Multicentric evaluation of an artificial intelligence model to stratify stage II colon cancer patients from whole slide images.
Bibliographic record
Abstract
3612 Background: Stage II colon cancer (IIA T3N0; IIB T4aN0; IIC T4bN0) represents nearly 25% of all colon cancers and includes a wide range of outcomes (5-year overall survival rate (OS) of 58.4% to 87.5%). We aim to test if a deep-learning-based analysis of whole slide histology images (WSI) can predict survival and highlight the most relevant morphologic characteristics underlying prognosis. Methods: We identified adult stage II colon cancer cases from the multimodal Cancer Genome Atlas (TCGA) and retrospective consecutive cases diagnosed between 2014-2019 (inclusive) from two independent centers (CHUM, Canada and IHP, France; IRB approved). We tested on these two independent datasets, an artificial intelligence (AI) algorithm trained on TCGA, using a cross-validation and cross-testing (80:20) framework. The model relies on a weakly-supervised attention-based pipeline that extracts survival driven histologic features from H&E WSI and assigns a risk score for each patient. The concordance index (c-index) was used as the primary outcome metric. Further testing of the survival score was performed with a multivariate Cox regression model. The stratification of the cohorts based on the risk scores was evaluated using Kaplan-Meier curves and log-rank test. 95% confidence intervals (CI) are provided. An adjusted two-tailed P value <0.05 was considered significant. The specific morphologic characteristics involved in the AI outcomes are under analysis. Results: The Discovery TCGA cohort consisted of n=463 colon cancer patients; 5-year OS: 68.7% (CI: 60.0%-75.9%). The external validation cohorts included (1) from CHUM, n=124 patients, 5-year OS: 67.0% (CI: 58.0%-75.0%), and (2) from IHP, n=123 patients, 5-year OS: 55.6% (CI 45.0%-67.0%). Cross-validation and testing yielded a c-index of 0.72 and 0.68 respectively, 0.67 for CHUM and 0.65 for IHP cohorts. After external validation, patients with a 'low-risk' score showed significantly higher 5-year OS than patients with a 'high-risk' score: CHUM: 75.0% (CI: 64.0%-84.0%) vs 53.0% (CI 38.0%-67.0%), P<0.05; and IHP 65.0% (CI: 44.0%-80.0%) vs 34% (CI: 21.0%-46.0%), P<0.01. Cox regression showed a significant effect of WSI-based survival score on 5-year OS: TCGA cohort HR=8.3 (CI: 3.1-12.8), P<0.001; CHUM: 7.6 (CI 2.7-21.8), p<0.005; IHP: 5.5 (CI 2.1-16.8), P<0.005. Conclusions: AI-based risk scoring for stage II colon cancer consistently correlated with 5-year OS across multiple independent cohorts, achieving good performances. These findings highlight the potential of modern computational pathology methods requiring minimal supervision to improve risk stratification of stage II colon cancer and patient care.
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 | 0.004 | 0.006 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.000 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
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".