Introduction: Clinical deployment pole for enteroscope AI
In all the fields of medical endoscopes, intestinal mirrors AI is the most mature and fast-paced course of clinical deployment. Olympus Cadie has been awarded a price increase of 7.3% in the incidence of adenomas (ADR) in the RCT at the EAGLE multi-centre. This achievement marks the cross-section of intestinal lenses from "research" to "clinical markers."
In 2026, intestinal mirrors ABA continued to break down in real time detection and precision. PESNet achieved end-to-end delays of 12.6 ms/frames, meeting 40 ms of clinical budget.
The special value of intestinal cosmolar AI is "reduced leakage." In colon cancer screening, adenoma can be as high as 20-30%, and adenoma can grow into cancer in the years to come. The core value of the AIS system is not "discovery more" but "inside" — to minimize the incidence of leakage and to actually achieve the screening reduction goal.
Current state of the industry: FDA approval and multi-centre RCT certification of gold standards
The only area in the medical video AI that has reached the dual milestone of "FDA approval plus multi-centre RCT validation" is the intestinal mirror AI.
At the regulatory level, Olympus Cadie has been awarded FDA-clared and CE-Marced, the first real-time cloud-based CADE application to be approved through FDA.
On the evidence-based level, the EAGLE experiment published in December 2025 in the npj Digital Medicine, the most important multi-centre RCT in the field of enteroscope AI, included 841 patients, 22 goggles, showing an increase of 7.3 per cent in absolute values for the CADDIE team ADR; an increase of 93 per cent in the detection rate of the glandoma (>10mm), a 57 per cent increase in the incidence of non-salent adenoma and a 230 per cent increase in the SSL. These data not only demonstrate the overall effectiveness of AI, but also reveal the unique advantage of AI in the "probable disease" - the most susceptible to manual examination of parsemic and non-fatal disease, and the most significant increase in the detection rate of AI.
Technically, PESNet (Frontiers in 2025) achieved end-to-end delay of 1080p12.6 ± 0.3 ms/frames, split 4.4 ms, prompt integrated 0.6 ms, prototype search<0.2ms，远低于40ms临床预算。这一超低延迟使AI可以在不影响检查流程的情况下实时运行。StyleGAN增强训练（2025年npj Digital Medicine）使用合成超15万张结直肠新生物图像，将mAP从0.86提升至0.93，减少外部验证泛化差距。
On a market scale, the global market for colon cancer screening is expected to be over $3 billion in 2026, with AI-aided colon lenses being the fastest growing area. China’s incidence of colon cancer continues to rise, with huge demand for enteric lenses.
2026 Frontline breakthroughs: EAGLE test, PESNet and StyleGAN
The core breakthrough of the intestinal mirror AI for 2025-2026 is focused on three areas.
The first is the milestone of the EAGLE experiment. As the first intestinal lens I multi-centre RCT, EAGLE provides evidence of the highest level of medical evidence. 7.3% of the ADR absolute value increases seem modest, but at the population level it means thousands more adenomas per year, preventing hundreds of colon cancers. The increase in the incidence of gland cancers by 93% and the increase in SSL by 230% are more striking — these are the most easily missed disease-transformation types.
The second is a break in real-time partition. PESNet not only detects saliva, but also achieves precision partition (4.4 ms), providing a technical basis for guidance on the marking and removal of the saliva boundary in real-time. The study also reported a 26% reduction in the salivating saliva and a 15% reduction in the edge of residual tumors after cold removal, confirming AI’s value in the whole process of examination and treatment.
The third is enhanced generation data. SteileGAN is training in the YOLOv5 test model for the synthesis of over 150,000 colons. The breakthrough is from 0.86 to 0.93. This breakthrough addresses the imbalance in medical image data and the scarcity of rare pathologies, providing new pathways for model pantraction.
The 1 million-circle cortical mirror data sets of Longway technology provide a key support for these frontier studies. The data sets are fully documented for go-ins, reaching the deepest part, back-to-blind/end-end remix colon markings, retroscopy and colon sections, with intestinal preparation ratings, salivation size/form/spectification, removal of integrity and pathology validation data.
Longhui Tech data set: 1 million cases of systematization of intestinal mirror video
The Lianhui technology endoscope video-image data set, with a scale of 1 million cases, systematically covers the entire process of colonoscopy.
