A researcher's catalogue
Freely available clinical and health datasets, from ICU records to imaging, genomics, signals and population surveys. Each entry has a short description, the original article and how to get access.
Every resource here is free for research. "Free" means no fee to obtain the data. Commercial use, cloud compute or some sponsor datasets may have their own terms. Fee-based sources such as UK Biobank, CPRD, HCUP NIS and MarketScan are deliberately left out.
Detailed, de-identified records from intensive care units, operating rooms and emergency departments: vitals, labs, medications, notes and outcomes.
Medical Information Mart for Intensive Care IV
The flagship open critical-care database: de-identified EHR data for patients admitted to the ICU or emergency department of Beth Israel Deaconess Medical Center, Boston (2008–2019). Includes demographics, vitals, labs, medications, procedures, diagnoses and survival, organised in modular hospital (hosp) and ICU (icu) tables.
Johnson AEW et al. Sci Data 2023;10:1
Medical Information Mart for Intensive Care III
The predecessor of MIMIC-IV covering ~46,000 ICU patients at BIDMC (2001–2012), including bedside charting, labs, medications and >2 million free-text clinical notes. Still the most cited ICU dataset and the basis of many benchmarks.
Johnson AEW et al. Sci Data 2016;3:160035
A 100-patient subset of MIMIC-IV with identical structure, released openly so anyone can learn the schema and test code before applying for full access.
Emergency department module of MIMIC-IV with ~425,000 ED stays: triage, vital signs, medication reconciliation, dispensed medications and diagnoses. Links to MIMIC-IV hospital and ICU records.
De-identified free-text discharge summaries (~330,000) and radiology reports (~2.3 million) linked to MIMIC-IV patients — a major resource for clinical NLP and LLM research.
Echocardiogram DICOM studies (~500,000 videos) from patients in MIMIC-IV, linkable to their EHR data and notes for multimodal cardiac research.
eICU Collaborative Research Database
Multi-centre ICU data for >200,000 admissions across 208 US hospitals (2014–2015) collected through the Philips eICU telehealth programme. Complements single-centre MIMIC for external validation.
Pollard TJ et al. Sci Data 2018;5:180178
The first freely accessible European ICU database: ~23,000 ICU admissions at Amsterdam UMC (2003–2016) with high-resolution vitals, ventilator settings, labs and drugs (~1 billion data points).
Thoral PJ et al. Crit Care Med 2021;49:e563
High time-Resolution ICU Dataset
Data for ~34,000 ICU admissions at Bern University Hospital with physiological parameters sampled every 2 minutes — notably higher resolution than most ICU datasets. Developed for early warning of circulatory failure.
Hyland SL et al. Nat Med 2020;26:364
Salzburg Intensive Care Database
ICU data for ~27,000 admissions at University Hospital Salzburg (2013–2021) including per-minute vital signs, therapies, labs and cost-related information.
Rodemund N et al. Intensive Care Med 2023;49:987
Paediatric Intensive Care Database
A large paediatric ICU database from the Children's Hospital of Zhejiang University School of Medicine (2010–2018) covering ~13,000 children, structured similarly to MIMIC.
Zeng X et al. Sci Data 2020;7:14
High-fidelity intraoperative data from 6,388 surgical patients at Seoul National University Hospital: synchronised waveforms (ECG, arterial pressure, EEG/BIS, ventilator), numeric vitals, anaesthetic drugs and perioperative clinical data.
Lee HC et al. Sci Data 2022;9:279
Perioperative research dataset of ~130,000 surgical cases from Seoul National University Hospital, combining intraoperative vitals with pre- and post-operative EHR data and outcomes.
Lim L et al. Sci Data 2024;11:655
Multimodal Clinical Monitoring in the Emergency Department
118,385 adult ED visits at Stanford (2020–2022) pairing continuous vitals and waveforms (ECG, PPG, respiration) with structured EHR data and radiology reports.
Kansal A et al. Sci Data 2025
Medical Informatics Operating Room Vitals and Events Repository
Operating-room data from ~58,000 surgical patients at UC Irvine Medical Center, including intraoperative waveforms, vitals, medications and perioperative EHR records.
Samad M et al. JAMIA Open 2023;6:ooad084
Hourly ICU data for ~40,000 patients from three hospital systems, released for early prediction of sepsis. A compact, ready-to-model benchmark.
Reyna MA et al. Crit Care Med 2020;48:210
Northwestern ICU Database
Harmonised ICU data for 25,000+ patients at Northwestern Medicine (2020–2022) in a MIMIC-IV-compatible structure, enabling multi-centre validation.
Critical care database of patients with infection, Zigong Fourth People's Hospital
ICU records of patients with infection at Zigong Fourth People's Hospital, China (2019–2020), with vitals, labs, drugs and outcomes.
22.6 million clinical time-series events derived from MIMIC-IV discharge summaries for temporal reasoning research.
Multimodal pulmonary-embolism dataset of 19,438 patients: CT pulmonary angiograms, radiology impressions and longitudinal structured EHR data with outcome labels.
Huang SC et al. NeurIPS 2023 Datasets & Benchmarks
High-resolution anaesthesia monitoring data (vital signs and waveforms) from 32 surgical cases at the Royal Adelaide Hospital.
Liu D et al. Anesth Analg 2012;114:584
14,845 admissions of 12,238 patients to the cardiac unit of Hero DMC Heart Institute, Ludhiana, India (2017–2019) with labs, comorbidities and outcomes.
Large longitudinal research cohorts, national registries, multi-site EHR enclaves and realistic synthetic records for prototyping.
NIH programme building a cohort of 1 million+ diverse US participants with EHRs, surveys, physical measurements, wearables and whole-genome sequences, analysed in a cloud Researcher Workbench.
All of Us Investigators. N Engl J Med 2019;381:668
National COVID Cohort Collaborative
A secure enclave harmonising EHR data from >20 million patients (COVID-19 positive and controls) across US health systems into the OMOP model.
Haendel MA et al. JAMIA 2021;28:427
Open-source synthetic patient generator plus downloadable synthetic populations (e.g. SyntheticMass) with complete lifetime records in FHIR, C-CDA and CSV — no privacy restrictions.
Walonoski J et al. JAMIA 2018;25:230
Medicare Claims Synthetic Public Use Files
Synthetic Medicare beneficiary, inpatient, outpatient, carrier and prescription-drug claims (2008–2010) that mirror the structure of real CMS claims for training and software development.
~100,000 diabetic inpatient encounters from 130 US hospitals with labs, medications and 30-day readmission outcome — a classic teaching and ML benchmark.
Strack B et al. Biomed Res Int 2014:781670
Surveillance, Epidemiology, and End Results Program
Population-based cancer registry data covering ~48% of the US population: incidence, stage, treatment and survival since 1975, analysable with the free SEER*Stat software.
Cancer Data Access System
Individual-level data, images and biospecimen information from the PLCO cancer screening trial (~155,000 participants) and the National Lung Screening Trial (~54,000 participants).
NLST Research Team. N Engl J Med 2011;365:395
Biologic Specimen and Data Repository Information Coordinating Center
Individual-level data from >250 NHLBI-funded cohorts and trials such as Framingham, ARIC, MESA, CARDIA, WHI and SPRINT — a gold mine for cardiovascular and lung epidemiology.
Data and biosamples from NIDDK-funded studies in diabetes, obesity, kidney, liver and digestive diseases (e.g. DPP, TODAY, CRIC).
Parkinson's Progression Markers Initiative
Longitudinal clinical, imaging (DaTscan, MRI), genetic and biomarker data from thousands of people with and at risk of Parkinson's disease.
Marek K et al. Prog Neurobiol 2011;95:629
National Alzheimer's Coordinating Center
Standardised longitudinal clinical and neuropathology data from US Alzheimer's Disease Research Centers (>50,000 participants).
Pooled Resource Open-Access ALS Clinical Trials
Harmonised records of >11,000 ALS patients pooled from completed clinical trials — the largest ALS clinical trial dataset.
Atassi N et al. Neurology 2014;83:1719
Small classic dataset of 303 patients from Cleveland (plus Hungary, Switzerland, Long Beach) with 14 clinical attributes for predicting coronary artery disease — widely used in teaching.
Flagship Bridge2AI dataset on type 2 diabetes: 2,280 participants with retinal imaging, ECG, continuous glucose monitoring, wearables, environmental sensors and surveys (~3.9 TB).
