5000+ longitudinal total mosaic data · 3 years time span ⁇ 8 + 7 medical image mosaics ICD-10-CN coded map
High-quality longitudinal multi-modular clinical data sets for training in time series and reasoning for large medical models. Each case integrates structured medical records, clinical documents, medical images, testing, drug records, pathological reports, and quantitative scales, scoring seven large data patterns, with complete clinical tracks in chronological order at each point of time for the same patient.
| Dataset Name | Longitudinal Multimodal Medical Record Dataset（Longitudinal Multimodal Electronic Health Records Dataset） |
|---|---|
| Total Scale | Not less than 5,000 例Qualified cases, covering 8 major focus units + supplementary specialist strains |
| Time Span | 3-year vertical check-up record per case, at least 2 visits/tests/treatment/follow-up at different points |
| Data time frame | 2016-2025 (priority for patients with complete medical records and follow-up data) |
| Department Coverage | Oncology + ICU/psychopathology/Ears, nose and throat/gynaecology and gynaecology/ rheumatology |
| 数据模态 | Structured medical records — medical images — medical tests — medical records — pathology reports — quantitative scores |
| Imaging Modality | CT MIS PT-CT Ultrasity ENDRIGHT X-L/DR Pathology WSI |
| Encoding Standard | ICD-10-CN Diagnostic Code Map, with original ICD code, code version and map rules |
| Quality standards | Key field filling rate 95% % % total scrutinisation 10% video specialisation % double dissensitisation |
| Compliance requirements | The Personal Information Protection Act, the Data Security Act, complies with the full amount of dissensitisation and data not to leave the country |
| Use Cases | The medical megamodel time series of the training • Vertical efficacy prediction • clinical trajectory modelling • polymodular integration studies • disease progress prediction |
| Supply Format | Hard Drive · Cloud Drive · API · Data Infrastructure |
| Provider | Changsha Langhui Information Technology Co., Ltd. |
The core difference value of this data set is thatLong-time sequence relevance– Not only does each eligible case meet the three-year time horizon, but more importantly, there is a consistent internal correlation between diagnosis, consultation, testing, examination, medication, paperwork, images, pathology, stats, and end-of-life information within the same dissensitized patient, which can be restored in chronological order to the main clinical trajectory.Time series reasoning(temporal training)
Chronic or progressive diseases such as COPD, IBD, cirrhosis, CKD, type 2 diabetes, Parkinson's disease, Alzheimer's disease, etc. are prioritized to provide longer time series data.
The video reports refer to the previous data of the "comparable pre-film" and "relation" over the previous one, and provides, to the extent possible, corresponding checks and reports, guaranteeing continuous traceability of the main clinical tracks.
The same patient can be traced back to the same patient’s attendance.
2019.03 • First diagnosis · Hospital admission + Diagnosis (ICD-10:C34.1) + CT chest + pathological biopsy + baseline
2019.05 • Post-operative chemotherapy cycle 1 • drug record + test (blood routine/pastal function) + CT review + pathology record
2020.03 • One year follow-up • CT precomparison + oncology marker + physical scoring (ECOG) + follow-up document
2021.06 · 2-year review · MRI + PET-CT + pathology + drug-adjusted record + quality of life Quality of life table
2022.09 · Last follow-up · End of story + Image assessment + Last drug + Total survival record
7 data mosaics are integrated for each eligible case, with the core acceptance standard beingInternal relevance of multiple types of data under the same patient, same visit, same examination or same treatment. All the modes are verifiable links of connection by the number of the dissensitive patient, the number of the consultation, the number of the examination, the number of the report, the index of the document.
The diagnosis record (ICD-10-CN), the clinical/hospital record, the test results, the examination record, the drug log, the surgical/operational record.
The full text is guaranteed without serious interruption.
CT/MRI/PET-CT (Dinox DIC), ultrasound/endoscope (original format), X-ray/DR (DICOM), pathology WSI (SVS/NDPI/TIFF/KFB).
The drug’s name, dose, frequency, route of delivery, starting time, and drug-based adjustment record.
