AI summary · may contain errors
Oncoformer predicts cancer development within one year and diagnoses tumors using electronic health records and chest X-rays, trained on Chinese patients and tested on a UK cohort.
- In the news
Moving in Cancer Discovery.
Rebuilt
AI summary · may contain errors
Oncoformer predicts cancer development within one year and diagnoses tumors using electronic health records and chest X-rays, trained on Chinese patients and tested on a UK cohort.
AI summary · may contain errors
Foundation models are transforming cancer AI, but no single model consistently excels across every task, motivating model-combination strategies for pathology and drug selection.
Newest first · last 60 days
AI summary · may contain errors
A system-level framework for clinical readiness of foundation and agentic AI in computational pathology is proposed, outlining requirements across evidence, validation, economics, and governance.
AI summary · may contain errors
Adjusted AUC under nuisance-model misspecification can drop below one half under Gaussian latent-index assumptions with exact linear-index strata, with a closed-form correlation boundary.
AI summary · may contain errors
Hist2Prot infers single-cell spatial proteomic profiles from hematoxylin and eosin histology, trained on ~51 million paired cells and validated across ~72 million additional cells.
AI summary · may contain errors
BraTS-Path 2024 benchmark for glioblastoma sub-regions shows top AI performance at Accuracy=0.69 and MCC=0.58, despite strong intra-tumoral heterogeneity and annotation variability.
AI summary · may contain errors
CNN-based dense breast tissue segmentation models achieve the best standalone performance under limited pixel-level supervision on 596 expert-annotated mammograms, balancing accuracy and computational cost.
AI summary · may contain errors
A risk-model-agnostic framework for US women maps absolute breast cancer risk cut points to screening intervals, producing a risk-scaled gradient (annual at 2%-3%, annual with supplemental imaging/counseling at ≥3%).
AI summary · may contain errors
A multimodal AI digital pathology score predicts distant metastasis after salvage radiotherapy in 533 post–radical prostatectomy biochemical recurrence patients, with a subdistribution hazard ratio of 2.17 per SD.
AI summary · may contain errors
Pretrained MobileNet with a differential-attention module best distinguishes high-grade from low-grade serous carcinoma on the UBC-OCEAN slide-level task, achieving AUC 0.862 ± 0.005.
AI summary · may contain errors
CADe device availability in Veterans Health Administration colonoscopy sites increased adenoma detection odds (OR 1.22, 95% CI 1.15–1.28) versus usual care.
AI summary · may contain errors
A deep learning model using routine whole-slide histopathology from early-stage melanoma predicts distant metastatic risk with added prognostic value beyond AJCC stage, achieving a best-fold balanced accuracy of 0.88.
AI summary · may contain errors
An integrated LDCT-based early lung cancer detection program at two Ethiopian tertiary centers established a working infrastructure, with 73.9% of participants never having smoked.
AI summary · may contain errors
Combining endogenous fluorophore concentration and local histogram entropy improves normal versus cancerous lung tissue discrimination, reaching a mean ROC AUC of 98.4% ± 0.7% in five-fold cross-validation.
AI summary · may contain errors
Self-supervised learning on over 7,000 multiplex immunofluorescence images from 1,800+ patients identifies previously unknown prognostic spatial patterns, outperforming segmentation-based analyses.
AI summary · may contain errors
MorphX, an interpretable deep-learning framework, predicts survival in hepatocellular carcinoma (HCC) and lung adenocarcinoma (LUAD) with concordance indices of 0.775–0.799 in discovery cohorts.
AI summary · may contain errors
PAIR-IBD spatial quantification distinguishes pathologist-defined active ulcerative colitis from histological remission by detecting higher compartment-specific immune densities and crypt injury, including 2.1-fold LP neutrophils.
AI summary · may contain errors
Spatial immunophenotyping and multimodal immune profiling improve 10-year survival prediction in a cohort of 664 Malaysian breast cancer patients, reaching AUROC 0.71 with IHC/spatial features.
AI summary · may contain errors
The NCI convened the Foundation Models for Cancer Workshop (March 24–26, 2026), drawing 2,000+ registrants worldwide and highlighting key requirements for reliable oncology foundation models.
AI summary · may contain errors
HRDPath predicts homologous recombination deficiency in high-grade serous ovarian carcinoma with an AUC of 0.846, outperforming previous methods across three patient cohorts.
AI summary · may contain errors
Class visualizations and activation atlases in a pathology foundation-model setting retain class-associated morphology, but separability drops and ambiguity rises for morphologically overlapping colorectal and multi-organ cancer classes.
AI summary · may contain errors
FAHR-FaceSurvival, fine-tuned on 34,363 cancer patient photographs, predicts mortality with robust performance, with the highest-risk quartile showing an adjusted hazard ratio of 3·23 versus the lowest.
AI summary · may contain errors
SQUALL, a multimodal foundation model, was pretrained on histMol using 1.93 billion spatially aligned histology–transcriptomics spots/bins from 3,446 sections, enabling learned morphology–molecular correspondences.
AI summary · may contain errors
FOCUS, a multimodal generative foundation model for spatial transcriptomics, jointly improves molecular and spatial fidelity and recovers biologically consistent patterns across immunology, development, and oncology.
AI summary · may contain errors
The Transformer-based deep learning model using gene expression plus preoperative clinical data predicts sentinel lymph node macro-metastasis in SCAN-B node-negative T1–T2 breast cancer, improving ROC AUC to 0.693.
Journals this month
Topics moving, week to 17 Nov 2026
By field