Dataset Overview
Definition：Million-Scale Upper Abdominal CT Imaging Datasetis provided byChangsha Langhui Information Technology Co., Ltd.A database of mega-medicine images, based on the DataAssetsAPI platform. It covers liver, gall bladder, pancreas, spleen, adrenal gland, kidney and lymphal lymphal ligature seven large organs, 1 million + DICOM image slices, severely de-identified and divided through the 5-dimensional Quality Rating System (S+/S/A/B/C).
| Dataset Name | Millions of upper-breathed CT image label data sets |
| Total Data Volume | 1 million + DICOM image slices (including sweep + enhanced sequences) |
| Data Format | Original DICOM format, keep full scanning parameter metadata |
| Covered Organs | 7 major organs: liver, gall bladder, pancreas, spleen, adrenal gland, kidney, abdominal lymphoma |
| Matrix Resolution | ≥ 512 × 512, with some sequences at 1024 × 1024 |
| Scan Type | Top abdomen flattening CT, boosting CT with patients as a reference |
| Layer thickness distribution | 1.0 mm - 1.5 mm thin layer CT ratio of approximately 80% |
| Device Distribution | GE 256-slice CT accounts for about 80%, with equipment from other manufacturers serving as a diversity supplement |
| Control current range | 200 – 400 MA (> 800 MA anomaly data excluded) |
| Path phase | Mainly early- and mid-stage disease courses, excluding post-treatment, post-operative and late-stage cancer images |
| Enhance CT interval | The same patient has to swipe and enhance the sequence. < 7 |
| Quality Requirements | Free of severe motion artifacts and metal artifacts |
Organ and disease coverage
The data set covers seven major abdominal organs, each of which contains a finely marked description of the common pathogen type.
Liver
Gall bladder
Pancreas
Spleen.
Adrenal gland.
Kidney.
Imaging Data Specifications
| Specification Item | Detailed Description |
|---|---|
| Data Format | Original DICOM format, complete metadata (scan parameters, layers, re-nuclear, etc.) |
| Matrix Resolution | ≥ 512 × 512, with some sequences at 1024 × 1024 |
| Scan Type | Top abdomen flattening CT, boosting CT with patients as a reference point. |
| Layer thickness distribution | 1.0 mm - 1.5 mm thin layer CT ratio of approximately 80%, with the remaining layer thick as a complement to diversity |
| Device Distribution | GE 256-slice CT accounts for about 80%, with equipment from other manufacturers serving as a diversity supplement |
| Control current range | General upper abdomen scan 200-400 ma, excluding data above 800 ma |
| Path phase | Mainly early- and mid-stage disease courses, excluding post-treatment, post-operative and late-stage cancer images |
| Enhance CT interval | The same patient has less than a week between the clean-up and the enhanced sequence. |
| Quality Requirements | Free of severe motion artifacts and metal artifacts |
Field Data Dimensions
| Field Name | Data Type | Dimension Description |
|---|---|---|
| patient_anon_id | String(24) | Dissensitive patient unique identification, linked to multiple patient check-ups |
| study_uid | String(64) | DITC STUDY Information UID, Identification Single CT Check |
| series_uid | String(64) | DICOM Series Information UID, identifying the same scan sequence |
| slice_index | Integer | Image Serial Number, single check can contain 200-600+ slices |
| scan_type | Enum | PLAIN | ENHANCED_ARTERIAL | ENHANCED_VENOUS | ENHANCED_DELAYED |
| modality | Enum | CT | CECT | CT_SPIRAL |
| slice_thickness | Float | Layer thickness 0.625-5.0 mm, main force data set 1.0-1.5 mm, or approximately 80% |
| matrix_size | String | Rebuild matrix sizes: "512 x 512" or "1024 x 1024." |
| kvp | Integer | Pyramid voltage 80-140 kVp |
| tube_current_ma | Integer | Controlled current 200-400 MA (excluded > 800 MA anomaly data) |
| pixel_spacing | String | Pixels spacing: 0.5\0.5 mm to 0.9\0.9 mm |
| organ_labeled | Array[Enum] | List of organs marked ["Liver", "Galblebradder", "PANCREAS", "SPLEEN", "ADRENAL", "KIDNY", "LYMPH_NODE"] |
| lesion_present | Boolean | Existence of a stove (based on radiology reports) |
| lesion_type | Enum | MALIGNANT | BENIGN | INFLAMMATORY | CYSTIC | NORMAL |
| lesion_diameter_mm | Float | Maximum path (mm) of the stove, from the image section measurements or label team labels |
| lesion_location | String | Diagram description (e.g. LIVER_SEGMENT_VIII) |
| pathology_confirmed | Boolean | Is there a pathological report confirming the diagnosis? |
| age_group | Enum | 20-30 | 30-40 | 40-50 | 50-60 | 60-70 | 70+ |
| gender | Enum | MALE | FEMALE |
| quality_score | Enum | S_PLUS ⁇ S ⁇ A ⁇ B ⁇ C (Five-D Mass Slotting) |
| motion_artifact | Enum | NONE ⁇ MILD ⁇ MODERATE (Several motion hypocrisy excluded) |
| report_summary | Text | Summary of the Image Section report, description of the disease and diagnosis |
Embedding and reporting information
Image Section report
The section of the stove, morphological characteristics, pathology type, stage and severity assessment
Pathology Report
Source of the screening organization, nature of the disease, degree of differentiation, criteria for case confirmation
Diversity coverage
Each disease category covers different ages, genders and ensures modelability
Data Security and De-identification
All data has been thoroughly de-identified, with all information that could directly or indirectly identify an individual removed; only parameters and labels relevant to medical research are retained。
| Processing Type | Detailed Description |
|---|---|
| De-identified Fields | All personal identifiers such as name, ID card number, phone number, address and hospitalization number |
| Reserved Fields | DICOM scan parameters, image pixel data, disease stove signs, pathological diagnostic information |
De-identified JSON Data Sample
The following is a structured JSON model of de-identified processing.