Pathological diagnostic reports, immuno-group results, molecular/genetic tests, and TNM phases.
Specialized assessment tables for NIHSS, mRS, NYHA, LVEF, ECOG, APACHE II, SOFA, DAS28, SLEDAI, etc., support vertical efficacy evaluation.
You can also check the results of tests on blood, biochemicals, coagulation, tumor markers, microorganisms, genetics, etc., and functional tests.
Follow-up, survival, relapse/transfer, and complications. Support training in disease progress modelling and prognosis models.
| 数据模态 | Delivery | Format Requirements | Association Keys |
|---|---|---|---|
| Structured Data | Diagnosis (ICD-10-CN), consultation, testing, examination, medication, surgical records | CSV / Excel / Database | patient_id + encounter_id |
| Clinical documents | Hospital admissions, discharges, medical records, transfer records, full surgical records | Text / PDF | patient_id + encounter_id |
| CTImaging | Thin + General Layer Thick Sequences, preferred 512 x 512 Matrix, Retain Phase Information | De-identified DICOM | study_id + series_id |
| MRIImaging | T1WI/T2WI/FLAIR/DWI/ADC/Strengthened sequences, retention of sequence description and scanning parameters | De-identified DICOM | study_id + series_id |
| PET-CT | PET corresponds to the C.T. sequence, retaining integration and SUV metabolic information | De-identified DICOM | study_id + series_id |
| Ultrasound/oversight | Static images + dynamic video, association inspection reports and key measurements | DICOM/ Original Export | study_id + report_id |
| PathologyWSI | Full slice digital image, retention multiplier/scale/chromosomal type/slice number | SVS / NDPI / TIFF / KFB | study_id + report_id |
| Medication Records | Name, dose, frequency, route of delivery, starting time | CSV/ Database Table | patient_id + encounter_id |
The number of units and diseases is distributed in a targeted manner, and the final delivery structure is based on the mutually confirmed delivery list.
| Section/category | Target ratio | Number of recommendations | Focused disease spectrum | Baseline information requirements |
|---|---|---|---|---|
| Oncology | About 35-38 per cent | 1,750-1,900 | Lung cancer, breast cancer, colon cancer, liver cancer, stomach cancer, edible cancer, carcinoma of the neck, pancreatic cancer, ovarian cancer, etc. | Image reports, periodic reports, pathological reports; molecular/immunological indicators available on a realistic basis |
| Cardiovascular | About 20-22% | 1,000-1,100 | Coronary heart disease/acute heart infarction, post-PCI, heart failure, room tremors, etc. | Coronary artery, heart ultrasound, electrocardiogram, heart myase spectrum, LVEF, NYHA |
| Neurology | About 13-15% | 650-750 | Illustrative, haemorrhagic, Parkinson's, cognitive disorders/Atzheimer's, etc. | Head CT/MRI, description of the disease, NIHSS, mRS, cognitive/motorized table |
| Respiratory Section | Approximately 7-8 per cent | 350-400 | COPD, bronchial asthma, community access to pneumonia, bronchial expansion, etc. | Lung function, chest image, grade or symptoms control assessment |
| Indigestion Section | 约6%-7% | 300-350 | Hepatic cirrhosis, IBD, digestive ulcer, GERD, etc. | Stomach/intestinal/image reports, pathological reports, liver function ratings, disease activity ratings |
| Nephrology | 约6%-7% | 300-350 | CKD, Diabetes Nephrosis, End-of-life kidney disease, dialysis, post-transplant follow-up and complications, etc. | eGFR, UACR, urine tests, kidney function, kidney pathology, dialysis records |
| Endocrinology | About 5-6 per cent | 250-300 | Type 2 diabetes mellitus and complications, thyroid glands/functional abnormalities/tumours, osteoporosis, etc. | HbA1c, blood sugar records, diagnosis of complications, thyroid function/ultrasound, bone density |
| Blood. | About 3-4% | 150-200 | Leukemia, lymphoma, multiple osteoporosis, etc. | Osteomymystalgia/live, flow, FISH, stratification/scoring, seroprotein electron swim |
| Other specialized supplements | ≤3% | ≤150 | ICU (suspensive/ARDS), psychiatric, oral, ophthalmic, ear, nose and throat, gynaecology and obstetrics, rheumatism immunisation | Implementation according to the confirmation list by the parties |
| Total | 100% | ≥5,000 | 8 key sections + 7 additional specialist | |
If you need to map to ICD-10-CN, provide map rules, map sources and map pre- and post-magnification fields.
