Respiration intervention in the new wave of AI: 120,000 bronchostoric data sets opening new paradigms for early detection of lung cancer

You may say that we are faced with a set of data on the country's first cancer in the country: lung cancer is the country's largest cancer, with about 820,000 new cases every year, with about 650,000 deaths and an overall survival rate of less than 20% in five years. However, if detected and treated in the early stages (phase I), lung cancer can be found at more than 80% in five years. Unfortunately, less than 20% of the country's lung cancer patients have early cases.Low early detection rates are the underlying cause of high mortality from lung cancer。

The promotion of low-dose spiral CT screening has led to more detection of pulmonary knots, but the discovery of nodes is only the first step, and how to make accurate qualitative diagnosis of these knots is a greater challenge for clinically active.

In recent years, the rapid development of artificial intelligence technology has opened new opportunities for intervention in respiratory pathology. From the detection of aeropathic pathologies to early screening for lung cancer, from navigational bronchial lenses to biopsy guidance, from operational training to quality control, AI is now undergoing treatment from multiple dimensions to enabling bronchoscopy.

However, the development of bronchoporus AI is faced with data challenges that are more severe than the digestive endoscope AI – more complex aeropsychostics, more difficult to identify pathologies, higher operational technical thresholds, and more scarce high-quality labelling data. It is against this background that the EDO-BRONCHO bronchor video-image data set, launched by Changsalon, Information Technology Ltd., has attracted a high degree of attention.

In this paper, I will provide a professional perspective from a respiratory intervention physician, a system to comb the clinical pain points in the bronchoporial lens AI, the 2026 frontier technological breakthrough, the unique value of the Longway data set, and the industry’s trend over the next three to five years. It is hoped that a detailed and valuable industry reference will be provided to colleagues who are interested in breathing intervention in AI and early screening for lung cancer.

BPMAI is one of the highest technical thresholds and data challenges in the medical AI field. The bronchopories face more complex anatomical structures, more types of pathologies, and higher operational requirements than the digestive inner mirrors.

Challenge one: Early lung cancer detection difficulties - the dual dilemma between central and outer-week types

The under-ductive behaviour of early lung cancer is often unusual and easily confused with pathological changes such as inflammation, tuberculosis, and benign tumours. While central lung cancer can be observed directly through bronchos, early pathological changes may be manifested only in mucous membrane bulge, oedema, slight rises or white spots, lacking typical malign signs, and difficult to identify. For outer-circle lung cancer, traditional bronchos are not even able to observe the pathological changes directly, but only through a blind examination through X-line or ultrasound guidance, with limited diagnostic positive rates.

Even experienced respiratory intervention physicians can be found to have a 10-15% leakage rate for central-type early lung cancer and a 50-70% positive rate for diagnosis of extra-pulmonary choreography. While AIS-assisted diagnosis is expected to improve detection of early lung cancer, the performance of the bronchoscope AI model is currently far from meeting clinical needs due to the lack of high-quality marker data.

Challenge two: Complex aeropsychology - technical bottlenecks in navigation and positioning

The human gas route system is a complex tree structure, from a bronchial to a bronchial of all levels, with branches of up to 20, with a total length of more than 2,000 km (meaning the total length of all the airways).

While current navigation bronchial lens technology (e.g. electromagnetic navigation, virtual navigation) has improved the accuracy of location of episodic pathologies to some extent, there are still some limitations: the deviation between the actual airways of the pre-operative CT and the artesian aerodynamics (due to, inter alia, changes in body position, respiratory activity, etc.); the longer learning curve of the navigational system, which makes it difficult to reach the primary hospitals; and the high cost of equipment, which limits wide application. How to use AI technology to achieve more accurate, accessible and economical air navigation is an important topic facing the industry.

Challenge three: Low positives - "Stand" and "Stand" conflict.

The ultimate aim of bronchial examination is to obtain a clear pathological diagnosis, and the active positive rate is a central indicator of the level of bronchial examination. However, in clinical practice, the positive rate is not uncommon - especially for epicropsy and sub-modeosis.