Genomic sequences and deep phenotypes of patients with undiagnosed rare conditions, released annually through dbGaP.
Standardised clinical data on 700,000+ hospitalised and outpatient COVID-19 patients from 60+ countries.
Eight weeks of CGM glucose, insulin, meals, exercise, sleep and wearable data for 12 people with type 1 diabetes.
Marling C, Bunescu R. KDH Workshop 2020
Nationally representative household, health-examination and ageing surveys — the backbone of epidemiology and health-policy research.
National Health and Nutrition Examination Survey
Continuous US survey combining interviews, physical examinations and laboratory tests (~5,000 people per year) on nutrition, chronic disease, biomarkers and environmental exposures.
Behavioral Risk Factor Surveillance System
The world's largest continuously conducted telephone health survey (>400,000 US adults per year) on risk behaviours, chronic conditions and preventive-service use.
National Health Interview Survey
Annual cross-sectional household survey on the health status, health-care access and insurance of the US civilian population since 1957.
Medical Expenditure Panel Survey
Panel survey of US families, their medical providers and employers detailing health-care use, expenditures, insurance and prescribed medicines.
Youth Risk Behavior Surveillance System
Biennial school-based survey of US high-school students on violence, substance use, sexual behaviour, diet, physical activity and mental health.
National Survey on Drug Use and Health
Annual nationally representative US survey of tobacco, alcohol and drug use, mental illness and treatment, with public-use files and an online analysis tool.
National Longitudinal Study of Adolescent to Adult Health
Nationally representative cohort of US adolescents (grades 7–12 in 1994–95) followed into adulthood with social, behavioural, biomarker and genetic data across five waves.
Harris KM et al. Int J Epidemiol 2019;48:1415
Demographic and Health Surveys
Nationally representative household surveys from 90+ low- and middle-income countries (incl. Pakistan, India's NFHS, Bangladesh, Nigeria) on fertility, maternal and child health, nutrition, HIV and malaria, with GPS cluster data.
Corsi DJ et al. Int J Epidemiol 2012;41:1602
Multiple Indicator Cluster Surveys
Household surveys in 100+ countries producing internationally comparable indicators on children and women: nutrition, immunisation, WASH, education and child protection.
Microdata from WHO STEPwise surveys of non-communicable disease risk factors (tobacco, diet, blood pressure, glucose) in dozens of countries, plus GSHS school surveys.
Health and Retirement Study
Longitudinal panel of ~20,000 Americans over 50 tracking health, cognition, work, wealth and family since 1992; the template for a global family of ageing studies.
Sonnega A et al. Int J Epidemiol 2014;43:576
English Longitudinal Study of Ageing
Biennial panel of people aged 50+ in England since 2002 covering health, biomarkers, cognition, economics and social participation; harmonised with HRS.
Steptoe A et al. Int J Epidemiol 2013;42:1640
Survey of Health, Ageing and Retirement in Europe
Cross-national panel of ~140,000 people aged 50+ in 28 European countries and Israel on health, socio-economic status and social networks.
Börsch-Supan A et al. Int J Epidemiol 2013;42:992
China Health and Retirement Longitudinal Study
Nationally representative survey of Chinese residents aged 45+ (~19,000 people) with health, biomarkers, function, income and family data since 2011.
Zhao Y et al. Int J Epidemiol 2014;43:61
China Health and Nutrition Survey
Longitudinal household survey (1989 onward) on nutrition, health, economics and the effects of social change across Chinese provinces.
Popkin BM et al. Int J Epidemiol 2010;39:1435
Korea National Health and Nutrition Examination Survey
South Korea's national health-examination survey (since 1998) with interviews, physical exams, labs and nutrition data, modelled on NHANES.
Kweon S et al. Int J Epidemiol 2014;43:69
Harmonised data files and documentation for the HRS family of ageing studies (HRS, ELSA, SHARE, CHARLS, LASI, MHAS, KLoSA and more) for cross-country comparison.
Harmonised, documented versions of NHIS (1963–present) and MEPS with a custom data-extract builder that makes multi-year analysis simple.
Brazilian Longitudinal Study of Ageing
Nationally representative cohort of Brazilians aged 50+ (9,412 at baseline) covering health, function, care and economics; part of the HRS family.
Lima-Costa MF et al. Am J Epidemiol 2018;187:1345
National Neighborhood Data Archive
US neighbourhood-level measures (food, healthcare, transport, pollution, amenities) by census tract and ZIP code for social-determinants research.
Aggregated indicators, burden-of-disease estimates, mortality and cancer statistics for countries and regions worldwide.
WHO's portal of 1,000+ health indicators for 194 member states — mortality, disease burden, health systems, SDG targets — with an OData API.
Global Burden of Disease Study
Estimates of deaths, DALYs, prevalence and risk-factor burden for 370+ diseases and injuries in 204 countries from 1990 onward.
GBD 2021 Diseases and Injuries Collaborators. Lancet 2024;403:2133
Country-level time series on health financing, service coverage, nutrition, population dynamics and outcomes, downloadable via DataBank or API.
Curated, well-documented global datasets and charts on health topics (causes of death, vaccination, COVID-19, child mortality, life expectancy) with GitHub-hosted data.
Mathieu E et al. Nat Hum Behav 2021;5:947
Deaths by country, year, sex, age and ICD cause as reported by national civil-registration systems since 1950.
High-quality, harmonised life tables, death counts and population exposures for ~40 countries, the reference source for demographic mortality research.
IARC's estimates of cancer incidence, mortality and prevalence for 185 countries, plus Cancer Over Time, Cancer Tomorrow and Cancer Survival modules.
Bray F et al. CA Cancer J Clin 2024;74:229
Online query system for US public-health data: births, deaths (multiple cause), cancer statistics, vaccine adverse events, environmental and population data.
Public-use micro-data files for every US birth, death and fetal death record, released annually by NCHS.
Model-based small-area estimates of 40 chronic-disease measures for US counties, cities, census tracts and ZIP codes.
Country-level HIV estimates — prevalence, incidence, AIDS deaths, treatment cascade and key populations — from UNAIDS modelling.
EU-wide statistics on causes of death, health-care resources and expenditure, health status (EHIS) and accidents at work.
US federal catalogue of thousands of health datasets from HHS agencies (CDC, CMS, FDA, NIH) with API access.
Public Medicare and Medicaid datasets: provider utilisation and payments, Part D prescribers, hospital quality and cost reports.
Brazilian Ministry of Health information system: hospital admissions (SIH), mortality (SIM), births (SINASC), notifiable diseases (SINAN) and more.
High-resolution gridded population, demographic, births and urbanisation estimates for every country — essential denominators for health mapping.
Chest X-ray, CT, MRI, mammography, ultrasound and echocardiography collections, many with labels or segmentations.
377,110 chest radiographs (DICOM) with 227,835 free-text radiology reports from BIDMC, linkable to MIMIC-IV. A JPG version with NLP-derived labels (MIMIC-CXR-JPG) is also available.
Johnson AEW et al. Sci Data 2019;6:317
224,316 chest radiographs of 65,240 Stanford patients labelled for 14 observations with uncertainty labels; the basis of a well-known benchmark.
Irvin J et al. AAAI 2019;33:590
112,120 frontal chest X-rays from 30,805 patients at the NIH Clinical Center with 14 text-mined disease labels and some bounding boxes.
Wang X et al. CVPR 2017
160,000+ chest X-rays from Hospital San Juan (Spain) with Spanish reports and 174 radiographic findings, 27% manually annotated by radiologists.
Bustos A et al. Med Image Anal 2020;66:101797
18,000 chest X-rays from two Vietnamese hospitals with radiologist bounding-box annotations for 22 local findings and 6 global diagnoses.
Nguyen HQ et al. Sci Data 2022;9:429
5,000 four-view full-field digital mammography exams with BI-RADS, density and finding-level annotations from Vietnamese hospitals.
Nguyen HT et al. Sci Data 2023;10:277
~30,000 chest X-rays (from NIH ChestX-ray14) re-annotated by radiologists with bounding boxes for lung opacity suggestive of pneumonia.
Shih G et al. Radiol Artif Intell 2019;1:e180041
Large archive of de-identified cancer imaging collections (CT, MRI, PET, pathology) — e.g. LIDC-IDRI, CBIS-DDSM, NLST — often linked to TCGA genomics and clinical data.