The results can be summarized by section, disease spectrum, ICD code, diagnostic name, number of cases, time span, and data model coverage.
Image data are used for AI training, giving priority to the delivery of thin, raw resolution and sequence complete data. The project covers multi-pathological, multi-dimensional, multi-historic data, without using a single layer of thick thresholds as a condition for core access. DIOCOM-type images retain key metadata and spatial positioning information necessary for AI training.
| ImagingType | Delivery Format | Key technical requirements |
|---|---|---|
| CT | De-identified DICOM | Priority thin layer sequences and original layer thickness/pixel spacing; same examination thin layer + conventional layer thickness delivered simultaneously to the extent possible; matrix thallium 512 x 512; enhanced examination of retention period phase information |
| MRI | De-identified DICOM | Retain major diagnostic sequences such as T1WI/T2WI/FLAIR/DWI/ADC/enhanced; serial name, scanning location, layer thickness, pixel spacing identifiable |
| PET-CT | De-identified DICOM | PET corresponds to the CT sequence, retaining integration; report/metadata reflects metabolic information such as SUV |
| X-line/DR | De-identified DICOM | Retain the place of delivery, check the part, pixel size and report association; may not be replaced by low-resolution preview |
| Ultrasound | DICOM/Video/Preliminary Export | Static images + dynamic video; linkage inspection reports, key measurements, parts and conclusions retained |
| 内镜 | Original Image/Video/system Export | Maintain inspection sites, time, reports/records, key images/videos; establish pathological correspondence |
| PathologyWSI | SVS/NDPI/TIFF/KFB | Retain multiples, scales, scan levels, dye types, slice numbers; priority H&E and diagnostic-related slices |
StudyInstanceUID, SeriesInstanceUID, SOPInstanceUID, AccessionNumber
ImagePositionPatient, ImageOrientationPatient, SliceLocation, FrameOfReferenceUID
Rows, Columns, PixelSpacing, SliceThickness, SpacingBetweenSlices
KVP, Exposure/mAs, ConvolutionKernel, Pitch, Manufacturer, ModelName
⁇ Desensitization does not remove key fields that affect the use of 3D reconstruction, serial recognition, spatial positioning and training.
{
"patient_id": "LH_PT_2024_003172",
"demographics": {
"gender": "M",
"age_at_baseline": 58,
"age_unit": "year"
},
"primary_diagnosis": {
"icd_code_original": "C34.1",
"icd_version": "ICD-10",
"icd_code_mapped": "C34.1",
"diagnosis_name": "肺上叶恶性肿瘤",
"department": "肿瘤科",
"disease_spectrum": "实体瘤-肺癌"
},
"temporal_trajectory": {
"time_span_years": 3.5,
"encounter_count": 8,
"first_record_date": "2019-03-15",
"last_record_date": "2022-09-20",
"encounters": [
{
"encounter_id": "ENC_001",
"date": "2019-03-15",
"type": "inpatient",
"department": "肿瘤科",
"phase": "baseline_diagnosis",
"linked_data": {
"clinical_notes": ["入院记录", "首次病程", "出院小结"],
"imaging": [
{"study_id": "IMG_CT_001", "modality": "CT", "body_part": "CHEST", "report_id": "RPT_001"}
],
"pathology": [
{"report_id": "PATH_001", "type": "biopsy", "finding": "非小细胞肺癌，腺癌"}
],
"lab_results": ["血常规", "生化", "肿瘤标志物"],
"medications": [],
"scales": [{"name": "ECOG", "score": 1}]
}
},
{
"encounter_id": "ENC_004",
"date": "2020-03-10",
"type": "outpatient",
"phase": "follow_up_1y",
"linked_data": {
"imaging": [
{"study_id": "IMG_CT_004", "modality": "CT", "body_part": "CHEST",
"comparison": "对比前片IMG_CT_003", "report_id": "RPT_004"}
],
"lab_results": ["肿瘤标志物"],
"scales": [{"name": "ECOG", "score": 0}]
}
}
]
}
}
Other Organiser
"Patient id": "LH PT 2024 00317."