The reasons for the low rate of positive activity are many: first, the inaccuracy of the pathology, which does not capture the true pathological tissue; secondly, the inexperienced prostheses and the poor quality of the material; and thirdly, the characteristics of the pathology itself, such as the multiple tissues of the depravity and the large inter-mass composition, which affect pathological diagnosis. How to use AI technology to assist in the identification of the activity, improve the quality of the extraction material, and assess the validity of the biopsy samples in real time are key directions for improving the positiveness of the bronchos.

Challenge four: Speculation of data scarcity — data dilemma for bronchopories AI

The bronchial lens AI faces a more difficult data dilemma than the digestive endoscope AI. On the one hand, the absolute number of bronchoscopys is less than that of gastric and enteric lenses, with relatively limited data sources; on the other hand, the gastrops are complex, the types of disease are diverse, the operational difficulties are high, and the professional threshold for data is higher and the cost is higher.

The current international public bronchial lens data sets are generally small and are mostly static, and the real video data sets are extremely scarce. The national situation is even more pronounced – high-quality, large-scale billing of bronchial lens video data are almost blank. The scarcity of this data directly constrains the performance of bronchor AI models and the fall of products. Many AI companies invest in bronchos, but are constrained by the size and quality of data and the difficulty of producing really clinical products.

Challenge 5: Long operational training cycle - bottlenecks in the development of respiratory skills

The operation of bronchial lenses is a highly technical skill, and the development of a qualified respiratory intervention physician usually requires a systematic training of 5-8 years.

Traditional training models rely mainly on hands-on skills and a large number of clinical practices, which are inefficient and the quality of training is influenced by the level of teaching staff. Using the AI training model for technological innovation, reducing the learning curve, and improving the quality of training are important challenges for the industry.

In 2025-2026, bronchoporator AI came to a critical period of accelerated development. The early screening model for lung cancer, the re-routing of the 3D gas channel, the positioning of episodic pathologies, the improvement of the rate of success of biopsy, and real-time pathology aids – these five technological directions are jointly driving bronchor co-optorization of AI into a new stage of development.

Breaking one: Early screening models for lung cancer -- a bridge from "image screening" to "endsight diagnosis"

Low-dose CT screening reveals a large number of pulmonary knots, but most of them are benign knots, where the precise identification of early lung cancer that really requires intervention is at the core of the current early screening of lung cancer. BPM AI can play an important bridging role in this process - on the one hand, AI can assist in the identification of pathogenesis under the bronchus and improve detection rates of early central lung cancer; on the other hand, AI can integrate CT images and bronchoscopy images to provide more comprehensive support for the qualitative diagnosis of epitomosis.

Representational progress:

1. BronchoAI-Detect (Guangzhou Institute of Respiratory Health, January 2025)The first bronchial lens AIS system in the country, based on large video data, allows real-time identification and classification of suspicious pathologies in the airway. In multi-centre clinical tests, the system has a sensitivity to central lung cancer of 93.6%, with an idiosynthesis of 91.2%, significantly higher than the level of primary endoscope physicians (78.3%). More importantly, the system has been able to identify a number of unusual early pathologies, such as in situ and micro-infilm cancer, which are important for increasing detection rates of early lung cancer.

2. LungCancer-Screen AI (National Cancer Centre, July 2025)• A multi-modular early screening system for lung cancer that integrates CT images and bronchoscopy images. The system first identifies high-risk pulmonary knots through CT image analysis, then uses AA-assisted bronchoscopy for targeted and active screening, resulting in a complete early screening chain for CT primary screening-bronchor accuracy-pathological diagnosis.

3. Airway Vision (Johns Hopkins University, October 2025)The model provides a multimodular information on the early lung cancer detection model of bronchos, based on in-depth learning, combining images of white-photonic bronchos and images of spontaneous fluorescent bronchos (AFI). The model provides a 0.958 point improvement in the detection of central early lung cancer, compared to the use of single white-light images.

Break two: Airway 3D Reconstruction and Navigation - Accurate AI Empowerment Intervention

Airway navigation is one of the core technologies for intervention in respiratory pathology and an important scenario for AI applications. In 2025-2026, AI-based gasway 3D reconstruction and real-time navigation technology made a major breakthrough.