Clark K et al. J Digit Imaging 2013;26:1045
1,018 thoracic CT scans with lung-nodule annotations by four radiologists — the standard dataset for lung-nodule detection (LUNA16 derives from it).
Armato SG et al. Med Phys 2011;38:915
Curated, DICOM-converted subset of the Digital Database for Screening Mammography with updated ROI segmentations and pathology-verified labels.
Lee RS et al. Sci Data 2017;4:170177
Emory Breast Imaging Dataset
3.4 million screening and diagnostic mammograms from 116,000 racially diverse women with BI-RADS, pathology and demographic data.
Jeong JJ et al. Radiol Artif Intell 2023;5:e220047
Ten 3D segmentation tasks (brain tumour, heart, liver, hippocampus, prostate, lung, pancreas, hepatic vessels, spleen, colon) for benchmarking generalisable algorithms.
Antonelli M et al. Nat Commun 2022;13:4128
1,228 CT scans with segmentations of 117 anatomical structures (organs, bones, muscles, vessels), released alongside the popular TotalSegmentator tool.
Wasserthal J et al. Radiol Artif Intell 2023;5:e230024
Kidney Tumor Segmentation Challenge
Contrast-enhanced abdominal CT scans with expert segmentations of kidneys, tumours and cysts (KiTS19/21/23).
Heller N et al. Med Image Anal 2021;67:101821
Liver Tumor Segmentation Benchmark
131 training and 70 test contrast-enhanced abdominal CT scans with liver and tumour segmentations from seven institutions.
Bilic P et al. Med Image Anal 2023;84:102680
888 CT scans from LIDC-IDRI with standardised nodule annotations and an evaluation framework for nodule detection.
Setio AAA et al. Med Image Anal 2017;42:1
Raw k-space and DICOM data for knee, brain, prostate and breast MRI from NYU Langone, built for accelerated MRI reconstruction research.
Knoll F et al. Radiol Artif Intell 2020;2:e190007
10,030 apical four-chamber echocardiography videos with expert tracings, ejection fraction and volumes from Stanford Medicine.
Ouyang D et al. Nature 2020;580:252
2D echocardiographic sequences from 500 patients with expert segmentations of left ventricle, myocardium and left atrium.
Leclerc S et al. IEEE Trans Med Imaging 2019;38:2198
Automated Cardiac Diagnosis Challenge
Cine cardiac MRI from 150 patients across five diagnostic groups with ventricular and myocardial segmentations.
Bernard O et al. IEEE Trans Med Imaging 2018;37:2514
Breast Ultrasound Images Dataset
780 breast ultrasound images from 600 women in Egypt, labelled normal, benign or malignant with ground-truth masks.
Al-Dhabyani W et al. Data Brief 2020;28:104863
108,000+ labelled retinal OCT images (CNV, DME, drusen, normal) and 5,800 paediatric chest X-rays (pneumonia vs normal) from the landmark Cell transfer-learning paper.
Kermany DS et al. Cell 2018;172:1122
40,561 musculoskeletal radiographs of the upper extremity from 12,173 patients labelled normal/abnormal by radiologists.
Rajpurkar P et al. arXiv 2017
Portal hosting many Stanford imaging datasets free for research: CheXpert, EchoNet, DDI, CT colonography, knee MRI, BrainMetShare and more.
NLM search engine over 3.7 million images from PubMed Central articles plus the Indiana University chest X-ray collection (7,470 images with reports).
Demner-Fushman D et al. JAMIA 2016;23:304
18 standardised, lightweight 2D and 3D biomedical image datasets (pathology, chest X-ray, dermatology, retina, CT organs…) in MNIST-like format for fast prototyping and teaching.
Yang J et al. Sci Data 2023;10:41
Open benchmark of tens of thousands of chest X-rays and CT scans assembled from multiple public sources for COVID-19 detection (COVID-Net).
Wang L et al. Sci Rep 2020;10:19549
Chest X-ray and CT studies of COVID-19 patients from the Valencian region with radiological findings, reports and lab results.
Free open-access database of ~59,000 medical images and ~12,000 teaching cases indexed by diagnosis, organ system and modality.
Medical Imaging and Data Resource Center
Open commons of 500,000+ de-identified imaging studies (chest X-ray, CT, MRI) with clinical data, initially built for COVID-19.
1,255 cervical spine MRI exams with vertebra and disc segmentations from Duke University Health System.
Pre-operative dynamic contrast-enhanced breast MRI of 922 invasive breast cancer patients with imaging features, pathology and outcomes.
Saha A et al. Br J Cancer 2018;119:508
1,506 breast DCE-MRI cases from four TCIA collections with expert tumour segmentations and harmonised clinical data.
Enriched screening mammography set with cancers, biopsy-proven benign findings and negatives, released after the RSNA breast-cancer-detection challenge.
~874,000 head CT slices from 25,000 exams annotated by neuroradiologists for five haemorrhage subtypes.
Flanders AE et al. Radiol Artif Intell 2020;2:e190211
Knee MRI studies labelled for twelve abnormal findings from the RSNA challenge.
RSNA International COVID-19 Open Radiology Database
Expert-annotated chest CT (240 studies) and chest radiographs (1,000) of COVID-19 from multiple institutions.
Tsai EB et al. Radiology 2021;299:E204
82 head CT scans of traumatic brain injury patients with intracranial haemorrhage segmentations.
Hssayeni MD et al. Data 2020;5:14
9,000+ non-contrast head CT scans labelled normal or abnormal (~1.3 TB) from Chinese hospitals.
32,000+ annotated lesions on 10,594 CT studies of 4,427 patients, mined from radiologist bookmarks at the NIH Clinical Center.
Yan K et al. J Med Imaging 2018;5:036501
Large-scale multi-organ annotations on thousands of abdominal CT volumes (AbdomenAtlas-8K and later releases) produced with AI-assisted expert review.
Qu C et al. NeurIPS 2023 Datasets & Benchmarks
17,000+ CT scans with pan-cancer lesion labels from the FLARE challenge for foundation-model segmentation.
Chest CT volumes with free-text findings linked to pixel-level 3D segmentations for grounded report generation.
Initial public release of 3,630 chest CT volumes from Duke with 83 abnormality labels extracted from reports.
Draelos RL et al. Med Image Anal 2021;67:101857
Brain and spine MRI volumes paired with radiology reports for vision-language modelling.
628 T1- and T2-weighted abdominal MRI scans of cirrhotic livers with expert segmentations.
40,967 chest X-rays from Hospital Israelita Albert Einstein, São Paulo, labelled for 14 findings.
Reis EP et al. Sci Data 2022;9:487
4,555 bilingual (Spanish/English) chest X-ray studies with sentence-level findings grounded to bounding boxes.
223,462 chest X-ray–report pairs from Stanford with pathology labels, RadGraph annotations and demographics.
Chambon P et al. arXiv 2024
7,643 paediatric echocardiogram videos with LV tracings and ejection fraction from Lucile Packard Children's Hospital.
Reddy CD et al. J Am Soc Echocardiogr 2023;36:482
12,000 parasternal long-axis echo videos with wall-thickness measurements for detecting LV hypertrophy and its causes.
Duffy G et al. JAMA Cardiol 2022;7:386
375 cine cardiac MRI studies from four scanner vendors and six centres with ventricular segmentations, for domain-generalisation research.
Campello VM et al. IEEE Trans Med Imaging 2021;40:3543
Point-of-care lung ultrasound clips (300,000+ frames) from 226 emergency department patients with labels.
100,000+ 3D medical shapes (organs, bones, vessels, instruments) extracted from public segmentation datasets.
Li J et al. arXiv 2023
Structured version of NLM MedPix CT and MRI cases with clinical notes, built for multimodal model training.
Non-profit platform of curated, outcome-linked medical datasets (ECG waveforms, biopsy slides, chest X-rays) from health systems worldwide.
Whole-slide histopathology, dermoscopic and clinical skin images, retinal fundus and OCT collections.
Whole-slide images of sentinel lymph nodes from breast-cancer patients with pixel-level metastasis annotations (1,399 WSIs) from Dutch medical centres.
Ehteshami Bejnordi B et al. JAMA 2017;318:2199
Prostate cANcer graDe Assessment Challenge
~11,000 prostate biopsy whole-slide images from Radboud and Karolinska with ISUP grade and Gleason labels.
Bulten W et al. Nat Med 2022;28:154
327,680 96×96 histopathology patches derived from CAMELYON16 with binary metastasis labels — an MNIST-sized pathology benchmark.
