Department of Endoscopy, Hiroshima University, Hiroshima, Japan.
Patients and Main Outcome Measurements
A total of 189 colorectal lesions were analyzed. Each lesion was observed by NBI magnifying endoscopy and classified according to microvascular features (ie, thickness and irregularity). Microvessel thickness was classified as invisible, thin, or thick, and microvessel irregularity was classified as invisible, regular, mildly irregular, or severely irregular. NBI endoscopic images were compared with histologic findings.
Results
With respect to microvessel thickness, invisible microvessels were found significantly more often in hyperplasia lesions, and thick microvessels were found significantly more often in carcinoma with submucosal massive invasion (sm-m) (P < .01). With respect to microvessel irregularity, invisible microvessels were found significantly often in hyperplasia lesions, and severely irregular microvessels were found significantly often in sm-m lesions (P < .01). Accuracy of diagnosis of sm-m on the basis of thick and severely irregular lesions was 100%.
Conclusion
Microvascular features determined by NBI magnification are associated with histologic grade and depth of submucosal invasion. These results indicate that NBI magnification is useful for the prediction of histologic diagnosis and selection of therapeutic strategies of colorectal tumors.
Section snippets
Lesions and colonoscopic observations
A total of 189 colorectal lesions from 163 patients who underwent endoscopic or surgical resection at Hiroshima University Hospital were analyzed. Each patient provided informed consent. Lesions were first detected by conventional views and then observed by magnified views with the NBI system. Two specialists with more than 10 years of experience in the field of colonoscopy performed the endoscopic observations in this study. The instruments used in this study were a magnifying videoendoscope
MVC and microvessel diameter as determined by immunohistochemistry
Immunohistologic features of microvessels in HP lesions and sm-m are shown in Figure 3. Endothelial cells were stained with an anti-CD34 monoclonal antibody. Microvessel diameter was small and of homogeneous size, and MVC was low in HP lesions. Microvessel diameter was of uneven size (small and large microvessels mixed together), and MVC was greater in sm-m lesions than in other lesions.
MVC ranged from 5 to 52, with a mean ± SD of 19.5 ± 8.4 μm (median, 20 μm). Therefore, lesions were
Discussion
Magnifying endoscopy provides details of the surface of the GI tract, which has made it possible to examine the pit pattern of colorectal tumors, even during routine colonoscopic examination by using dye spraying, such as indigo carmine.14, 15, 16, 17, 18, 19 The pit pattern classification of colorectal tumors (type I-V pit pattern), based on stereomicroscopy, as proposed by Kudo et al,18, 19 is reported to be useful for evaluating the histologic features of tumors. Assessment of pit pattern is
Endoscopic observation of tissue by narrow-band illumination
Opt Rev
(2003)
K. Gono et al.
Appearance of enhanced tissue features in narrow-band endoscopic imaging
J Biomed Opt
(2004)
Y. Sano et al.
Optical/digital chromoendoscopy during colonoscopy using narrow-band imaging system
Dig Endosc
(2005)
H. Machida et al.
Narrow-band imaging in the diagnosis of colorectal mucosal lesions: a pilot study
Endoscopy
(2004)
Y. Hamamoto et al.
Usefulness of narrow-band imaging endoscopy for diagnosis of Barrett's esophagus
J Gastroenterol
(2004)
K. Shibuya et al.
High magnification bronchovideoscopy combined with narrow band imaging could detect capillary loops of angiogenic squamous dysplasia in heavy smokers at high risk for lung cancer
Thorax
(2003)
S. Tanaka et al.
Conditions of curability after endoscopic resection for colorectal carcinoma with submucosally massive invasion
To analyze the consistency of endoscopic biopsy of colorectal high-grade intraepithelial neoplasia (HGIN) and pathological diagnosis and to explore the value of preoperative examination in differentiating HGIN from invasive carcinoma.
Clinicopathological data of 79 patients with colorectal HGIN undergoing preoperative endoscopic biopsy from January 2012 to December 2017 were retrospectively analyzed.
Pathologically, 57 cases (72.8%) were diagnosed as invasive carcinoma and 22 (27.8%) as HGIN. Tumor size ≥3 cm, ulcer on the surface of the lesion, HGIN without adenoma, platelet to lymphocyte ratio, neutrophil to lymphocyte ratio, lymph node enlargement, and spiculation of the peri-intestinal fat on computed tomography were associated with postoperative invasive carcinoma. Multivariate analysis showed that a longest diameter ≥3 cm, preoperative diagnosis of HGIN without adenoma, and spiculation of the peri-intestinal fat were independent factors for a postoperative diagnosis of invasive carcinoma. Depending on the weight of these three independent factors in binary logistic regression analysis, a comprehensive scoring model was established. When the score was ≥1.5, the sensitivity and specificity for the diagnosis of invasive carcinoma were 86.0% and 81.8%, respectively. Utilizing the prediction index, the area under the receiver operating characteristic curve was 0.869.
A diagnosis of colorectal HGIN by colonoscopy is poorly consistent with the postoperative pathological diagnosis. The scoring model established in this study for identifying colorectal infiltrating carcinoma is simple and feasible. When the comprehensive score is ≥ 1.5, an aggressive approach of surgical treatment is recommended.
