AI Science Tracker
EN

Southern Medical University

2 discoveries from 2022 to 2026

Discoveries

AI-measured retinal vessel features show modest associations with later colorectal cancer

Source check: Source checked. Not independent scientific confirmation.

Finding: Deep-learning retinal vessel measurements were associated with later colorectal-cancer diagnoses in 55,798 UK Biobank participants. Four vascular features retained associations after clinical adjustment, with modest effect sizes. This retrospective analysis of longitudinal records does not establish causal cancer mechanisms, validated screening accuracy or benefit from changing cancer-surveillance decisions.

Technology: RMHAS deep-learning retinal vascular segmentation and automated feature measurement

Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.

AI contribution
The pretrained RMHAS deep-learning pipeline segmented retinal structures at cohort scale, enabling quantitative vascular features linked to subsequent recorded cancer diagnoses. Cox regression alone and generic vessel-segmentation benchmarks are not treated as the enabling AI result.
Test and result
The source analyzed 55,798 initially colorectal-cancer-free participants with usable fundus images, documenting 652 incident cancers over median 11.32-year follow-up. RMHAS segmented arteries, veins and optic disc before quantifying vascular geometry. After adjustment, four features remained associated with incidence; greater arterial-tortuosity variability had a per-standard-deviation hazard ratio 1.03, 95% CI 1.01–1.06. Sensitivity analyses were reported, including exclusion of early diagnoses.
Limitation / still unknown
Observational associations have modest effect sizes and may reflect shared risk factors, reverse causation or residual confounding. Multiple-feature screening and selective imaging eligibility limit interpretation; 93% of participants were White. This was retrospective analysis despite prospective original cohort collection. No independent screening model, incremental clinical utility, causal angiogenesis or reduced cancer mortality was established by the represented finding. Supplementary results were not independently reconstructed.
Evidence sources and annotation context

Evidence context (descriptions follow the cited sources; test type is a catalog classification):

Evidence sources: https://link.springer.com/article/10.1007/s40123-026-01418-w

Source checked · editorial details

The source page was checked against this entry. Independent confirmation is not assessed by this label.

Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.

Review date: 21 Nov 2026. Review scope: Read actual primary UK Biobank eligibility, RMHAS segmentation/feature-extraction, clinical outcome linkage, covariate/FDR/sensitivity-analysis Methods and incidence Results/limitations. Original RMHAS development paper and supplements not independently read. Specific measured cohort association, not screening deployment or causal mechanism, represented. Exact original article front DOI/title and fresh PubMed primary relationship metadata checked..

No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.

Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.

Learned stromal expression score associates with colon-cancer outcomes

Source check: Source checked. Not independent scientific confirmation.

Finding: A 31-gene score selected using elastic-net and penalized survival learning was associated with worse relapse-free and overall survival in retrospective colon-cancer data. A higher score also showed adverse-outcome associations in a separate TCGA cohort. The result is a clinical expression–outcome association, not proof of chemotherapy resistance or a treatment-selection benefit.

Technology: Elastic-net learning; LASSO-Cox feature selection

Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.

AI contribution
Elastic-net models learn stromal-group discriminators, and repeated penalized Cox fits select the expression features that constitute the measured-outcome-associated score.
Test and result
The score was developed using five GEO datasets totalling 990 cases and examined in 382 TCGA-COAD cases. Outcome-informed resampling and expression cutoffs selected genes; the final score subtracts the mean expression of negatively signed selected genes from the positive group. The source reports adjusted survival associations.
Limitation / still unknown
Prognostic outcomes inform development, selection and cutoffs; derivation-cohort associations are not independent validation. Treatment comparisons are observational and do not identify causal benefit or intrinsic resistance. Computational drug-response and pathway predictions are not measured clinical response. Supplemental implementation and treatment-confounding details were not inspected; separate experimental mechanisms are not represented.
Evidence sources and annotation context

Evidence context (descriptions follow the cited sources; test type is a catalog classification):

Evidence sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC9118126/

Source checked · editorial details

The source page was checked against this entry. Independent confirmation is not assessed by this label.

Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.

Review date: 21 Nov 2026. Review scope: Exact primary front and selected computational/clinical Methods and corresponding high-level Results; operational laboratory protocols excluded from review.

No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.

Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.

Open the interactive profile for the timeline, or browse all organizations and model families.