P0371 Building a Robust Artificial Intelligence Solution for Use in Ulcerative Colitis Clinical Trials
Bibliographic record
Abstract
Abstract Background Artificial Intelligence (AI) is increasingly used to assess Ulcerative Colitis (UC) disease activity in clinical trials, with the goal of matching or exceeding expert performance in a reproducible manner. Certai has been created from previous AI work as a model that conforms to the new definitions within the Modified Mayo Endoscopic Score criteria. Methods Using the self-supervised DINOv2 method, Certai was pre-trained on 845 colonoscopic procedures and then refined with annotations from the proprietary Software for Intelligent Annotation (SIA) platform. This browser-based tool, with advanced playback controls and a UC scoring interface, enabled video annotation by eight global IBD specialists and central readers, and seven additional specialists in training on SIA. Labellers underwent rigorous onboarding to meet minimum thresholds for Intra-Class Correlation Coefficient (ICC) and Quadratic Weighted Kappa (QWK). Certai’s architecture features two Vision Transformer models: a Quality Control (QC) model and a Scoring model. The QC model excludes frames with poor clarity, inadequate bowel prep, non-colonic views, chromoendoscopy, or biopsy procedures. The Scoring model uses multi-headed outputs to assess UC severity, grading vascular pattern, bleeding, ulcers/erosions, friability, and erythema. Results A total of 8.9 million frames from 39 videos were labelled across six categories, resulting in 2.17 million merged labels determined by majority vote. For inter-rater agreement on modified MES in the labelling process, the overall ICC among the onboarded labellers was 0.86, and the QWK was 0.88. On a validation set of colonoscopy procedures, Certai achieved 100% agreement with human central readers on modified MES scores. The ICC among three expert labellers was 0.922 and rose to 0.942 with the addition of Certai. There was an additional increase of the ICC to 0.955 when Certai was paired with just one expert labeller. Similarly, QWK scores rose from 0.914 for two expert labellers to 0.961 with Certai and one expert labeller. Conclusion Certai represents an advance in UC disease activity assessment, meeting modified MES requirements. With further specialist labelling of an additional several hundred videos at the detailed frame level, a robust version of Certai will enable new quality standards in speed of central reading and consistency. Future applications may include stand-alone AI reads with human sign-off or a 2 + 1 reader model incorporating AI as one reader.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.019 | 0.052 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.003 | 0.001 |
| Open science | 0.002 | 0.003 |
| Research integrity | 0.002 | 0.002 |
| Insufficient payload (model declined to judge) | 0.010 | 0.004 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".