Veeling BS et al. MICCAI 2018
7,909 microscopic images of breast tumour tissue from 82 patients at four magnifications, labelled benign or malignant with subtypes.
Spanhol FA et al. IEEE Trans Biomed Eng 2016;63:1455
The largest public collection of dermoscopic images of skin lesions (>400,000), with diagnoses and metadata, home of the annual ISIC melanoma challenges.
10,015 dermoscopic images of common pigmented skin lesions across 7 diagnostic categories, >50% histopathology-confirmed.
Tschandl P et al. Sci Data 2018;5:180161
16,577 clinical skin-condition images across 114 conditions annotated with Fitzpatrick skin type to study fairness in dermatology AI.
Groh M et al. CVPR Workshops 2021
Diverse Dermatology Images
656 biopsy-proven clinical images emphasising Fitzpatrick skin types V–VI and uncommon diseases, built to expose bias in dermatology algorithms.
Daneshjou R et al. Sci Adv 2022;8:eabq6147
2,298 smartphone images of skin lesions from 1,373 Brazilian patients with clinical metadata across six diagnoses.
Pacheco AGC et al. Data Brief 2020;32:106221
~88,000 retinal fundus photographs graded 0–4 for diabetic retinopathy, released for the 2015 Kaggle competition.
~5,600 fundus images from rural Indian screening programmes (Aravind Eye Hospital) graded for diabetic retinopathy severity.
Indian Diabetic Retinopathy Image Dataset
516 high-resolution fundus images with pixel-level lesion annotations, DR/DME grades and optic disc/fovea locations.
Porwal P et al. Data 2018;3:25
1,748 macula-centred fundus images from 874 diabetic examinations, widely used to validate DR screening algorithms.
Decencière E et al. Image Anal Stereol 2014;33:231
Digital Retinal Images for Vessel Extraction
40 colour fundus images with manual retinal-vessel segmentations — the classic vessel-segmentation benchmark.
Staal J et al. IEEE Trans Med Imaging 2004;23:501
Retinal Fundus Glaucoma Challenge
1,200 fundus images with glaucoma labels and optic disc/cup segmentations from Chinese hospitals.
Orlando JI et al. Med Image Anal 2020;59:101570
100,000 H&E image patches of human colorectal cancer and normal tissue in nine tissue classes.
Kather JN et al. PLoS Med 2019;16:e1002730
630,000 colorectal tissue image tiles across nine tissue classes for training pathology classifiers.
Nuclei instance segmentation and classification (~200,000 nuclei) across 19 tissue types.
Gamper J et al. arXiv 2020
220,000+ crowdsourced and expert-reviewed nucleus labels from TCGA breast cancer slides.
Amgad M et al. GigaScience 2022;11:giac037
Prostate cancer histopathology: 2.6 million patches and 4,675 whole-slide scans with diagnoses.
Koziarski M et al. Sci Rep 2024;14:6780
Pathology images of the same tissues captured with different whole-slide scanners and smartphones, for studying domain shift.
3D pathology image volumes of prostate and oesophageal tissue with binary labels for deep-learning triage of 3D microscopy.
27,558 segmented thin blood-smear red-cell images, evenly split between Plasmodium-parasitised and uninfected.
Rajaraman S et al. PeerJ 2018;6:e4568
200 dermoscopic images of melanocytic lesions with expert segmentations and diagnoses from Hospital Pedro Hispano, Portugal.
Mendonça T et al. IEEE EMBC 2013
170 non-dermoscopic clinical images of melanoma and naevi from the University Medical Center Groningen.
Giotis I et al. Expert Syst Appl 2015;42:6578
1,011 cases with clinical and dermoscopic images annotated with the seven-point melanoma checklist and diagnoses.
Kawahara J et al. IEEE J Biomed Health Inform 2019;23:538
1,947 teledermatology cases with differential diagnoses from multiple dermatologists across 419 conditions, for studying label uncertainty.
2,201 clinical skin images across 174 diseases with quizzes, released with the Seoul National University dermatology AI studies.
Objective skin-tone measurements linked to EHR data (167 variables) to study pulse-oximetry and other device bias.
Brazilian Multilabel Ophthalmological Dataset
16,266 retinal fundus photographs from 8,524 Brazilian patients with demographics and multi-label diagnoses.
Nakayama LF et al. PLOS Digit Health 2024;3:e0000454
5,164 smartphone retinal images from 1,291 diabetic patients in Brazil with DR grades and systemic data.
747 fundus images with gold-standard glaucoma labels based on full ophthalmic examination.
47,247 fundus fluorescein angiography images from 10,790 reports with lesion annotations, for explainable report generation.
High-Resolution Fundus Image Database
45 high-resolution fundus images (healthy, diabetic retinopathy, glaucoma) with vessel segmentations.
Budai A et al. Int J Biomed Imaging 2013:154860
750 fundus images with optic disc and cup boundaries marked by six ophthalmologists (4,500 annotations) for glaucoma research.
Almazroa A et al. Proc SPIE 2018;10579
39 fundus images from a UK diabetic screening programme with manual vessel segmentations, covering comorbidities.
Holm S et al. J Med Imaging 2017;4:014503
712 fluorescein-stained ocular surface images with corneal ulcer segmentations and type/grade labels.
Deng L et al. Sci Data 2020;7:23
Fundus images with optic disc and cup segmentations, curated for deep-learning glaucoma assessment.
Brain MRI, fMRI, PET and EEG collections for ageing, dementia, development and psychiatric research.
Alzheimer's Disease Neuroimaging Initiative
Longitudinal MRI, PET, CSF/blood biomarkers, genetics and cognitive data from >2,000 participants (normal, MCI, Alzheimer's) since 2004.
Petersen RC et al. Neurology 2010;74:201
Open Access Series of Imaging Studies
Brain MRI (and PET in OASIS-3/4) with clinical and cognitive data across normal ageing and Alzheimer's disease, from Washington University's Knight ADRC.
Marcus DS et al. J Cogn Neurosci 2007;19:1498
Open platform hosting 1,000+ BIDS-formatted neuroimaging datasets (MRI, MEG, EEG, iEEG, PET) contributed by researchers worldwide.
Markiewicz CJ et al. eLife 2021;10:e71774
High-quality multimodal MRI (structural, resting/task fMRI, diffusion) and behavioural data from ~1,200 healthy young adults, plus Lifespan and disease extensions.
Van Essen DC et al. NeuroImage 2013;80:62
Autism Brain Imaging Data Exchange
Resting-state fMRI and structural MRI from >2,000 individuals with autism and typical controls aggregated from 24+ international sites.
Di Martino A et al. Mol Psychiatry 2014;19:659
Adolescent Brain Cognitive Development Study
Largest long-term study of brain development in the US: ~11,800 children followed from age 9–10 with MRI, cognition, mental health, substance use and genetics.
Casey BJ et al. Dev Cogn Neurosci 2018;32:43
Federated archive of human-subjects mental-health research data (clinical, imaging, genomics) from hundreds of NIH-funded studies, including ABCD and HCP lifespan.
~600 MR images (T1, T2, PD, MRA, DTI) from healthy subjects scanned at three London hospitals.
Brain Tumor Segmentation Challenge
Multi-institutional multiparametric MRI (T1, T1-Gd, T2, FLAIR) of gliomas and other brain tumours with expert segmentations of tumour sub-regions.
Menze BH et al. IEEE Trans Med Imaging 2015;34:1993
The world's largest open clinical EEG corpus (>30,000 recordings) from Temple University Hospital with physician reports; subsets annotated for seizures, artefacts and abnormality.
Obeid I, Picone J. Front Neurosci 2016;10:196
Long-term scalp EEG of 22 paediatric subjects with intractable seizures at Boston Children's Hospital, with annotated seizure onsets.
EEG recordings of 14 adult epilepsy patients from the University of Siena with seizure annotations by neurologists.
Detti P et al. Processes 2020;8:846
1,500+ 64-channel EEG recordings from 109 volunteers performing real and imagined motor tasks; widely used for brain–computer interface research.
Schalk G et al. IEEE Trans Biomed Eng 2004;51:1034
R package bundling a diverse collection of ready-to-use datasets on the brain, nervous system and neurological and psychiatric disorders, each documented with its original source and licence.
Multiparametric MRI of 630 glioblastoma patients with tumour segmentations, radiomic features, genomics and survival.