Other Organiser "Patient anon id": "PAB-CT-20241003-01582", "Study uid": "1.2.840.113619.2.xx.xx.xx.xx.xx.xx.xx.xx." "series uid": "1.2.840.113619.2.xx.xx.xx.xx.xx.yyy." "slace index": 245, "scan type": "ENHANCED ARTERIAL", "Modality": "CECT", "slice thickness": 1.25, "matrix size": "512x512" "kvp": 120, "tube current ma": 350, "Pixel spacing": "0.7" "organ labelled": ["Liver", "PANCREAS", "KIDNY", "Lesion present": true, "Lesion type": "malignant," "Lesion diameter mm": 32.5, "Lesion location": "Liver SEGMENT VIII", "Pathology confront": true, "age group": "50-60", "gender": "MALE", "quality score": "S", "motion article": "NONE", "Report summary": "The eighth part of the right-hand liver shows a low density of about 32.5 x 28.0 mm, with a significantly enhanced scanned artery period, and a clear threshold period and delay period, considering the possibility of primary liver cell cancer. The pancreas and the kidneys are not clearly abnormal." ♪ I'm sorry ♪
AI Training Use Cases
Disease detection and partitioning
Applying model: NU-Net / YOLOv8-3D / Mask R-CNN / SAM-Med. Based on the thin-layer CT-to-end multi-organosis and partitioning model, support is provided for simultaneous analysis of seven large organs.
Good and evil classification
Applying model: ResNet-3D / Swin Transformer / DenseNet. Using enhanced CT multiphase phase characteristics, training in autometric identification models for vascular tumours vs. malignant tumours, to assist in the LI-RADS standardized classification.
Joint multi-organ analysis
Applying model: NNU-Net Multi-Category / TransNet / TotalSegmentator. Multi-organ stoves, such as liver, pancreas and kidneys, are separated and classified simultaneously in the same CT scan, with an upper abdominal panorama diagnosis.
Radiomics and Prognosis Prediction
Applying models: PyRadiomics + XGBoost / Random Forest. Extract 1000+D image grouping features from the partition mask, build models such as tumour prognosis, lymphoma transfer risk assessment.
Automatic Imaging Report Generation
Applying model: Multimodular LLM / Vision-Language Model. Video-text pairing training automated report generation system, output structured image reporting, more than 40% efficiency in radiology.
lymph knot transfer prediction
Applying model: GNN chart neural network / 3D CNN. Cervical lymphoma lymphoma ligation detection and transfer risk rating, supporting tumour ration and treatment programme decision-making.
Data Acquisition and Quality Control System
Raw data collection
Collecting raw DICOM data from the Cooperative Image Centre and removing PHI from the water stream by de-sensitive water lines
Image Mass Control Filter
Automatically detect mass problems such as motion hypothyms, metal permutations, incomplete scanning ranges, and remove unsatisfactory images
The medical kit is marked and reviewed.
RIO by medical video expert, double review + arbitration mechanism to ensure accuracy
Stored by Quality Grade
S+/S/A/B/C-5 based on image quality, marking integrity, pathological recognition
Frequently asked questions
A customised upper abdominal CT data program?
Changsha Langhui Information Technology Co., Ltd. The DataAssetsAPI platform is dedicated to providing high-quality, compliant medical image data assets to AI enterprises and scientific institutions.
Data Sample Preview · Customized Data Solutions · Professional Technical Support