The latter method makes it easier for the endoscopist to localize the polyp but may be distracting for histologic assessment. Output of CADx also includes 2 kinds of patterns: pathology prediction32,34,37,50 (eg, non-neoplastic or neoplastic) and endoscopic classification that can then be extrapolated to the histopathology of the polyp29,31 (eg, Narrow-band Imaging [NBI]-International Colorectal-Endoscopic [NICE] classification,51 Sano classification,52 or Hiroshima classification53). Similarly, each type of output has pros and cons.
Recent breakthroughs in artificial intelligence (AI), specifically via its emerging sub-field “deep learning,” have direct implications for computer-aided detection and diagnosis (CADe and/or CADx) for colonoscopy. AI is expected to have at least 2 major roles in colonoscopy practice—polyp detection (CADe) and polyp characterization (CADx). CADe has the potential to decrease the polyp miss rate, contributing to improving adenoma detection, whereas CADx can improve the accuracy of colorectal polyp optical diagnosis, leading to reduction of unnecessary polypectomy of non-neoplastic lesions, potential implementation of a resect-and-discard paradigm, and proper application of advanced resection techniques. A growing number of medical-engineering researchers are developing both CADe and CADx systems, some of which allow real-time recognition of polyps or in vivo identification of adenomas, with over 90% accuracy. However, the quality of the developed AI systems as well as that of the study designs vary significantly, hence raising some concerns regarding the generalization of the proposed AI systems. Initial studies were conducted in an exploratory or retrospective fashion by using stored images and likely overestimating the results. These drawbacks potentially hinder smooth implementation of this novel technology into colonoscopy practice. The aim of this article is to review both contributions and limitations in recent machine-learning-based CADe and/or CADx colonoscopy studies and propose some principles that should underlie system development and clinical testing.
The Japan NBI Expert Team (JNET) classification is the first universal narrow-band imaging magnifying endoscopic classification of colorectal tumors. Considering each type in this classification, the diagnostic ability of Type 2B is the weakest. Generally, clinical behavior is believed to be different in each gross type of colorectal tumor. We evaluated the differences in the diagnostic performance of JNET classification for each gross type (polypoid and superficial) and examined whether the diagnostic performance of Type 2B could be improved by subtyping.
We analyzed 2933 consecutive cases of colorectal lesions, including 136 hyperplastic polyps/sessile serrated polyps, 1926 low-grade dysplasias (LGDs), 571 high-grade dysplasias (HGDs), and 300 submucosal (SM) carcinomas. We classified lesions as polypoid and superficial type and compared the diagnostic performance of the classification system in each type. Additionally, we subtyped Type 2B into 2B-low and 2B-high based on the level of irregularity in surface and vessel patterns, and we evaluated the relationship between the subtypes and histology, as analyzed separately for polypoid and superficial types. We also estimated interobserver and intraobserver variability.
The diagnostic performance of JNET classification did not differ significantly between polypoid and superficial lesions. Ninety-nine percent of Type 2B-low lesions were LGDs, HGDs, or superficial submucosal invasive (SM-s) carcinomas. In contrast, 60% of Type 2B-high lesions were deep submucosal invasive (SM-d) carcinomas. The results were not different between each gross type. Interobserver and intraobserver agreements for Type 2B subtyping were good, with kappa values of .743 and .786, respectively.
Type 2B subtyping may be useful for identifying lesions that are appropriate for endoscopic resection. JNET classification and Type 2B sub classification are useful criteria, regardless of gross type.
The Japan NBI Expert Team (JNET) was established in 2011 and has proposed a universal narrow-band imaging (NBI) magnifying endoscopic classification of colorectal tumors. The aim of this study was to evaluate the clinical usefulness of the JNET classification for colorectal lesions.
We analyzed 2933 colorectal lesions, which were diagnosed by NBI magnifying observation before endoscopic treatment or surgery. The colorectal lesions consisted of 136 hyperplastic polyps/sessile serrated polyps (HPs/SSPs), 1926 low-grade dysplasia (LGD), 571 high-grade dysplasia (HGD), 87 superficial submucosal invasive (SM-s) carcinomas, and 213 deep submucosal invasive (SM-d) carcinomas. We evaluated the relationship between the JNET classification and the histologic findings of these lesions.
The sensitivity, specificity, positive and negative predictive values, and accuracy of Type 1 lesions for the diagnosis of HP/SSP were, respectively, 87.5%, 99.9%, 97.5%, 99.4%, and 99.3%; of Type 2A lesions for the diagnosis of LGD were 74.3%, 92.7%, 98.3%, 38.7%, and 77.1%; of Type 2B lesions for the diagnosis of HGD/SM-s carcinoma were 61.9%, 82.8%, 50.9%, 88.2%, and 78.1%; for Type 3 lesions for the diagnosis of SM-d carcinoma were 55.4%, 99.8%, 95.2%, 96.6%, and 96.6%, respectively.
Types 1, 2A, and 3 of the JNET classification were very reliable indicators for HP/SSP, LGD, and SM-d carcinoma, respectively. However, the specificity and positive predictive value of Type 2B were relatively lower than those of others. Therefore, an additional examination such as pit pattern diagnosis using chromoagents is necessary for accurate diagnosis of Type 2B lesions.