Bakas S et al. Sci Data 2022;9:453
Pre-operative brain MRI with glioma segmentations for 146 patients from Nigeria, addressing under-representation of African populations.
Adewole M et al. arXiv 2023
Anatomical Tracings of Lesions After Stroke
955 T1-weighted MRIs of stroke patients with manually segmented lesions from 33 cohorts worldwide.
Liew SL et al. Sci Data 2022;9:320
Longitudinal MRI, segmentations, treatment (gamma knife) and outcome data of patients with brain metastases from NYU Langone.
Simultaneous high-density EEG and eye-tracking recordings from 356 participants (47+ hours) for gaze estimation benchmarks.
Kastrati A et al. NeurIPS 2021 Datasets & Benchmarks
ECG, PPG, EEG/sleep, heart and lung sounds, bedside waveforms and consumer-wearable recordings.
48 half-hour two-lead ambulatory ECG recordings with beat-by-beat cardiologist annotations — the reference standard for arrhythmia-detection algorithms since 1980.
Moody GB, Mark RG. IEEE Eng Med Biol Mag 2001;20:45
21,799 clinical 12-lead, 10-second ECGs from 18,869 patients with SCP-ECG diagnostic, form and rhythm statements annotated by cardiologists.
Wagner P et al. Sci Data 2020;7:154
~800,000 diagnostic 12-lead ECGs from ~160,000 patients linked to MIMIC-IV, with machine measurements and cardiologist report text.
12-lead ECGs from >45,000 patients at Shaoxing People's Hospital with rhythm and diagnostic labels from expert physicians.
Zheng J et al. Sci Data 2020;7:48
345,779 12-lead ECGs from 233,770 Brazilian patients (a 15% sample of the CODE telehealth cohort) with labels for six abnormalities and mortality follow-up.
Ribeiro AH et al. Nat Commun 2020;11:1760
~88,000 annotated ECGs pooled from multiple international databases (CPSC, INCART, PTB, Georgia, Chapman, Ningbo) with SNOMED-CT diagnostic labels.
Perez Alday EA et al. Physiol Meas 2020;41:124003
8,528 short single-lead ECGs from AliveCor devices labelled normal, AF, other rhythm or noisy.
Clifford GD et al. Comput Cardiol 2017;44
Continuous single-lead ECG from 11,000 patients (up to two weeks each) with ~2.7 billion labelled beats — a huge resource for self-supervised ECG learning.
Tan S et al. arXiv 2019
Bedside-monitor waveforms (ECG, ABP, PPG, respiration) and numeric trends recorded from ICU patients, partly matched to the clinical databases.
53 eight-minute recordings of ECG, PPG and impedance pneumography with manual breath annotations, extracted from MIMIC-II.
Pimentel MAF et al. IEEE Trans Biomed Eng 2017;64:1914
197 whole-night polysomnographic recordings (EEG, EOG, chin EMG) with hypnograms scored by technicians — the standard sleep-staging benchmark.
Kemp B et al. IEEE Trans Biomed Eng 2000;47:1185
Polysomnography and clinical data from >40,000 sleep studies across major cohorts (SHHS, MESA, CFS, CHAT, MrOS, APPLES).
Zhang GQ et al. JAMIA 2018;25:1351
70 overnight single-lead ECG recordings with minute-by-minute apnoea annotations for ECG-based sleep-apnoea detection.
552 intrapartum CTG recordings (fetal heart rate and uterine contractions) with umbilical-artery pH and clinical outcomes from Brno University Hospital.
Chudáček V et al. BMC Pregnancy Childbirth 2014;14:16
3,000+ phonocardiogram recordings from multiple databases labelled normal or abnormal.
Liu C et al. Physiol Meas 2016;37:2181
5,272 heart-sound recordings from 1,568 children in Brazil with murmur annotations (timing, grade, shape) by experts.
Oliveira J et al. IEEE J Biomed Health Inform 2022;26:2524
920 annotated lung-sound recordings from 126 patients (healthy, COPD, asthma, pneumonia, etc.) with crackle and wheeze labels.
Rocha BM et al. Physiol Meas 2019;40:035001
25,000+ crowdsourced cough recordings with self-reported COVID-19 status and expert annotations on a subset.
Orlandic L et al. Sci Data 2021;8:156
Breathing, cough, vowel and counting audio from ~2,700 Indian participants with COVID-19 status and symptoms.
Bhattacharya D et al. Sci Data 2023;10:397
Wrist- and chest-worn multimodal sensor data (BVP, EDA, ECG, EMG, respiration, temperature, accelerometry) from 15 subjects during stress and amusement.
Schmidt P et al. ICMI 2018
Wrist PPG, accelerometer and chest ECG from 15 subjects during daily-life activities for motion-robust heart-rate estimation.
Reiss A et al. Sensors 2019;19:3079
Four months of Fitbit data (steps, sleep, heart rate, SpO2, temperature) plus surveys and ecological momentary assessments from 71 participants in the wild.
Yfantidou S et al. Sci Data 2022;9:663
Smartphone accelerometer and gyroscope signals from 30 volunteers performing six daily activities — a classic activity-recognition dataset.
4.9 million 30-second PPG segments from MIMIC-III ICU patients with rhythm labels, for PPG-based arrhythmia research.
100,000 12-lead ECGs from Columbia University Irving Medical Center paired with echocardiography-derived structural heart disease labels.
Poterucha TJ et al. Nature 2025
Voice and speech recordings with clinical data from participants with voice, neurological, respiratory and mood disorders (NIH Bridge2AI).
53,449 breathing, cough and voice recordings from 36,116 participants with COVID-19 test status and symptoms.
Xia T et al. NeurIPS 2021 Datasets & Benchmarks
Smartwatch accelerometer and gyroscope recordings from 469 participants (Parkinson's, differential diagnoses, healthy) during structured assessments.
AMP-PD: longitudinal CSF protein/peptide data for progression prediction; FoG: lower-back accelerometry from lab, home and daily-living settings for freezing-of-gait detection.
Facial videos with skin-tone measurements and reference heart rate for evaluating camera-based (remote PPG) heart-rate measurement across skin tones.
ECG and autonomic recordings from nine volunteers during propofol-induced loss and recovery of consciousness.
255 anonymised overnight polysomnography recordings from the Danish Center for Sleep Medicine with expert sleep stages.
Polysomnography from 25 healthy adults and 55 obstructive sleep apnoea patients, each scored by five sleep technologists.
Guillot A et al. IEEE Trans Neural Syst Rehabil Eng 2020;28:1955
Germline and cancer genomics, transcriptomics, proteomics, metabolomics, single-cell atlases and GWAS summary statistics.
Harmonised genomic and clinical data from TCGA (33 cancer types, ~11,000 patients), TARGET, CPTAC and other NCI programmes, with web portal and API.
Grossman RL et al. N Engl J Med 2016;375:1109
Interactive portal for exploring and downloading processed multidimensional cancer genomics datasets from 400+ studies, with clinical annotations.
Cerami E et al. Cancer Discov 2012;2:401
International Cancer Genome Consortium platform with clinically annotated tumour genomes from >100,000 patients (successor to the PCAWG/ICGC 25K data).
ICGC/TCGA PCAWG Consortium. Nature 2020;578:82
Catalogue Of Somatic Mutations In Cancer
Expert-curated database of somatic mutations in human cancer, including Cancer Gene Census, mutational signatures and drug-resistance mutations.
Tate JG et al. Nucleic Acids Res 2019;47:D941
Whole-genome sequences of 3,202 individuals from 26 populations, the reference catalogue of human genetic variation maintained by the International Genome Sample Resource.
1000 Genomes Project Consortium. Nature 2015;526:68
Genome Aggregation Database
Allele frequencies and constraint metrics from >800,000 exomes and genomes across diverse ancestries — essential for variant interpretation.
Karczewski KJ et al. Nature 2020;581:434
Variant browser and allele frequencies from >150,000 deeply sequenced genomes in NHLBI's TOPMed programme; individual-level data via dbGaP.
Taliun D et al. Nature 2021;590:290
Public archive of relationships between human genetic variants and phenotypes with supporting evidence and clinical-significance assertions from labs worldwide.
Landrum MJ et al. Nucleic Acids Res 2018;46:D1062
Curated collection of all published genome-wide association studies with >600,000 variant–trait associations and full summary statistics for many studies.
