Abstract
BACKGROUND & AIMS: Barrett’s epithelium measurement using widely accepted Prague C&M criteria is highly operator dependent. By reconstructing the surface of the Barrett’s area in 3D from endoscopy video, we propose a novel methodology for measuring the C&M score automatically. This 3D reconstruction provides an extended field of view and also allows to precisely quantify the Barrett’s area including islands. We aim to assess the accuracy of the extracted measurements from phantom and demonstrate their clinical usability. METHODS: Advanced deep learning techniques are utilised to design estimators for depth and camera pose required to map standard endoscopy video to a 3D surface model. By segmenting the Barrett’s area and locating the position of the gastro-oesophageal junction (GEJ) we measure C&M scores and the Barrett’s oesophagus areas (BOA). Experiments using a purpose-built 3D printed oesophagus phantom and high-definition video from 98 patients scored by an expert endoscopist are used for validation. RESULTS: Endoscopic phantom video data demonstrated a 95 % accuracy with a marginal ± 1.8 mm average deviation for C&M and island measurements, while for BOA we achieved nearly 93 % accuracy with only ± 1.1 cm² average deviation compared to the ground-truth measurements. On patient data, the C&M measurements provided by our system concord with the reference provided by expert upper GI endoscopists. CONCLUSIONS: The proposed methodology is suitable for extracting Prague C&M scores automatically with a high degree of accuracy. Providing an accurate measurement of the entire Barrett’s area provides new opportunities for risk stratification and the assessment of therapy response.
Introduction: The Challenge of Quantifying Barrett’s Oesophagus
Barrett’s oesophagus (BE) is a significant precancerous condition where the normal squamous lining of the esophagus is replaced by columnar epithelium due to chronic gastro-oesophageal reflux. Accurate monitoring is crucial for cancer risk stratification. Currently, clinicians rely on the "Prague C&M" criteria, which measure the circumferential (C) and maximal (M) extent of the affected area relative to the gastro-oesophageal junction (GEJ).
However, manual measurement is notoriously subjective and prone to inter-observer variability. Endoscopists must estimate lengths by observing markings on the endoscope handle while navigating a dynamic, non-rigid environment. Furthermore, standard Prague criteria often overlook "islands" of Barrett’s tissue, which can still harbor malignancy. There is a clear clinical need for an objective, automated system to quantify the total Barrett’s Oesophageal Area (BOA) and track subtle changes over time.
AI-Driven 3D Surface Reconstruction
To address these limitations, researchers developed a real-time AI pipeline that transforms standard 2D endoscopy video into a 3D surface model. The system utilizes several advanced deep learning components:
- Depth and Pose Estimation: A Feature Pyramid Network with a ResNeXt-101 backbone predicts the distance from the camera to the esophageal wall. Deformable convolutions are employed to handle the complex, non-rigid geometry of the esophagus.
- Real-Time Segmentation: A lightweight neural network identifies Barrett’s tissue and the GEJ "starting line" in real-time.
- 3D Mosaicking and Quantification: By combining depth, pose, and segmentation, the system builds an extended 3D field of view. Shape-fitting algorithms then automatically calculate C and M scores, as well as the total BOA. The system can be calibrated using the known diameter of biopsy forceps (2.8 mm) to provide measurements in millimeters.
Validation and Clinical Performance
The system was validated using a three-tiered approach: synthetic data for initial training, a 3D-printed esophagus phantom for physical ground-truth validation, and a clinical cohort of 98 patients.
On the phantom, the system achieved 95% accuracy for C&M measurements (±1.8 mm deviation) and 93% accuracy for total area (±1.1 cm² deviation). In the clinical study, the AI’s measurements showed strong concordance with expert endoscopist assessments (Spearman correlation > 0.90).
Key insights from the clinical data include:
- Consistency: The system demonstrated high reproducibility across consecutive endoscopy visits, as shown in the high correlation between visit-1 and visit-2 measurements.
- Therapeutic Monitoring: The AI successfully quantified tissue reduction in patients undergoing ablation therapy. In some cases, while C and M scores dropped to zero, the system still detected significant residual "islands" of tissue (over 26 cm²), highlighting its superior sensitivity compared to manual scoring.
- Real-Time Utility: Operating at approximately 27 frames per second, the system provides live feedback, potentially acting as a "virtual biopsy" map to document precise sampling locations.
Conclusion and Future Outlook
This AI-driven 3D reconstruction methodology provides a robust, objective alternative to manual Prague C&M scoring. By automating measurements and quantifying the entire Barrett’s area, it offers new opportunities for precise risk stratification and monitoring of therapeutic responses. While this single-center pilot study is promising, future work will focus on multi-center validation and integrating these tools into "smart" endoscopes to ensure high-quality data acquisition across diverse clinical settings.
Summary updated