In terms of data collection, all images are derived from the cooperative three-acadet hospital digestive internal medicine and endoscopy centres. Full records are recorded for goggles, reaching the deepest part, back-to-backs/rest intestines (if arrived), backscans and colon sections. Retention equipment is originally exported for full screening videos and key original maps, and original resolution, frame ratio, coding, colour/enhanced mode and time sequences are maintained.
In the labelling system, each data includes structured fields such as intestinal preparation ratings (standardized ratings such as BBPS), entanglements, retrospectation, hologram size/form/spectation, removal of integrity, pathology and complications. The same chronology perspective is not counted as many cases to ensure the accuracy of the data count.
In terms of distribution, the pathological/positive check is defined as the subject; normal complete check, post-operative review, and limited quality check are organized separately.
In terms of quality control, the three-layer quality control process of the first bid for the ingestion physician + the deputy director and above, which is reviewed by the expert, and the consistency assessment + is used.
Data sets prioritize formal endoscopy reports, pathology, CT/MRI, treatment and follow-up data to support cross-modular learning and vertical change analysis.
Forward vision: from detection to full process support
The future development of intestinal cosmolar AI will go in three directions.
The first is "test" to "full process support." Currently AI is used primarily for salivating meat, and will be expanded to salivation (NICE/JNET), cut-off range guidance (AI labeling boundaries), removal integrity assessment (AI check residues) and blood loss risk predictions.
The second is from "single centres" to "multicenter extension." The EAGLE experiment has proved to be effective in multiple centres, but the quality of equipment, intestinal preparation, and differences in patient groups in different centres remain challenges.
The third is "aided diagnosis" to "quality control". AI not only supports diagnosis, but also serves as a quality control tool - auto-assessment of key quality control indicators such as intestinal preparation quality, retrospectation time, arrival range, etc., to provide data support for the quality management of the inner mirror centre.
In terms of industry competition, the first tier is the system with EAGLE-level RCT evidence and FDA approval. Longhuitech’s 1 million-case data sets provide a data base for high-quality models of training for chasers and innovators.
Conclusion: EAGLE experiment opens the era of the alignment of the AI mirror.
The intestinal mirror AI process, from concept to FDA approval and RTT validation at the EAGLE multi-centre, provides a model for clinicalization of medical images. 7.3% of ADRs are up, meaning that thousands of colon cancers can be prevented each year at the population level.
Long Salang Langhui Information Technology Ltd., with a 1 million case intestinal mirror video set as its anchor, is becoming a major data service provider in the field of enteroscope AI. At a critical time when CADIE has been approved by FDA and the EAGLE experiment has confirmed its validity, Long Huitech will continue to promote clinical coverage of straight-to-rest screening with high-quality data.
Clinical value of colon cosmorial AI and commercial path
The commercialization of colon coronary mirror AI led to a landmark breakthrough in 2026. The CADDIE system was approved by the FDA as the first AAD diagnostic facility for real-time detection of rectal entrails, and the multiple control tests at the EAGLE centre confirmed that AAA support can increase adenomas by about 10%, an increase that is of public health significance for reducing the incidence of and mortality from colonic cancer. These advances mark the formal entry of enteric cortical mirror AI from conceptual validation to clinical routine use.
The value of intestinal mirror AI is clear from the commercial path - reducing the incidence of leakage, upgrading ADRs, and reducing the cycle of specialist development. China’s incidence of colon cancer continues to rise, with more than 550,000 new cases occurring annually, and the demand for intestinal lens screening is huge but seriously inadequate. AI support can compensate to some extent for the shortage of endoscope physicians, while improving the quality of the examination.
From a data governance perspective, intestinal cosmopolitan training data need to cover both video screening, pathological reports, follow-up results and surgical records, creating a complete chain of evidence from screening to treatment to the prognosis. The Randhytech 1.7 million intestinal cosmosystem data sets were constructed with a rigorous cross-data-linking mechanism to ensure that each examination accurately corresponds to pathogens, treatment records and follow-up data. This deep-link data structure supports not only training meat detection models, but also the development of smart assistive systems from risk assessment to treatment decision-making, providing data-driven solutions for the whole management of colon cancer.