Sollis E et al. Nucleic Acids Res 2023;51:D977
Open database of published polygenic scores with variant weights, metadata and performance metrics for thousands of traits.
Lambert SA et al. Nat Genet 2021;53:420
GWAS summary statistics for ~2,400 disease endpoints from ~500,000 Finnish biobank participants linked to national health registries.
Kurki MI et al. Nature 2023;613:508
Multi-ancestry GWAS summary statistics for >7,000 phenotypes in UK Biobank across six continental ancestry groups — free, even though UK Biobank itself is fee-based.
Gene expression and eQTLs across 54 non-diseased tissues from ~1,000 post-mortem donors, showing how genetic variation regulates expression.
GTEx Consortium. Science 2020;369:1318
Encyclopedia of DNA Elements: functional genomics data (ChIP-seq, ATAC-seq, RNA-seq, Hi-C) mapping regulatory elements across human and mouse cell types.
ENCODE Project Consortium. Nature 2012;489:57
NCBI's public repository of >200,000 functional-genomics studies (microarray, RNA-seq, methylation, single-cell) submitted with publications.
Barrett T et al. Nucleic Acids Res 2013;41:D991
EMBL-EBI archive of functional-genomics experiments — Europe's counterpart to GEO — now served through the BioStudies platform.
The largest public repository of raw high-throughput sequencing reads, with cloud-hosted copies on AWS and Google Cloud.
database of Genotypes and Phenotypes
NIH archive of individual-level genotype, sequence and phenotype data from >1,500 studies (e.g. Framingham, TOPMed, GTEx raw data).
Mailman MD et al. Nat Genet 2007;39:1181
Repository of controlled-access human genetic and phenotypic data from biomedical studies worldwide (e.g. H3Africa, many cancer cohorts).
Freeberg MA et al. Nucleic Acids Res 2022;50:D980
Single-cell and spatial transcriptomics data from millions of cells across human tissues, contributing to a reference map of every cell type.
Regev A et al. eLife 2017;6:e27041
Curated, standardised collection of 1,000+ single-cell datasets (100M+ cells) with in-browser exploration and a Python/R Census API.
Maps of protein expression across human tissues, cells, organelles, blood and cancers using antibody-based imaging and transcriptomics.
Uhlén M et al. Science 2015;347:1260419
Mass-spectrometry proteomics and phosphoproteomics data from CPTAC tumour cohorts, harmonised and linked to GDC genomic data.
The world's largest repository of mass-spectrometry proteomics data, part of the ProteomeXchange consortium.
Perez-Riverol Y et al. Nucleic Acids Res 2022;50:D543
Cross-species, cross-technique repository for metabolomics experiments with raw spectra, metadata and reference metabolite information.
Yurekten O et al. Nucleic Acids Res 2024;52:D640
Metagenomic and 16S data from body sites of healthy adults (HMP1) and multi-omic disease cohorts (iHMP: IBD, prediabetes, pregnancy).
Human Microbiome Project Consortium. Nature 2012;486:207
Cancer Dependency Map: CRISPR/RNAi gene-dependency screens, drug sensitivity and multi-omic profiles for 1,000+ cancer cell lines (incl. CCLE).
Tsherniak A et al. Cell 2017;170:564
Genomics of Drug Sensitivity in Cancer
Drug-response data for ~1,000 cancer cell lines against hundreds of anticancer compounds, with genomic markers of sensitivity.
Yang W et al. Nucleic Acids Res 2013;41:D955
Integrates genetics, omics, literature and drug data to score and prioritise target–disease associations for drug discovery.
Ochoa D et al. Nucleic Acids Res 2023;51:D1353
Repository of immunology study data (clinical trials and mechanistic studies) from NIAID-funded research, curated into a standard model.
Bhattacharya S et al. Sci Data 2018;5:180015
Database of genomic variants and phenotypes from ~50,000 consented patients with rare developmental disorders, used for variant interpretation.
Firth HV et al. Am J Hum Genet 2009;84:524
Genome-wide association summary statistics for schizophrenia, bipolar disorder, depression, ADHD, autism and other psychiatric disorders.
100 million single-cell transcriptomic profiles of cancer cell lines under ~1,100 drug perturbations — the largest single-cell perturbation atlas.
Trial registries and platforms that share individual participant data from completed trials.
Registry and results database of >500,000 clinical studies in 220+ countries, with a modern REST API and bulk downloads.
Zarin DA et al. N Engl J Med 2011;364:852
Aggregate Analysis of ClinicalTrials.gov
A relational (PostgreSQL) copy of the entire ClinicalTrials.gov database refreshed daily, ideal for research on trial design and reporting.
Tasneem A et al. PLoS One 2012;7:e33677
International Clinical Trials Registry Platform
Single search portal over primary trial registries worldwide (incl. CTRI, ChiCTR, EU CTR, ISRCTN, ANZCTR, PACTR).
Independent non-profit platform to search and request de-identified individual participant data from 7,000+ trials contributed by pharma, academia and funders (now also hosting GSK's former CSDR requests).
Bierer BE et al. N Engl J Med 2016;374:2411
Yale Open Data Access Project
Independent review and sharing of individual participant data from trials contributed by Johnson & Johnson, Medtronic, Procter & Gamble and others.
Open platform sharing patient-level data from oncology trial comparator (and some experimental) arms, with built-in analytics tools.
Individual-level data from completed NCI-funded cancer clinical trials run by US cooperative groups.
De-identified data from completed trials of the NIDA Clinical Trials Network on substance-use disorder treatments.
Clinical notes corpora, biomedical QA benchmarks, the literature itself and the standard vocabularies that tie data together.
Annotated de-identified clinical notes from the i2b2/n2c2 shared tasks: de-identification, medication extraction, temporal relations, obesity, smoking status, emrQA and more.
14,049 clinician-annotated natural language inference pairs built from MIMIC-III notes.
Romanov A, Shivade C. EMNLP 2018
Annual XML baseline plus daily updates of 37 million+ biomedical citations and abstracts — the raw material for literature mining and biomedical language models.
Millions of full-text journal articles from PubMed Central licensed for reuse and text mining, in XML and plain text.
COVID-19 Open Research Dataset: ~1 million scholarly articles on COVID-19 and coronaviruses with full text, released by AI2 (final release 2022).
Wang LL et al. ACL NLP-COVID Workshop 2020
Multiple-choice questions from US, Chinese and Taiwanese medical licensing exams — the most common benchmark for medical LLMs.
Jin D et al. Appl Sci 2021;11:6421
194,000 multiple-choice questions from Indian AIIMS and NEET-PG entrance exams spanning 21 medical subjects with explanations.
Pal A et al. CHIL 2022 (PMLR 174)
Biomedical research question-answering dataset (yes/no/maybe) built from PubMed abstracts, with 1,000 expert-labelled instances.
Jin Q et al. EMNLP-IJCNLP 2019
Annual challenge datasets for large-scale semantic indexing (MeSH) and biomedical question answering with expert-curated gold answers.
~5,000 sample transcribed medical reports across 40 specialties — handy for teaching clinical NLP without any data-use agreement.
Integrates 200+ biomedical vocabularies (SNOMED CT, ICD, MeSH, RxNorm, LOINC…) into concepts with cross-mappings, plus the Semantic Network and SPECIALIST lexicon.
Bodenreider O. Nucleic Acids Res 2004;32:D267
WHO's current International Classification of Diseases with browser, coding tool and a free API for integration in health information systems.
Universal code system for laboratory tests, clinical measurements and documents, required by many interoperability standards.
Medical Subject Headings
NLM's controlled vocabulary thesaurus used to index PubMed, available as XML, RDF and via API.
Adverse-event reports, drug labels, drug–target and side-effect knowledge bases, and prescribing data.
Public APIs and bulk downloads for FDA data: adverse event reports (drugs, devices, food), recalls, drug labels and NDC directory.
Kass-Hout TA et al. JAMIA 2016;23:596
FDA Adverse Event Reporting System
Quarterly raw files of spontaneous adverse-event and medication-error reports submitted to FDA since 2004 — the main source for disproportionality analyses.
Vaccine Adverse Event Reporting System
US national early-warning system collecting reports of adverse events after vaccination, with downloadable raw data since 1990.
Public access to aggregated suspected adverse-reaction reports for medicines authorised in the European Economic Area.
Public window into VigiBase, the WHO global database of 35 million+ individual case safety reports from 150+ countries.
