HCC MRI Image Marking Data Set - Langhui Technology Data Products

Dataset Overview

Dataset Name	HCC MRI Multi-Sequence Image Embedding Data Set
Total Data Volume	52,800 cases (with a notationed image 3.17 million frames)
Data Format	DICOM 3.0 / NifTI / JSON label
Series Type	T1WI / T2WI / DWI / ADC / DCE multi-stage enhancement
Data Source	Radiology unit of the cooperative medical facility (multicentres, multi-equipment, ethical clearance)
Annotation Method	Independent marking of 2 radiologists + 1 chief physician arbitration
Privacy protection	DICOM Head Information de-identification, facial characterization, HIPAA/GDPR
Update Frequency	Quarterly increments, annual average increase of 8,000 - 12,000

Detailed Description of Data Dimensions

Field Name	Data Type	Value Range / Description
patient_id	String(32)	IID, Format: P{center}{seq}
study_uid	String(64)	DICOM Story Information UID
age_group	Enum	18-30 / 31-45 / 46-60 / 61-75 / 76+
gender	Enum	M / F
sequence_type	Enum	T1WI \| T2WI \| DWI \| ADC \| DCE_ART \| DCE_PVP \| DCE_DELAY
slice_thickness	Float	1.0 - 8.0 mm
pixel_spacing	Array[2]	[x_spacing, y_spacing] mm
matrix_size	Array[2]	[256-1024, 256-1024]
lesion_label	Enum	HCC \| ICC \| MET \| HEM \| FNH \| CYST \| NORMAL
li_rads	Enum	LR-1 \| LR-2 \| LR-3 \| LR-4 \| LR-5 \| LR-M \| LR-TIV
lesion_size	Object	{long_axis_mm, short_axis_mm, volume_ml}
segmentation	Array	Polygon vertex coordinates (physical coordinates, mm)
seg_mask_path	String	Split the mask NiffTI file relative path
b_value	Integer	0 / 50 / 400 / 800 / 1000 (DWI Sequence)
annotator_consensus	Float	Dice coefficient among labels: 0.0 - 1.0
qc_status	Enum	PASS \| FLAGGED \| REJECTED

De-identified JSON Data Sample

[ { "patient_id": "P01A003824", "study_uid": "1.2.840.113619.x.masked.240518.001", "age_group": "46-60", "gender": "M", "sequence_type": "DCE_ART", "slice_thickness": 3.0, "pixel_spacing": [0.68, 0.68], "matrix_size": [512, 512], "lesion_label": "HCC", "li_rads": "LR-5", "lesion_size": {"long_axis_mm": 42.3, "short_axis_mm": 28.7, "volume_ml": 18.2}, "segmentation": [[125.3, 210.7], [130.1, 215.2], [128.9, 220.8], [120.5, 218.4], [118.2, 212.0]], "seg_mask_path": "masks/P01A003824_DCE_ART_HCC.nii.gz", "annotator_consensus": 0.92, "qc_status": "PASS" }, { "patient_id": "P02B015637", "study_uid": "1.2.840.113619.x.masked.240312.002", "age_group": "31-45", "gender": "F", "sequence_type": "T2WI", "slice_thickness": 5.0, "pixel_spacing": [0.94, 0.94], "matrix_size": [384, 384], "lesion_label": "HEM", "li_rads": "LR-1", "lesion_size": {"long_axis_mm": 12.1, "short_axis_mm": 8.5, "volume_ml": 0.46}, "segmentation": [[340.5, 155.2], [345.8, 158.0], [343.1, 162.7], [338.2, 160.1]], "seg_mask_path": "masks/P02B015637_T2WI_HEM.nii.gz", "annotator_consensus": 0.88, "qc_status": "PASS" }, { "patient_id": "P03C027419", "study_uid": "1.2.840.113619.x.masked.240115.003", "age_group": "61-75", "gender": "M", "sequence_type": "DWI", "slice_thickness": 5.0, "pixel_spacing": [1.17, 1.17], "matrix_size": [256, 256], "b_value": 800, "lesion_label": "ICC", "li_rads": "LR-M", "lesion_size": {"long_axis_mm": 55.8, "short_axis_mm": 41.2, "volume_ml": 49.6}, "segmentation": [[89.2, 78.4], [95.6, 82.1], [94.8, 88.7], [87.3, 91.5], [82.1, 85.2]], "seg_mask_path": "masks/P03C027419_DWI_ICC.nii.gz", "annotator_consensus": 0.95, "qc_status": "PASS" }, { "patient_id": "P01A003825", "study_uid": "1.2.840.113619.x.masked.240518.004", "age_group": "46-60", "gender": "M", "sequence_type": "DCE_PVP", "slice_thickness": 3.0, "pixel_spacing": [0.68, 0.68], "matrix_size": [512, 512], "lesion_label": "HCC", "li_rads": "LR-5", "lesion_size": {"long_axis_mm": 40.1, "short_axis_mm": 26.9, "volume_ml": 15.2}, "segmentation": [[122.0, 208.1], [127.1, 213.0], [126.3, 218.2], [118.0, 216.0], [115.5, 209.8]], "seg_mask_path": "masks/P01A003825_DCE_PVP_HCC.nii.gz", "annotator_consensus": 0.93, "qc_status": "PASS" }, { "patient_id": "P04D008992", "study_uid": "1.2.840.113619.x.masked.240707.005", "age_group": "76+", "gender": "F", "sequence_type": "DCE_DELAY", "slice_thickness": 4.0, "pixel_spacing": [0.82, 0.82], "matrix_size": [480, 480], "lesion_label": "NORMAL", "li_rads": "LR-1", "lesion_size": null, "segmentation": null, "seg_mask_path": null, "annotator_consensus": null, "qc_status": "PASS" } ]