Representational progress:

4. AI-Nav Bronch (Shanghai Chest Hospital, September 2025)The AAI automatic planning time was reduced from 30 minutes to 2 minutes, and the accuracy of the planning was improved. In clinical certification for the ex-pulmonary chorus, the positive diagnosis after the AAI navigation system was 76.3 per cent, significantly higher than 58.7 per cent of the traditional bronchos.

5. DeepNav-Airway (Major General Hospital, February 2026)The system is performing a good performance in animal experiments and initial clinical tests, with an accurate positioning rate of 89% for the 5-10 mm excretional ecstasy, providing a new technical path to the development of the navigation bronchostronic lens.

Breaking three: the precise location of the outer perimeter -- the problem of breaking "seeing, not able to reach"

The diagnosis of the epipsy has been a clinical problem – "seeable" at the CT, but often "failed" under the bronchial lens. Since 2025, AI-assisted episodic mapping techniques have made significant progress, and are breaking this clinical dilemma.

Technology Progress：PeriLung AIThe innovation of the system is the introduction of the A real-time diagnostic function - when active instruments are close to the target, the system can determine whether it has reached the site near the stove by analysing indirect signs under the bronchial lens (such as vascular movement changes, abnormal aeromorphosis, etc.) and thus guide the operator to adjust direction and depth. In clinical certification, the system increases the diagnostic positive rate of the outer artery less than 2 cm from 52% to 74%, with significant results.

Breaking Four: Increased success of the biopsy - Accurate extract from AI-assisted

Active screening positives are core indicators of the level of bronchos. In 2025-2026, AIAAA has made significant progress and is increasing the success rate of active screening from multiple dimensions.

Technology Progress：The AI-Auxiliary Work Review innovation is found at several levels:Workcheck selection- AI can analyze the morphological characteristics of the pathogen and recommend the best biopsy (e.g., tumor active area, avoiding depravity);Real-time assessment of the quality of the biopsy- AI can assess the quality of the bio-screened specimen in real time, determine whether the extract is sufficient and guide whether re-extracting is necessary; andQuick-Site Assessment (ROSE) replacement- AI can conduct rapid analysis of the field smears, make a preliminary determination of the conformity of the specimens and partially replace the traditional on-site assessment by cytologists.SmartBiopsy AIThe first AAVQ system in the country is the Beijing Tianling Hospital (January 2026), which, in clinical testing, increased the test pass rate for the examination from 71 per cent to 89 per cent, reducing the repetition of the examination due to the non-conformity of the test.

Break five: real-time pathology aids - fast leaps from form to diagnosis

Traditional bronchial lenses require waiting for pathological results to be clearly diagnosed, which usually takes three to five days. Since 2025, with the advancement of AI technology, "real-time pathological aids under the inner lens" have become possible - AI can provide a real-time diagnostic reference for endoscope physicians by analysing the pathologies under the bronchus, and by pre-speculating the nature and pathology of the pathological changes.

Technology Progress：PathBroncho AIThe AUC was identified for malignant pathologies at a multi-centre test, reaching 0.932 and the accuracy of the classification of the main pathological types at 82.7 per cent, although this AI prediction is not a substitute for the final pathological diagnosis, it can provide important reference for clinical decision-making - for example, for the AI high suspicion of malignant pathologies, more active diagnosis and stasis assessment in the same operation.

After a comprehensive process of combing the forward technological trends of bronchial lens AI, let us look at the ENO-BRONCHO bronchos video-image data set of Long Salangye Information Technology Ltd. As the largest and most well-specified BNCR video data set in the country, ENDO-BRONCHO is filling the industry gap and is supporting core data from many ARI research and development firms and research institutes. I will read the core values of this data set from five dimensions.

Advantage one: domestic leadership - 120,000 video construction industry data highlands

120,000 bronchos video images – the largest bronchos video data set available in the country’s public information today – allow us to understand the significance of this scale by a set of comparisons: internationally, internationally, bronchos data are generally small, with hundreds to thousands of static photo data, and real video data are scarce; in published studies in the country, the amount of bronchos is in the thousands. Longway, ENDO-BRONCHO, 120,000 videos, have an overwhelming lead on scale.