Data-mined drug side-effect (OFFSIDES) and drug–drug interaction (TWOSIDES) signals derived from FAERS with confounding correction.
Tatonetti NP et al. Sci Transl Med 2012;4:125ra31
Side effects and indications of marketed medicines extracted from drug package inserts and mapped to MedDRA.
Kuhn M et al. Nucleic Acids Res 2016;44:D1075
Comprehensive drug knowledge base: chemistry, targets, pharmacology, interactions and pathways for ~16,000 drugs.
Knox C et al. Nucleic Acids Res 2024;52:D1265
Manually curated database of bioactive drug-like molecules with >20 million bioactivity measurements against targets.
Zdrazil B et al. Nucleic Acids Res 2024;52:D1180
The world's largest free chemistry database: compounds, substances, bioassays, patents and safety information.
Kim S et al. Nucleic Acids Res 2023;51:D1373
Official FDA-approved structured product labels (package inserts) for prescription and OTC drugs, downloadable in SPL XML.
Normalised names and identifiers for clinical drugs linking many drug vocabularies, with the free RxNav APIs.
Nelson SJ et al. JAMIA 2011;18:441
The Anatomical Therapeutic Chemical classification and Defined Daily Doses used worldwide for drug-utilisation research.
66 AI-ready datasets and benchmarks spanning target discovery, ADMET, drug–drug interactions, clinical trial outcomes and more.
Huang K et al. NeurIPS 2021 Datasets & Benchmarks
1,036 FDA-approved drugs ranked by risk of drug-induced liver injury from labelling and literature.
Chen M et al. Drug Discov Today 2016;21:648
LLM-extracted, clinician-checked toxicity summaries and ratings across eight organ systems for 2,418 FDA-approved drugs.
Pathogen genomes, surveillance line lists, AMR, tuberculosis, malaria, influenza and historical notifiable-disease data.
Country-reported TB notifications, treatment outcomes, drug resistance, TB/HIV, financing and WHO incidence estimates since 2000.
Weekly virological surveillance data on influenza (and SARS-CoV-2/RSV) from national influenza centres in 150+ countries since 1995.
Geospatial estimates of malaria prevalence, incidence and intervention coverage plus parasite-rate and vector-occurrence surveys, with an R package.
Global Antimicrobial Resistance and Use Surveillance System
Country-reported surveillance data on resistance in priority bacterial pathogens and antimicrobial consumption, with an interactive dashboard.
Interactive maps and trends of antibiotic resistance and antibiotic consumption for countries worldwide, including the Drug Resistance Index.
Industry AMR surveillance datasets (e.g. Pfizer ATLAS, GSK SOAR, J&J, Paratek) with millions of isolate-level susceptibility results.
Bacterial and Viral Bioinformatics Resource Center
Integrated genomes, AMR phenotypes, protein families and analysis tools for bacterial and viral pathogens (merger of PATRIC, IRD and ViPR).
Olson RD et al. Nucleic Acids Res 2023;51:D678
Real-time clustering of bacterial and fungal pathogen genomes from food, environment and patients, with AMR gene calls (AMRFinderPlus).
Portal for searching and downloading viral sequences (incl. SARS-CoV-2, influenza, dengue, mpox) with metadata.
Global initiative sharing influenza, SARS-CoV-2, RSV, mpox and other virus genomes with associated epidemiological metadata.
Shu Y, McCauley J. Euro Surveill 2017;22:30494
Real-time phylogenetic tracking of pathogen evolution (SARS-CoV-2, influenza, mpox, dengue, measles…) with open curated datasets and tools.
Hadfield J et al. Bioinformatics 2018;34:4121
Digitised US notifiable-disease reports since 1888 and global disease counts standardised for research on historical epidemics and vaccination impact.
van Panhuis WG et al. N Engl J Med 2013;369:2152
API of real-time US epidemic indicators — cases, hospitalisations, claims, symptom surveys, influenza-like illness — from Carnegie Mellon's Delphi group.
Reinhart A et al. PNAS 2021;118:e2111452118
The iconic global daily COVID-19 case and death time series (Jan 2020–Mar 2023) that powered the Johns Hopkins dashboard.
Dong E et al. Lancet Infect Dis 2020;20:533
Interactive EU/EEA surveillance data for ~50 communicable diseases including AMR (EARS-Net), with downloadable tables.
Open, de-identified line-list case data for outbreaks (COVID-19, mpox, Marburg, H5N1) curated from public sources.
350,000+ waiting-room thermal images and 21 million audio samples (coughs, speech) from a university clinic for influenza-like-illness surveillance.
Al Hossain F et al. Proc ACM IMWUT 2020;4:1
Daily index of COVID-19 protective policy stringency for US states and other jurisdictions during 2020.
General platforms that host or index many health datasets — good places to keep searching.
The home of MIMIC, eICU and hundreds of open and credentialed physiological, clinical and imaging datasets, plus annual Computing in Cardiology challenges.
Goldberger AL et al. Circulation 2000;101:e215
Long-running repository of 600+ curated datasets, including dozens of classic clinical ones (breast cancer Wisconsin, heart disease, diabetes, CKD, hepatitis).
Platform hosting hundreds of biomedical image-analysis challenges with downloadable datasets and leaderboard evaluation.
Collaborative research platform hosting data for DREAM challenges, AMP-AD, BraTS and many consortia, with fine-grained access controls.
Community data platform with thousands of medical datasets and competition data (RSNA, SIIM, APTOS, PANDA) plus free GPU notebooks.
The world's largest archive of social and behavioural science data, including health and ageing collections (NACDA) and many NIH-funded studies.
UK's largest collection of social, economic and population data — including ELSA, Understanding Society, Health Survey for England and birth cohorts.
Cloud-hosted public datasets you can analyse in place — e.g. gnomAD, 1000 Genomes, TCGA open tier, CAMELYON, EMBED, NIH CXR.
CERN-run general-purpose open repository where many medical datasets (TotalSegmentator, CODE-15%, COUGHVID…) are deposited with DOIs.
Python package bundling curated medical and public-health datasets (cancer, chronic disease, hospital management, COVID-19, environmental health) for teaching and analysis.
The largest hub for ML datasets, including thousands of medical text, imaging and benchmark datasets with one-line loaders.
Open research data repository with thousands of medicine, health and life-science datasets deposited with DOIs.
Generalist repository hosting many published clinical and biomedical datasets (e.g. Kermany OCT, PAD-UFES-20).
Datasets built to train or evaluate medical AI: clinical QA and reasoning, report generation, error detection, safety and bias tests, and image–text corpora.
525 physician-authored clinician tasks with rubrics for evaluating how well LLMs support professional clinical work.
840 AI-generated medical texts (reports, summaries, notes) graded by physicians for factual errors and patient-safety risk.
Questions, model responses and rubric scores used to benchmark health AI systems: 500 questions across five systems plus 4,023 English-language questions.
300 physician-written clinical tasks in a simulated FHIR-based EHR environment for benchmarking LLM agents that act on patient records.
Jiang Y et al. arXiv 2025
983 clinician-written instructions over 276 longitudinal EHRs with 303 reference responses, for evaluating LLMs on real EHR tasks.
Fleming SL et al. AAAI 2024
Longitudinal structured EHR data for 6,739 Stanford patients with 15 few-shot prediction tasks, released with a pretrained foundation model.
Wornow M et al. NeurIPS 2023 Datasets & Benchmarks
405 MIMIC-IV-derived examples for benchmarking extraction of ICD-10 diagnosis concepts from clinical text.
Patient questions paired with clinician-rewritten questions and grounding EHR note excerpts, from the BioNLP 2025 shared task on answering patient questions from records.
13,070 propositions from human- and LLM-written brief hospital courses (MIMIC-III) annotated for factual support against the EHR.
2,168 clinical notes from four types with LLM fact decompositions and entailment labels for evaluating factuality in clinical text.
Munnangi M et al. arXiv 2024
3,848 clinical texts with injected or real medical errors for benchmarking error detection and correction by LLMs (MEDIQA-CORR 2024).
Ben Abacha A et al. arXiv 2024
Ambient Clinical Intelligence benchmark: 207 doctor–patient conversation transcripts paired with visit notes for note-generation research.
Yim W et al. Sci Data 2023;10:586
1,700 short doctor–patient conversations with corresponding summaries (clinical note sections) for dialogue summarisation.