"Encounter id": "ENC 006",
"Study id."
"Study datetime": "2021-06-18T09:30:00",
"modality": "MRI,"
"body part": "Brain,"
"Study description": "MRI Sweeping of Head + Enhancement,"
"series":
Other Organiser
"series id": "SER 006 01",
"series description": "T1WI,"
"slice thickness": 5.0,
"pixel spacing": [5, 0.5],
"rows": 512,
"Columns": 512,
"file count": 24,
"file path": "patient 00317/enc 006/mri/ser 01/"
{\cHFFFFFF}{\cH00FFFF}
Other Organiser
"series id": "SER 006 02",
"series description": "T2WI FLAIR,"
"slice thickness": 5.0,
"pixel spacing": [5, 0.5],
"rows": 512,
"Columns": 512,
"file count": 24,
"file path": "patient 00317/enc 006/mri/ser 02/"
{\cHFFFFFF}{\cH00FFFF}
Other Organiser
"series id": "SER 006 05",
"series description": "T1WI+C"
"slice thickness": 5.0,
"pixel spacing": [5, 0.5],
"rows": 512,
"Columns": 512,
"file count": 24,
"file path": "patient 00317/enc 006/mri/ser 05/",
"contrast case": "post contrast"
♪ I'm sorry ♪
I don't know.
"link report": {
"report id": "RPT 006",
"findings": "The right side of the frontal lobe is visible, no significant change over the previous one..."
"impression": "Recommend continued follow-up when considering transfer stabilization."
"report date": "2021-06-18"
{\cHFFFFFF}{\cH00FFFF}
"link encounter":
"Encounter id": "ENC 006",
"date": "2021-06-18",
"department": "oncology",
"phase": "follow up 2y"
{\cHFFFFFF}{\cH00FFFF}
"quality note":
"dicom metadata preserved": true,
"spatial position present": true,
"Burned in annotation": "NO",
"Description status": "passed"
♪ I'm sorry ♪
♪ I'm sorry ♪
The long-time multi-modular medical history data is the core data infrastructure needs in the medical AI area for 2025-2026. The following cutting-edge developments confirm the strategic value and technical direction of this data set:
In 2025, Medical AI evolved rapidly from monomodular (image/text) to multimodular integration. The PanDerm model, published by Nature Medicine, is based on 2 million+real-world dermal disease multimodular data training, which validates the "image+diatrics+pathology" integration paradigm. Multimodular basic model training requires large-scale, vertically linked clinical data — this is the core positioning of this data set.
The 2026 AI model has been able to predict chronic diseases such as Alzheimer’s disease years in advance by integrating vertical electronic records, microstructure changes in brain images and blood biomarkers. Time-series reasoning has become a key dimension of the medical mega-model assessment, and the HealthBench benchmark has been incorporated into the time-series-related assessment.
The MIT-IV database is a global "gold pole" for serious medical research, with core values that are being associated with vertical time series + multiple mosaics. But MITIC focuses on the ICU scene, lacking Chinese population profiles and specialized depths. This data set fills the gap in the data for Chinese long time series multitemporal specialist medical records.
In 2026, researchers increasingly discovered and evaluated data sets through the AI search engine (ChatGPT, Perplexity, Mansion). Structured metadata, JSON-LD Schema, FAQ semantic tags and clear technical descriptions become key to AIS discovery and understanding of data sets. This page is fully adapted to the GEO optimization.