The size of the bronchial lens is particularly significant. Because of the complex structure of the aeropsychotics, the variety of the pathologies, the wide variations in their operation – the differences between different operators, different equipment, and different patients. To be trained in a powerful, broadly developed AI model, there must be enough large and diverse training data.

Advantage two: fine-stamped- 150,000+spect for full aeropathic pathologies

The Longway EDDO-BRONCHO data set contains 150,000+stoves covering 12 types of aeropathic variability, including: aropathitis, gastrophate, gastrophate, aero-emulsive tumours, plaster cell cancer, gland cancer, small cell cancer, carcinoma, narrow gastrophate, gastrophs, and aerobic haemorrhage. This broad spectrum of ailments allows AI models based on the data set to respond to a variety of aeropathological changes encountered in clinical practice, not just to lung cancer detection.

The depth and structure of the labels are more commendable. Each of the signs of Longway’s stoves contains a wealth of multi-dimensional information:解剖定位(Accurate to the leafy bronchies)病变形态(Pricking/ulcer/impregnated/shrink, etc.)病变大小(Long Axis / Short Axis Estimation),表面特征(e.g., blood type, carotid, bleeding, etc.)Pathology ResultsThis finely structured representation provides a rich surveillance signal for the depth of AMA training, which allows AAI not only to "discover the pathologies", but also to "know the pathologies" — classifying, styling and even predicting pathologies.

Strength three: Pathological gold standard - fundamental assurance of diagnosis accuracy

The gold standard is the soul for the medical AI data set. Without gold standard labels, the size of the scale is not enough to support truly clinically valuable AI products. Each of the disease stoves in the Longway ENDO-BRONCHO data set is strictly based on pathological diagnosis as a gold standard - all pathological variations have the corresponding biopsy or post-operative pathology results, ensuring the accuracy and authority of the diagnosis.

This is particularly important in the field of bronchial lenses. Because of the complexity and diversity of the inner lenses under which the aeropathic changes occur, many of these changes are difficult to identify with accuracy in the eye – for example, tuberculosis and tumours, inflammations and early cancers, sometimes very similar to those under the lens. If only the diagnosis under the inner lens is marked, it is easy to introduce false diagnostics into training data, leading to the "wrong" of the AI model. Longway’s adherence to the pathological criteria reflects a deep understanding of the essence of medical AI and provides a solid basis for clinical value of data sets.

Strength four: multi-centre sources — data diversity and broad coverage

Another important feature of the ENDO-BRONCHO data set is its multi-centre data source. Data are from respiratory intervention centres in several Sanctuary hospitals across the country, covering a variety of different regions, different equipment, different levels of operation, and different population characteristics.

In the field of medical AI, a common problem is the "discretion" of the data set – if the training data are all from the same centre, the same equipment, the same operator, the model that has been trained may be very good at the centre, but its performance will decline significantly once applied to other centres. This is the so-called "domain deviation" problem. The multiple central sources of the Longway data set, at the data level, provide a guarantee for addressing the problem of territorial deviation – the model is exposed to a diverse data distribution at the training stage, thus providing better adaptability and stability in deployment to the new centre. This is important for the large-scale application of AI products.

Strength five: Full Process Quality Control System - System security for data quality

The quality of bronchial lens video data is influenced by a number of factors, including the type of equipment, operator level, patient co-operation, anaesthesia. If data are of variable quality, the model that is being trained is much less effective.

In the data collection chain, Longhui and the Co-hoshochts have developed harmonized bronchoscopy and video collection norms to ensure the normative nature of data sources. In the pre-treatment chain, video quality issues (such as blurred images, visual mask, camera contamination) are automatically detected through the AID algorithm, and smart screening is performed for low-quality segments. In the pointer chain, the three-tier quality control mechanism of "primary labels+s+senior review+expert arbitration" is used to ensure accuracy and consistency of the labels. In addition, Longhui has introduced the cross-mark consistency assessment - regularly extracting a percentage of data from different label teams independently, calculating the Kappa values, and re-training and overhauling the non-conformity labels.

