Questions about DovaVision UC?

DovaVision UC is an AI-driven clinical research platform that standardizes endoscopic assessment of ulcerative colitis (UC) from colonoscopy videos. It provides objective, reproducible disease severity scoring to support trials, clinical practice, and training. The platform delivers high-resolution, frame-level analysis of mucosal healing and disease severity, going beyond traditional video-level Mayo Endoscopic Scores (MES).

Unlike traditional MES scoring that provides a single summary score per video, DovaVision UC offers multi-layered analysis:

• Frame-level severity to show disease variability across the colon

• Pixel-level characteristics identifying specific markers like ulcers, erosions, erythema, and loss of vascular pattern

This enables a more granular, objective, and interpretable assessment of disease activity.

DovaVision UC offers a three-layered explainability model to avoid the 'black box' problem in AI.  

1. Video-level: Overall MES-like score summarizing procedure severity

2. Frame-level: Classification of each frame into High, Medium, Low, Normal, or Unscorable severity

3. Pixel-level: Identification of specific disease features driving each frame's classification

The layered outputs from video-level scores down to pixel-level feature identification, the clear mapping to disease characteristics and the severity classifications reveal the rationale behind every output, ensuring transparency and trust in the analysis and providing a clear, interpretable tool that clinicians and researchers can trust and validate.

• Improved data quality and consistency across sites and readers

• Automated workflows to reduce manual scoring burdens

• Standardized scoring (Mayo, UCEIS) scalable across global studies

• Explainable outputs that align with regulatory expectations

• High ICC performance supporting reproducibility and reliability

• Detects subtle disease features that may be missed by humans

• Improves quality control in trial imaging datasets

Frame-level analysis is a key feature of DovaVision UC. It classifies each frame into High, Medium, Low, Normal, or Unscorable severity, enabling detailed tracking of disease progression and response to treatment over time. This approach captures subtle intra-procedural variations in inflammation that a single video-level score may miss, offering more actionable insights for research purposes.

• Standardized MES and frame-by-frame sub-score outputs per video

• UCEIS score

• Data export as JSON and burned-in heatmaps

• Built in Quality Control systems to ensure only scorable frames are assessed
• Built on a large-scale, self-supervised vision transformer, trained on millions of GI images
• Refined with expert-labeled datasets (multiple central readers per video, majority voting)
• Benchmarked against ICC/QWK standards

• Erythema
• Vascular patterns
• Bleeding severity
• Ulcers/erosions
• Friability
• Biopsy-related findings
• (All captured at frame level)

• Pre-trial: Patient screening, site calibration
• During trial: Data quality control, scoring consistency, reader augmentation
• Post-trial: Retrospective dataset analysis, regulatory reporting, publication support

• Secure cloud upload or API-based integration
• On average it takes about 15 minutes to analyze a procedure and output both a video overlay and JSON data. Outputs in standard formats (JSON, visual overlays, video with scoring burn-ins)
• Compatible with research platforms and trial databases
• Designed for rapid adoption with minimal site disruption

• Trial Phases: Supports Phase 1–3 and post-marketing studies
• Drug Development: Strengthens clinical endpoints with objective, surrogate scoring
• Retrospective Data: Enables re-analysis of legacy datasets for new insights
• Scalability: Automated analysis supports large, global multi-site trials

DovaVision UC’s scoring is derived from frame-level disease characteristics, which are aggregated to produce a comprehensive video-level score. This means it can:

• Show which disease features (e.g., ulcers, erythema) contributed to each frame's classification
• Presents interpretable severity categories across the full procedure
• Generate unique visual outputs like heatmaps and more