Artificial intelligence (AI) has been attracting considerable attention as an important scientific topic in the field of medicine. Deep‐leaning (DL) technologies have been applied more dominantly than other traditional machine‐learning methods. They have demonstrated excellent capability to retract visual features of objectives, even unnoticeable ones for humans, and analyze huge amounts of information within short periods. The amount of research applying DL‐based models to real‐time computer‐aided diagnosis (CAD) systems has been increasing steadily in the GI endoscopy field. An array of published data has already demonstrated the advantages of DL‐based CAD models in the detection and characterization of various neoplastic lesions, regardless of the level of the GI tract. Although the diagnostic performances and study designs vary widely, owing to a lack of academic standards to assess the capability of AI for GI endoscopic diagnosis fairly, the superiority of CAD models has been demonstrated for almost all applications studied so far. Most of the challenges associated with AI in the endoscopy field are general problems for AI models used in the real world outside of medical fields. Solutions have been explored seriously and some solutions have been tested in the endoscopy field. Given that AI has become the basic technology to make machines react to the environment, AI would be a major technological paradigm shift, for not only diagnosis but also treatment. In the near future, autonomous endoscopic diagnosis might no longer be just a dream, as we are witnessing with the advent of autonomously driven electric vehicles.

Objective: To evaluate the visibility of colorectal lesions using a novel image processing algorithm, texture and color enhancement imaging (TXI), that allows the acquisition of brighter images with enhanced color and surface structure.Methods: During August–September 2019, patients referred for endoscopic treatment were prospectively recruited. Electronic data acquired while observing colorectal lesions using white light imaging (WLI) were obtained and recorded: WLI, TXI mode1 (with color enhancement), and TXI mode2 (without color enhancement) videos were constructed. The lesions were also recorded using narrow‐band imaging (NBI) from the same perspective as WLI. Four video clips (WLI, TXI mode1, TXI mode2, and NBI) were made per lesion. Thereafter, video files for evaluations were prepared by randomly arranging all video clips. Finally, visualization scores were evaluated by four endoscopists, and the WLI, TXI mode1, TXI mode2, and NBI results were compared.Results: Overall, 22 patients with 68 lesions were recruited; the video file for evaluation subsequently comprised 272 randomly arranged video clips. Mean visualization scores using WLI, TXI mode1, TXI mode2, and NBI were 70.0 (±20.1), 80.5 (±18.6), 75.6 (±18.1), and 69.0 (±20.6), respectively. Mean visualization scores for flat lesions using WLI, TXI mode1, TXI mode2, and NBI were 64.1 (±21.2), 76.5 (±20.18), 71.8 (±19.4), and 64.2 (±22.0), respectively. Visualization scores using TXI mode1 were significantly better than those using WLI, TXI mode2, or NBI.Conclusions: TXI enables improved visualization of colorectal lesions, even flat lesions, than WLI and NBI. TXI may allow better detection of colorectal lesions, although further prospective studies are required.

This study aimed to evaluate the diagnostic utility of the ultra-thin endoscope (UTE) for superficial squamous cell carcinoma (SSCC) compared to magnifying endoscopy (ME) under narrow-band imaging. Participants underwent endoscopic examination, and images of pharyngeal and esophageal SCCs, as along with suspicious SSCC lesions, were collected using UTE and ME on the same day. Three image catalogs (UTE, ME-1, and ME-2) were created and reviewed by three expert endoscopists. ME-1 and ME-2 contained the same endoscopic images. The primary endpoint was the intra-observer agreement for diagnosing SCC. Eighty-six lesions (SCC = thirty-nine, non-SCC = forty-seven) in 43 participants were identified. The kappa values for the intra-observer agreement between UTE and ME-1 vs. the control (ME-1 vs. ME-2) were 0.74 vs. 0.84, 0.63 vs. 0.76, and 0.79 vs. 0.88, respectively. The accuracies for diagnosing SCC by UTE and ME-1 were 87.2% vs. 86.0%, 78.0% vs. 73,2%, and 75.6 vs. 82.6%, respectively, with no significant differences (p > 0.05). The rates of lesions that were diagnosed with confidence by UTE and ME-1 were 30.2% vs. 27.9%, 55.8% vs. 62.8%, and 58.1% vs. 55.8%, respectively. UTE demonstrates substantial diagnostic performance for SSCC in the pharynx and esophagus.