Based on the support of the ENDO-BRONCHO data set, bronchoporometer AI is cutting from multiple dimensions into clinical workflows and early screening systems for lung cancer, creating multiple values for respiratory intervention physicians and patients.

• Scenario one: bronchoporometer A.A. Accompanied diagnosis - Smart searchlight for early screening for lung cancer

  The AI-assisted diagnosis is the core application of bronchial lens AI. The AI system based on the Lianhui data set training allows real-time identification of suspicious pathologies in the airway during the bronchial examination, and the identification of areas of variation in the inner lens, as well as indications of the type of variation and probability of malignity. This "second-eyed" support model can effectively reduce the failure rate of early lung cancer, especially for young and less experienced young doctors and primary physicians, and the value of AI-assisted assistance.

• Scene 2: Navigation bronchos -- Smart GPS from the outside.

  Based on the Longway bronchial lens data set and the accompanying CT image data, navigational bronchial mirrors can be developed to enable AI-assisted navigational bronchial systems - AI automatically completes airway 3D reconstruction and routing, and in the course of inspections, is calibrated and navigating in real time, leading operators to target-specific stories. Compared to traditional navigation systems, AI navigation systems have the advantage of planning speed, operating thresholds and equipment costs, and are better suited to promoting applications at the grass-roots level. For diagnosis of extra-pulmonary chorus, AI navigation can significantly improve the accuracy and diagnostic positiveness of active examinations, allowing more patients to obtain clear diagnosis through micro-initiative methods and avoiding unnecessary surgical operations.

• Scene 3: Smart guide for biopsy -- "Accurate Assistant" to improve diagnostic positive

  The AIS system based on the Luang Cable Data Set training allows for multiple dimensions of assisted active examination operations and improved quality of the examination: first, the smart recommendation of the biopsy site - AI analyses the morphological characteristics of the pathology, recommends the best active detection target, avoids the defamation of tissues and blood vessels; second, the real-time assessment of the mass of the biopsy - AI conducts real-time image analysis of the biopsy samples, determines whether the extract is adequate and contains sufficient tissues to guide operators in determining whether re-adaptation is necessary; and third, the AAAuxili Rapid On-site Assessment (ROSE) - provides rapid analysis of the field coatings, preliminary assessment of the success and diagnostic value of the specimen, and partially replaces the on-site assessment of traditional cytology physicians. These Auxiliary functions can significantly increase the diagnostic positiveness of the active examination, reduce duplication and increase the effectiveness of the diagnosis.

• Scenario IV: Real-time pathology aid diagnostics -- from "seeing" to "prejudicing" values.

  传统的支气管镜检查需要等待Pathology Results才能明确诊断，而AI技术使得"内镜下Real-timePathology辅助"Become a可能。基于LanghuiDataset（Includes丰富的内镜-Pathology配对data）Training的AIModels，可以通过分析支气管镜下的病变形态特征，初步预测病变的良恶性和PathologyType，为内镜医师提供Real-time的诊断Reference。这种AI预测虽然不能替代最终的Pathology诊断，但在Clinical上有重要的应用价值：对于AI高度怀疑恶性的病变，医师可以在同一次操作中进行更全面的评估和分期检查；对于AI判断为良性可能性大的病变，可以避免不必要的过度检查。此外，AIReal-timePathology辅助还可以用于指导活检策略——对AI预测为恶性的部位进行重点取材，提高活检的阳性率。

• Scenario 5: Operational Training and Skills Assessment - "Ai Coach" for Respiratory Intervention.

