AI in radiology
Software that reads scans alongside radiologists, catching cancers earlier and predicting who is at risk.
Hundreds of FDA-cleared radiology AI devices exist; oncology use cases include mammography reading (Transpara, Lunit INSIGHT, MASAI trial in Sweden showed 29% more cancers detected with 44% less workload), lung nodule detection and malignancy scoring (Sybil, Optellum), prostate MRI, and risk models (Mirai). Foundation models linking images with text are emerging.
How it works
Deep convolutional and transformer networks trained on labelled imaging; increasingly self-supervised on large unlabelled corpora.
- Scales expert reading
- Reduces workload and inter-reader variability
- Dataset shift across scanners and populations
- Regulatory lag for adaptive models
AI can take over one reader's work in double-reading screening programmes while finding more cancers. Whether the extra cancers found are ones that would have harmed women, and whether interval cancers fall, is the question the trial's primary endpoint will answer.
Lung screening works when it uses volumetric nodule management, and it works against a no-screening control. The protocol underpins the UK Targeted Lung Health Check programme and European recommendations. Benefit in women remains less precisely estimated.
Latest papers
topQuery for this technology: (TITLE:"artificial intelligence" OR ABSTRACT:"artificial intelligence" OR TITLE:"deep learning" OR ABSTRACT:"deep learning") AND (TITLE:"mammography" OR ABSTRACT:"mammography" OR TITLE:"lung cancer screening" OR ABSTRACT:"lung cancer screening" OR TITLE:"radiology" OR ABSTRACT:"radiology") AND (cancer OR tumor OR tumour OR oncology OR carcinoma OR lymphoma OR leukemia OR leukaemia OR myeloma OR sarcoma OR melanoma OR glioma). Results are unfiltered search hits about AI in radiology, not a curated reading list.
Pages like this
not linked directly; found by shared links- TechnologyDigital pathology & AI
Shares A dedicated fund for randomised trials of cancer AI with patient outcomes, Continuous prospective validation for every oncology AI tool after deployment, Oncology workforce, Federated training of pathology and radiology models across hospitals.
- BottleneckOverdiagnosis and false alarms
Shares AI malignancy scores to end repeat scans and biopsies for benign lung nodules, Multiparametric prostate MRI (PI-RADS), Require stage-shift or interval-cancer endpoints for AI in cancer screening, Set each woman's mammogram interval from her last mammogram, using AI risk.
- TermStage shift
Shares Require stage-shift or interval-cancer endpoints for AI in cancer screening, Molecular indolence classifiers bundled with every screening programme, Early detection, NELSON: volume-based CT screening reduces lung cancer deaths with fewer false alarms.
- CompanyOwkin
Shares Federated learning and privacy-preserving AI, Federated training of pathology and radiology models across hospitals, A federated learning consortium of cancer centres that jointly own the models, Pathology & radiology foundation models.
- IdeaNo clinical claims for imaging-derived biomarkers without phantom and standards compliance
Shares Aidoc, Lunit, CT (computed tomography), MRI.
- TrialNLST & NELSON (low-dose CT screening)
Shares NELSON: volume-based CT screening reduces lung cancer deaths with fewer false alarms, Low-dose CT lung screening, CT (computed tomography), Most lethal cancers are found late.
- TermSarcopenia
Shares Read muscle loss automatically from scans patients already have, Dose chemotherapy by muscle mass, not body surface area, Body composition (lean mass, fat mass, visceral fat), CT (computed tomography).
- Key paperNLST: yearly low-dose CT scans cut lung cancer deaths in heavy smokers
Shares NELSON: volume-based CT screening reduces lung cancer deaths with fewer false alarms, Low-dose CT lung screening, CT (computed tomography), Most lethal cancers are found late.