AI Science Tracker
EN

Xinjiang Medical University

3 discoveries from 2024 to 2026

Discoveries

Learned coronary-fat radiomics stratifies an angina-dominated clinical endpoint

Source check: Source checked. Not independent scientific confirmation.

Finding: A learned coronary-fat radiomics score was associated with subsequent cardiovascular readmissions in 578 prospectively collected patients. The composite endpoint was overwhelmingly worsening or unstable angina, with only two myocardial infarctions. The source reports an adjusted hazard ratio of 2.611 between score-defined groups; cutoff selection in the follow-up cohort and uncertain temporal model locking limit its interpretation as independent prognostic validation.

Technology: LASSO radiomics selection and stochastic-gradient-descent classifier of pericoronary fat CT features

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

AI contribution
Supervised feature selection and the learned radiomics classifier combined CT texture and intensity measurements into the score used for the represented clinical event-association analysis. The segmentation software is labeled AI-assisted by the source, but its undisclosed architecture is not invented here.
Test and result
A matched retrospective development set included 200 event cases and 200 controls, with a random 80:20 split and fivefold cross-validation. LASSO selected fourteen radiomics features and an SGD classifier was selected from five classifiers. A separate prospectively collected cohort included 578 patients with mean follow-up 29.17 months: 129 readmissions for worsening stable or unstable angina and two myocardial infarctions. Events were verified from records/interviews and adjudicated by two blinded cardiovascular specialists. The source reports adjusted HR 2.611, 95% CI 1.806–3.775, for score-defined high versus low groups.
Limitation / still unknown
The Methods describe ROC-derived cutoffs in the prospective cohort, so the reported risk separation is not a fully independent test of a prespecified threshold. Five-year retrospective outcome ascertainment overlaps the chronology of 2022 validation enrollment; model locking before those outcomes was not established. Matched case-control sampling and a restricted age-based clinical comparator limit calibration and comparative claims. Feature preprocessing isolation is incompletely specified. Single-center data and an angina-dominated composite do not establish broad hard-event prediction, causal inflammation mechanisms or clinical intervention benefit.
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/s11547-026-02302-0

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 original full publisher clinical enrollment/event adjudication, radiomics/learned-classifier and validation Methods, development/later-cohort Results and limitations. Supplements were not independently read. Prospective collection is distinguished from unestablished temporal model/cutoff locking. 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.

A learned six-gene score associates with archived glioma survival

Source check: Source checked. Not independent scientific confirmation.

Finding: Random-forest and LASSO selection produced a six-gene score associated with overall survival in archived lower-grade glioma data. The source also reports survival separation in three additional datasets using each cohort’s own median cutoff. This does not establish a fixed clinical threshold, a causal tumor mechanism or benefit from a selected drug.

Technology: Random survival forest; LASSO

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

AI contribution
Random forest and LASSO selected overlapping genes, followed by multivariable Cox scoring; the learned selection enables the represented expression-to-survival association.
Test and result
The source divided 497 TCGA cases into 348 development and 149 internal-validation records. Six genes remained after intersecting selectors and Cox refinement; survival comparisons were repeated in CGGA325, CGGA693 and GSE16011.
Limitation / still unknown
Prognostic screening before the split and reuse of the full TCGA cohort as Test2 limit independence. The three additional datasets used cohort-specific medians, so fixed-threshold validation is unestablished. Dataset overlap, endpoint harmonization and full preprocessing isolation were not independently reconstructed. Outcome-selected clusters are not newly verified biological subtypes; inferred immune activity and drug sensitivity are outside this result.
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/PMC11402926/

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: Primary data/model-construction Methods and survival Results only; no operational laboratory procedures reviewed.

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 1/5 · 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.

A learned prostate-expression score associates with recorded progression

Source check: Source checked. Not independent scientific confirmation.

Finding: A CoxBoost and random-survival-forest score separated progression-free-survival groups across archived prostate-cancer datasets. Reported three-year AUCs ranged from 0.677 to 0.933 in the additional datasets. All those datasets informed final model selection, so they do not provide untouched external validation.

Technology: CoxBoost; random survival forest

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

AI contribution
CoxBoost and random survival forests fitted and selected a progression-related expression score; the learned model enables the represented survival association.
Test and result
The source trained on 494 TCGA records and considered 101 algorithm combinations. Five additional dataset groups, totaling 597 records as reported, were used to choose the best average concordance model. The final signature contained 13 genes.
Limitation / still unknown
The additional cohorts selected the final model, and group-specific median scores do not validate a fixed threshold. Cross-database patient independence and endpoint harmonization are not fully established. High training performance and retrospective associations do not establish calibrated clinical prediction, mast-cell causality or benefit from immunotherapy. Drug-sensitivity predictions are excluded.
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/PMC11336432/ · https://www.ncbi.nlm.nih.gov/research/bionlp/RESTful/pmcoa.cgi/BioC_json/PMC11336432/unicode

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: Original primary computational model-selection and cohort survival Methods/Results only.

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 1/5 · 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.