Ben Abacha A et al. EACL 2023
57 mock primary-care consultations with audio, transcripts and clinician-written notes for speech recognition and note generation.
Korfiatis AP et al. ACL 2022
~167,000 patient summaries extracted from PubMed Central case reports with patient–article and patient–patient relevance labels for retrieval tasks.
Zhao Z et al. Sci Data 2023;10:909
Multimodal case-report dataset from PubMed Central: 75,382 articles, 96,428 cases and 135,596 images with captions and labels.
Synthetic clinical notes for 70 fictional patients, released by NHS England for safe NLP development.
Collection of Indian clinical datasets for medical speech recognition, health-record parsing and clinical note generation.
Cardiology test reports (ECG, echo, Holter, CMR, CPET) released with Google's AMIE cardiology study.
2,658 fictional radiology reports written by radiologists in 14 languages, for multilingual radiology NLP.
Radiologist-annotated entities and relations in chest X-ray reports (RadGraph: 500 reports; RadGraph2 adds change-over-time relations), plus large auto-annotated sets.
Jain S et al. NeurIPS 2021 Datasets & Benchmarks
Radiologist annotations of clinically significant errors in 100 AI-generated chest X-ray reports.
Clinically meaningful synthetic errors injected into MIMIC-CXR reports for training and evaluating report-error detection.
Scene graphs linking anatomy, findings and comparisons for MIMIC-CXR images, with a 500-patient gold standard.
Eye-tracking data, dictated reports and abnormality ellipses from radiologists reading 3,032 MIMIC-CXR chest X-rays.
Bigolin Lanfredi R et al. Sci Data 2022;9:350
Eye-tracking and report dictation from a radiologist reading 1,083 MIMIC-CXR images.
Karargyris A et al. Sci Data 2021;8:92
Radiologist think-aloud reasoning traces and eye-tracking recorded while interpreting chest X-rays.
Experimental data from 227 radiologists reading chest X-rays with and without AI assistance, for studying human–AI collaboration.
697,000 radiology image–text pairs (400,042 chest X-ray pairs) assembled from public sources to train a small multimodal radiology model.
Zambrano Chaves JM et al. arXiv 2024
3,515 clinician-generated question–answer pairs about 315 radiology images, the first manually built medical VQA set.
Lau JJ et al. Sci Data 2018;5:180251
Medical visual question answering with image-grounded reasoning traces and a rubric-scored test split.
Multilingual wound-care visual question answering: 477 patient cases with images, queries and expert responses (MEDIQA-WV).
119,000 synthetic medical images with VQA annotations for training multimodal models without patient data.
24 million image–caption pairs from 6 million PubMed Central open-access articles for training biomedical vision-language models.
Lozano A et al. arXiv 2025
3.4 million image–mask–label triples across nine imaging modalities used to train the BiomedParse segmentation foundation model.
Zhao T et al. Nat Methods 2025;22:166
208,414 pathology images paired with natural-language descriptions curated from medical Twitter, used to train the PLIP model.
Huang Z et al. Nat Med 2023;29:2307
30,000 ECG instruction pairs with grounded, heartbeat-level interpretations for training explainable ECG language models.
1,031 hours of biomedical video–text instruction data mined from educational videos, with SurgeryVideoQA and MIMICEchoQA benchmarks.
6,245 test samples benchmarking video-language models on surgical and medical video understanding.
Open benchmark of questions on medical concepts (ICD-9/10, ATC and procedure codes) at three difficulty levels.
Shoham OB, Rappoport N. arXiv 2024
Seven adversarial question sets for surfacing health-equity-related biases and harms in medical LLM answers.
Pfohl SR et al. Nat Med 2024;30:3590
Benchmark of reasoning and memory tests built from multinational medical exams for measuring hallucinations in LLMs.
Pal A et al. CoNLL 2023
382 clinician-crafted adversarial prompts with 1,146 ChatGPT responses labelled for safety, privacy, hallucination and bias.
Clinical vignettes on pain management designed to measure social bias (race, gender) in medical QA systems.
Expert-annotated oncology progress notes (breast and pancreatic cancer) for evaluating LLM information extraction.
Sushil M et al. NEJM AI 2024
1,000 consumer health questions paired with expert summaries for question summarisation.
Ben Abacha A, Demner-Fushman D. ACL 2019
~1 million question–logical form and 400,000 question–answer pairs generated from n2c2 clinical notes for EMR question answering.
Pampari A et al. EMNLP 2018
1,000 Spanish clinical case reports manually coded with ICD-10 diagnoses and procedures (CLEF eHealth 2020).
9,845 synthetic German medical sentences with named-entity annotations generated by GPT (not expert-validated).
Frei J, Kramer F. J Biomed Inform 2023;144:104478
600 PubMed abstracts annotated for biomedical entities and eight relation types for relation extraction.
Luo L et al. Brief Bioinform 2022;23:bbac282
945 sentences from 911 PubMed abstracts manually annotated with pharmacogenomic entities and relations.
Legrand J et al. Sci Data 2020;7:3
Community library of 126+ biomedical NLP datasets in a uniform schema with Hugging Face loaders.
Fries JA et al. NeurIPS 2022 Datasets & Benchmarks
Open-access subset of the BRIDGE benchmark: 55 multilingual clinical text tasks from real-world notes for evaluating LLMs.
52,000 diverse medical instruction–response pairs generated with GPT-4 for instruction-tuning (AlpaCare).
Zhang X et al. arXiv 2023
410,000 biomedical instructions with preference data, used to train the UltraMedical model family.
Zhang K et al. NeurIPS 2024 Datasets & Benchmarks
~2.2 million medical reasoning samples for supervised fine-tuning of reasoning LLMs.
~200,000 synthetic medical reasoning conversations generated with GPT-OSS-120B.
~200,000 biomedical instruction samples (NER, RE, NLI, QA, classification) for instruction-tuning LLMs.
2.5 million synthetic clinical instruction rows spanning many clinical text tasks and output formats.
Healthcare-focused pretraining corpus of ~65 billion tokens curated from web and literature sources.
24,665 NHS-derived Q&A pairs, 2,354 conversations and 4,688 medical task pairs generated with the OpenGPT framework.
Maternal-health questions and answers in four languages for evaluating multilingual health chatbots.
204,855 clinical trials with relevance judgements for patient-to-trial matching and search evaluation.
Test collections from TREC Clinical Decision Support, Precision Medicine and Clinical Trials tracks (topics, documents, relevance judgements).
~479,000 clinical trials with automatically derived weak outcome labels for trial-outcome prediction.
Curated list of FDA clearances for AI-based image-interpretation (CAD) devices with evidence characteristics.
Gastrointestinal endoscopy, capsule endoscopy and surgical video datasets, including robot demonstrations.
Gastrointestinal endoscopy image and video collections from Norway: 8,000 labelled images (v2), polyp segmentation masks (SEG), 4.7M video-capsule frames (Capsule) and 110,000+ images (HyperKvasir).
Pogorelov K et al. ACM MMSys 2017
21 colonoscopy videos (5,525 frames) labelled with the Boston Bowel Preparation Scale for bowel-cleansing assessment.
Pogorelov K et al. ACM MMSys 2017
18,481 small-bowel capsule endoscopy images with bounding-box annotations for lesion detection.
4,000+ surgical videos (938 hours) across many procedure types for pretraining surgical video foundation models.
~240,000 surgical video clip–caption pairs for surgical vision-language pretraining; the open subset SurgLaVi-β is freely released.
778 hours of surgical-robot demonstration episodes released with NVIDIA's GR00T-H healthcare robotics model.
Large catalogues and search tools that index thousands of datasets held elsewhere. The catalogues are free to use, but some datasets they list charge fees, so check each dataset's terms.
UK national catalogue of 1,400+ health datasets (NHS records, cohorts, biobanks, trials, registries) with standardised metadata, search filters and a common data-access request workflow.
Community-curated list of ~300 medical AI datasets, benchmarks and resources (imaging, EHR, NLP, LLM evaluation, wearables) maintained by the BrainX AI community.
IHME's searchable catalogue of the world's surveys, censuses, vital registration and disease registries, linking to source data wherever available.
Search engine indexing dataset descriptions across thousands of repositories — useful for finding datasets not listed here.
Curated list of wearable-sensor and human-movement datasets (accelerometry, biomechanics, gait) such as AddBiomechanics and NHANES accelerometry.
No databases match. Try a broader term or .