We developed a computer-aided detection (CADe) system to detect and localize colorectal lesions by modifying You-Only-Look-Once version 3 (YOLO v3) and evaluated its performance in two different settings. The test dataset was obtained from 20 randomly selected patients who underwent endoscopic resection for 69 colorectal lesions at the Jikei University Hospital between June 2017 and February 2018. First, we evaluated the diagnostic performances using still images randomly and automatically extracted from video recordings of the entire endoscopic procedure at intervals of 5 s, without eliminating poor quality images. Second, the latency of lesion detection by the CADe system from the initial appearance of lesions was investigated by reviewing the videos. A total of 6531 images, including 662 images with a lesion, were studied in the image-based analysis. The AUC, sensitivity, specificity, positive predictive value, negative predictive value, and accuracy were 0.983, 94.6%, 95.2%, 68.8%, 99.4%, and 95.1%, respectively. The median time for detecting colorectal lesions measured in the lesion-based analysis was 0.67 s. In conclusion, we proved that the originally developed CADe system based on YOLO v3 could accurately and instantaneously detect colorectal lesions using the test dataset obtained from videos, mitigating operator selection biases.

The cumulative metastasis rate of esophageal squamous cell carcinoma (ESCC) pathologically invading the muscularis mucosae (pT1a-MM), based on lymphovascular invasion (LVI) evaluated by immunohistochemical (IHC) staining is unknown. This retrospective study included patients with endoscopically resected pT1a-MM ESCC. The primary endpoint was the metastasis rate of pT1a-MM based on LVI, evaluated using IHC and additional prophylactic therapy. The secondary endpoint was the identification of independent factors for metastasis based on lesion characteristics. The prognosis was also analyzed considering the impact of head and neck cancer. A total of 104 patients were analyzed, with a median follow-up of 74 months. The positive rate for LVI was 43.3% (45/104). In 33 patients, IHC was not performed at the time of clinical evaluation, 8 of whom exhibited LVI. However, these patients did not exhibit metastasis. The metastasis rates of patients without LVI, those with LVI and additional therapy, and those with LVI without additional therapy were 5.1%, 20.8%, and 0%, respectively. Lesion size ≥ 25 mm was the only independent factor for metastasis in multivariate analysis. The advantage of IHC for determining additional prophylactic therapy is limited for patients with pT1a-MM ESCC.

Background: The utility of contrast-enhanced harmonic endoscopic ultrasonography (CH-EUS) alone in the biliopancreatic region appears to be limited because it is highly dependent on the experience and skill of the endoscopist. Therefore, the present study aimed to validate the efficacy of CH-EUS in clinical practice. Methods: Between January 2018 and March 2019, 301 consecutive patients who underwent CH-EUS were prospectively enrolled in this study. The diagnostic performance of CH-EUS was compared with that of dynamic computed tomography (CT), magnetic resonance imaging (MRI), and all combinations (i.e., CH-EUS, dynamic CT, and MRI) using a Bonferroni correction. A multiple logistic regression analysis was performed to extract each disease that allowed the CH-EUS diagnosis to be consistent with the final diagnosis. Results: In multiple comparisons of diagnostic performance, no significant differences were observed among dynamic CT, MRI, and CH-EUS (p = 1.00), but the diagnostic performance was significantly higher when all modalities were combined (p < 0.001). Moreover, only intraductal papillary mucinous neoplasm comprising adenoma or carcinoma (IPMN, n = 161) showed significance with respect to the agreement with the final diagnosis (p = 0.006). Conclusions: Our results showed that CH-EUS-based diagnosis of IPMN may be possible in clinical practice. On the contrary, to accurately diagnose biliopancreatic diseases other than IPMN, comprehensive diagnosis using multiple modalities may be necessary, rather than relying on CH-EUS alone.