a similar CADx system and conducted prospective clinical studies; the sensitivity of their CADx system for tumorous lesions, including SSL, was 89%, with a specificity of 38%.
Other AI Applications
Recently, studies finding negative results for CADe performance have been reported [51]. This has led to the emergence of, and growing acceptance of, a novel concept known as CAQ. The ability of CADe to detect lesions is limited to instances in which polyps are visually displayed on an endoscopic monitor; if polyps or cancers are located in a blind spot, such as folds, residual feces, or behind flexures, CADe is completely ineffective. This limitation arises because it is solely an image analysis technology, meaning that it is only effective for visualized lesions.
The development of a system with fewer blind spots to guide colonoscopy could work in conjunction with CADe to further reduce missed polyps. Yao et al. [52] developed a CAQ that monitors endoscope withdrawal speed, providing a real-time alarm if withdrawal is too fast, which significantly improved the ADR. The monitoring of withdrawal speed likely encouraged more meticulous inspection of the colonic mucosa, altering endoscopists’ practices and improving accuracy. Liu et al. [53] developed an AI-based system to improve the fold examination quality during colonoscopy. The system evaluates video frames to enhance real-time quality control during colonoscopic withdrawal. This study demonstrates that the system’s assessments correlate strongly with expert evaluations and historical ADRs, highlighting its potential to assist endoscopists in reducing blind spots and improving the overall quality of colonoscopy. McGill et al. [54] explored the use of AI to identify and quantify blind spots during colonoscopy. By reconstructing 3D colon segments in real-time, the AI system highlights unvisualized areas, offering potential guidance to endoscopists to minimize blind spots during procedures. While these approaches show promise in enhancing the effectiveness of CADe systems, it is important to note that these systems are not yet approved by regulatory bodies.
Man-Machine Collaborations in Clinical Practice
Even as AI performance improves, how the doctor who ultimately makes the diagnosis interprets the output will remain critically important; if the CADe outputs many false positives (FPs) despite having good sensitivity, endoscopists may ignore the model’s suggestions. Zhang et al. [55] reported that higher FPs were inversely associated with ADR; they hypothesized that the occurrence of FP alerts may occasionally disrupt endoscopists’ concentration, possibly reducing their trust in the CADe. Their results demonstrated that an ADR decrease resulted when the FPs per minute exceeded five, indicating it is crucial to focus on reducing FPs to ensure the generation of trusted CADes. Okumura et al. [56] reported a comparison of FPs before and after additional machine learning, finding that additional machine learning could reduce FPs by 70%.
Man-machine collaboration with CADx is more complicated. In the case of polyp detection, it is relatively easy to determine whether a CADe is incorrect, whereas verifying CADx diagnoses is more challenging; this is because the only endoscopists who can identify that a CADx is misdiagnosing in real time are those who are more accurate than the CADx. Endoscopists need to be able to dismiss when the AI misdiagnoses and follow when it is correct. Reverberi et al. [57] investigated the effectiveness of human-AI collaboration in medical decision-making, specifically examining CADx; endoscopists demonstrated the ability to selectively accept correct AI advice and reject that which was incorrect. However, this selective decision-making can be particularly challenging, as nonexpert endoscopists may lack the skills required to reliably identify misdiagnoses made by CADx. Djinbachian et al. [58] conducted a prospective study comparing CADx and endoscopist diagnosis based on CADx output, finding that the overall accuracy of CADx was better than human diagnosis (77.2% vs. 72.1%, respectively). This result indicates that human performance may be affected by CADx diagnosis.
Conclusion
The introduction of deep learning has democratized AI technology, enabling the rapid development of AI systems for colonoscopy. Every day, numerous new studies on endoscopic AI are reported, adding to our understanding of its potential and limitations in this field. A few AI systems have already received regulatory approval and are being implemented in clinical settings. However, while AI has shown tremendous potential in enhancing detection and diagnostic capabilities in colonoscopy, it is essential to remember that the long-term benefits of AI usage, such as reducing cancer incidence and mortality, remain unproven. Because of this, the focus should be on conducting large-scale prospective studies to verify these benefits and assess the overall utility of AI in endoscopy.
Acknowledgment
We thank Lisa Oberding, MSc, from Edanz (https://jp.edanz.com/ac) for editing a draft of this manuscript.
Conflict of Interest Statement
Shin-ei Kudo and Masashi Misawa received speaking honoraria from Olympus Corporation and have ownership interest in the products of Cybernet Systems.
Funding Sources
This research has not received any specific grant from any funding body in the public, commercial, or not-for-profit sectors.
Author Contributions
All authors made substantial contributions to the study concept, data analysis, or interpretation; drafted the manuscript or revised it critically for important intellectual content; approved the final version of the manuscript to be published; and agreed to be accountable for all aspects of the work. Each author’s CRediT roles are as follows: Masashi Misawa: conceptualization, data curation, formal analysis, methodology, writing – original draft, project administration, writing – original draft, formal analysis, data curation, and investigation. Shin-ei Kudo: supervision and project administration.