Annotation Specifications and Quality Control

Annotation Standards

Classification of the disease by the diagnosis of the disease following the LI-RADS v2018 diagnostic criteria
Split the label using a 1mm layer-by-layer, interpolated values smooth
Multi-series alignment is based on the T1WI enhanced artery period, rigidity + imitation alignment, mi-like measures
Embedding tool: ITK-SNAP / 3D Slicer, uniform output in NNFTI format

Quality Control Process (Three-Tier Inspection)

Level I self-examination:After the doctor has done the mark, he checks all cases and corrects the obvious error.
Level 2 cross-checking:Another peer doctor cross-checked, Dice & lt; 0.80 returned the re-post.
Third-level expert check-ups:The chief doctor randomly extracts 15% of the total number of cases, overall consistent rate ⁇ 0.90.
Label Consistency:LI-RADS Classification Kappa coefficient 0.85

Quality Metrics

99.2%

Data integrity rate

0.92

Average Dice coefficient

0.87

LI-RADS Kappa

<1.5%

Denunciation rate

AI Training Use Cases

Hepatitis test and classification

Applying model: NU-Net / YOLOv8-3D / SwinUNETR / MONAI. A multi-sequence calibration image-to-end screening of cookstoves and LI-RADS automatic classification model, FROC & g; 0.85 clinical acceptable standards.

3D medical image split

Applying model: NU-Net / SegResNet / UNETR / MedSAM. Translating polygonal labels into a 3D bodyware mask to provide semantic splitting training in support of multiple categories (HCC/ICC/HEM, etc.).

Multi-series alignment and integration

Applying model: VoxelMorph / TransMorph / LapirN. Using the labels, training in a deformable network with multi-series data is provided, autospace aligns the different series MRI images.

Radiomics and Prognosis Prediction

Applying models: Radimitics (PyRadiomics)+XGBoost/ Random Forest. Extract 1000+D image group features from the partition mask, build models such as HCC Post-Reactivation Forecast, MVI Status Classification, etc.

Get This Dataset

HCC MRI image data sets provide business authorizations, including complete labelling documents and de-identified DICOM images.

137-5502-0164 Online Consultation

Related Datasets

Commercial Application Value

A. Aided diagnostic products

The training of the AIS-based liver (CADE/CADx) system, backed by standard data sets required for approval of the NMPA three types of certificate, is over $5 billion.

Image Cloud Platform

The smart structured reporting and quality control module embedded in the cloud PACS system enhances the Section ' s reporting efficiency by over 40% and reduces the rate of leakage.

Clinical trials of pharmaceutical plants

Provides independent image assessment (IRC) baseline data for clinical trials of medication/immunotherapy for liver cancer targets and accelerates the drug approval process.

HCC MRI Imaging Annotation Dataset