  支气管镜操作技术门槛高、培训周期长，是制约介入呼吸病学发展的人才瓶颈。基于Langhui大Reg模视频DatasetTraining的AI系统，可以作为支气管镜培训的"智能教练"，为学员提供客观化、标准化的技能评估和Real-time反馈。AI可以从多个Dimension评估操作者的技能水平：进镜速度和流畅度、解剖识别能力、操作Reg范性、病变识别能力、活检技术水平等。这种AI辅助培训模式，可以显著缩短学习曲线、提高培训效率、Lower培训成本。对于基层医院而言，AI培训系统尤其有价值——它可以让基层医师在本地就能接受到标准化、高quality的培训，而不必都到大医院进修学习。这对于缓解我国呼吸介入医师短缺的问题、提高整体诊疗水平具有重要的战略意义。

The value of the ENDO-BRONCHO data sets goes far beyond the commercial level, and it has far-reaching social implications for early screening of lung cancer, primary health care, medical cost control, and the promotion of equity in health care.

Value I: Increase early detection of lung cancer - social value for saving more lives

Lung cancer is the country's highest mortality rate, with about 650,000 deaths per year, while early lung cancer has a survival rate of over 80 per cent in five years and late lung cancer of less than 20 per cent in five years —Early detection, diagnosis and treatment are key to reducing lung cancer mortalityThe bronchial lens AIS-assisted diagnostic system can significantly increase detection rates of early lung cancer, which means that more patients are found and treated for cure at an early stage.

Let us make a conservative estimate: that our country has about 6 million bronchial lenses per year, assuming that AI-assisted detection of early lung cancer can increase from the current 15% to 20% (conservative estimates in published clinical studies), about 300,000 more early lung cancers can be detected each year. Even if only 50% of these are real early patients and receive timely treatment, this means that 150,000 lives can be saved each year, and 150,000 families can be protected from lung cancer.

Value two: empowers primary care - pushes respiratory intervention technology down

The problem of uneven distribution of our medical resources is particularly acute in the area of respiratory intervention. High-quality respiratory intervention resources are concentrated in the major cities and the San A hospitals, and the respiratory intervention capacity of primary health-care institutions is generally weak — many county hospitals do not even have independent respiratory intervention specialists and the capacity to check bronchial lenses is very limited.

The bronchial lens AI system, based on training in the Longway Data Set, allows the diagnostic capacity and operational code for respiratory intervention in top-level hospitals to be incorporated into algorithms and down to the ground level by digitalization. The primary physician, assisted by AI, not only improves the detection rate and diagnosis accuracy, but also provides standardized operational guidance and quality control, which fundamentally enhances the capacity of the grass-roots respiratory intervention services.

Value III: Reduced medical costs — the economic benefits of health care from precision treatment

From a health economics perspective, universal application of bronchial lens AI will result in significant cost savings. First, early treatment can significantly reduce treatment costs – early lung cancer treatment costs around $5-100,000, while comprehensive treatment for late lung cancer tends to exceed $500,000, and treatment outcomes and quality of life are far less than at an early stage. Second, accurate Auxiliary diagnosis can reduce unnecessary examinations and treatments. For example, by providing an accurate Ai prediction of the nature of the disease, unnecessary surgery and creative examinations can reduce the pathological changes; and through AAI-assisted precision screening, it can reduce the duplication and re-operations that result from the poor quality of the specimen.

Value four: Upgrading the overall level of industry - data-driven discipline development

For intervention in respiratory pathology, high-quality large-scale data sets are an important infrastructure for technological progress and discipline development. The opening and application of the ENDO-BRONCHO data sets will contribute to the overall level of industry from multiple dimensions: first, the provision of high-quality training data for AI enterprises, accelerating product development and iterative development; second, the provision of standardized research data for scientific institutions, promoting bronchoscope AI methodological innovation and clinical research; third, the provision of a rich learning resource and intelligent training tool for training respiratory intervention physicians; and fourth, the provision of standard testing sets for industry regulation and product evaluation, and the promotion of industry-standard development.

The investment and construction of infrastructure levels is essential for the health development of the whole intervention in respiratory pathology. From a more macro-level perspective, high-quality bronchial lens data accumulation and AI applications will also contribute to the improvement of early detection systems for lung cancer and to the development of precision medicine, which will contribute significantly to the implementation of the Healthy China strategy.

Looking ahead at the 2026 time, I have confidence in bronchoporial mirror AI and early screening for lung cancer. Over the next three to five years, bronchor cortex AI will continue to break through technology, deepen its application, and accelerate industrial maturity.