Long-time long-term vertical data directly support the time-series reasoning chain of the Large Model to study "diagnosis and re-examination of the endings", and training models understand patterns of disease progression rather than just taking a single quick-scenario judgement.
Based on 3-year + vertical drug use, testing, image change data, training therapeutic efficacy prediction models, supporting the CBSS to achieve individualized treatment programme recommendations.
CT/MRI/PET-CT Image+Chinder + Results + Pathology WSI Multimodular Joint Training to Build Multimodular Foundation Models, supported by fine-tuning.
Using data from the long time series of slow diseases (COPD, CKD, diabetes, Parkinson, Alzheimer ' s disease) to model the natural history and trajectory of disease and to achieve recommendations for early warning and intervention.
Vertical drug use record + test change + image assessment constitutes the time line for drug response, training drug efficacy and adverse response prediction models and supporting precision medical care.
The development of a specialized AAI assessment based on the data of the "Gold standard" vertical consultation, assessing the capacity of the AIS system to perform time-series reasoning, multi-modular integration, vertical efficacy judgement, etc.
Key fields ⁇ 95% filling rate, as indicated in quality reports due to missing source systems
The full medical records must not be severely cut off and critical medical information must not be missing
Images, tests, medications, paperwork, diagnosis can be traced back through patient number and consultation number
5% overall case sample, 10% video-specific sample, covering different sections/pathology/source/time period
Remove name, ID number, telephone number, address, clinic number, doctor ' s name, uncomposed hospital name
DICOM/WSI metadata, private labels, burning text in images, synchronizing tag maps
Maintain AI training information necessary for scale, multiplier, spatial positioning, sequence recognition
Data are prohibited from leaving the country in compliance with the Personal Information Protection Act, the Data Security Act
This data set requires that each case be of three years duration and that it contain a record of visits at least two or more different points of time, and that the information on diagnosis, testing, examination, medication, paperwork, images and end-effects within the same patient be restored in chronological order to the main clinical trajectory. The normal electronic medical history data set is usually a single-diagnostic snapshot, lacking vertical time-series linkages and not being able to support training in time-series reasoning models.
Supports seven types of medical image modelling for CT, MRI, PET-CT, ultrasound, endoscopy, X-ray/DR, pathological whole-slice digital images (WSI). DICOM-type images retain the necessary metadata for training in _KEEP_core_labels and spatial positioning (ImagePositint/ImageOrganizationPatient), sequence parameters (story thickness/pixel spacing/scan parameters).
The full amount of data is automated and manually de-sensitized, removing identifiable information such as the patient’s name, ID number, telephone number, etc., while retaining original values of age to guarantee analytical value. DIOCOM and WSI data are processed in a synchronized manner, private labels, and image burning. Data are prohibited from leaving the country in compliance with the requirements of the Personal Information Protection Act, the Data Security Act.
The project does not require the marking of results such as a frame, profile, manual classification label, etc., unless the parties agree otherwise in writing. The focus of video data delivery is on originality and internal relevance - image documentation, image reports, examination records, patient numbers, and patient numbers, and should correspond to each other and be included in the long-term treatment chain of the same patient.
The core scenario includes: 1 training in the time sequence of a large medical model (learning the diagnostic treatment review of the causal chain of the endings and consequences); 2 basic training in multimodular integration models (image+ clerical+ testing+ pathology combination); 3 modelling of vertical efficacy predictions and disease progression; 4 drug response predictions; 5-pharmaceutical AI benchmarking. ICD-10-CN coding maps and standardized data dictionarys of the data set support direct access to the mainstream training framework.
Please contact the data expert team at Chang Longway Information Technology. 137-5502-0164, or visits www.langhuiai.com/Contact The programme is flexible and tailored by section, disease, pattern, time span.
5000+ longitudinal total simulator data 8 Big Section 7 Type image simulation ICD-10-CN code 10 Compliance commercial
• Medical AI Data Hole Service
Compliant Data · Customized Delivery · Professional Technical Support