Dataset Overview

Dataset Name	Million-degree ultrasound image label dataset
Total Data Volume	1 Million+ Hypersonic Image Frame
Annotation Task	Image quality assessment (level 5), standard cut-off classification (category 11), organ partition (past), disease cooks partition (elastema)
	Liver; other abdominal organs can be added later
	JPEG / PNG, retaining acquisition metadata (device model, frame number, de-identified patient ID)
Annotation Format	JSON(labelme compatible) supports porygon / RLE / binary mark
	Hepatic_cyst, simple_cyst and complex_cyst
Point Precision	Error Between Segmentation Boundary and Actual Boundary < 5 pixel，

Four-large-pointed missions

Task I: Image quality assessment

Level 5 quality rating (very poor - excellent)Noise/false/out-of-focus assessmentSpectrum contact status judgementServer Control Feedback

Task two: Face classification

11 Class Standard Ctr.Level 4 rating for structural qualityCross-cutting/hort-cut by abdominal axle/swordFirst liver door/second liver door/ribs cut

Task three: Organ division

The liver is physically divided.Sector Area LimitMinimum scenario policy for the dark zonePolygon/RLE/Mask Multiformat

Task 4: Distinction of the stove

Hepatic cysts divided.Simple/ComplexAngiogenesis identificationFrame Sequence Consistency

11 standard cut-off tables

Decoration Name	Core autopsy mark
Axle arc of the abdominal aperture	Aperture axle, muscular foot, external side of the liver left (S2/3)
The sword slashes across the face.	The left of the door vein, the "worker" font, the liver radon.
Swords fall across the surface.	Lower vein liver long axis, tail leaf, vitro ligamental cord
Right side of the ribs.	Right-to-past liver substance, twirl vertex, liver and kidney interface
He's got a slash under his right-hand rib.	The right branch and branch of the door vein, the right half liver.
Right liver, maximum slope cut.	Hepatic right veins into IVC, right cheeks on the same screen as the lower liver.
Right side of the rib door.	Right branch of the door vein, right liver substance (rib windows)
Second liver cut the face.	The third liver veins are coming into the IVC, the Three Leafs.
Hepat-right kidney strutted face	Right back of the liver with right kidney cortex, Morrison's nest.
First liver door long axle cutting	The main veins are long, the "mice rat hunt."
First liver door cut across.	Door vein main stem crosses and fork shapes

Imaging Data Specifications

Specification Item	Detailed Description
Image Model	Campoon detects hepatic ultrasound, static images and video frames
	Liver; other abdominal organs can be added later
	JPEG / PNG, retaining acquisition metadata (device model, frame number, de-identified patient ID)
Annotation Format	JSON(labelme compatible) supports porygon / RLE / binary mark
	Hepatic_cyst, simple_cyst and complex_cyst
Point Precision	Error Between Segmentation Boundary and Actual Boundary < 5 pixel，

Data Security and De-identification

All data have been thoroughly de-identified, all information that directly or indirectly identifies individuals has been removed and only ultrasound image pixel data and labeled information are maintained.

Processing Type	Detailed Description
De-identified Fields	All personal identifiers such as name, ID card number, phone number, address and hospitalization number
Reserved Fields	Ultrasound image pixel data, quality rating, cut-face classification labels, organ/cook partition mask

Field Data Dimensions

Field Name	Data Type	Dimension Description
image_id	String(32)	Supersonic Images Monopoly
video_id	String(32)	de-identified ultrasound video identification, associated with frame sequences of the same video
patient_anon_id	String(24)	De-sensitive patients ' unique identifier
exam_type	Enum	Abdomen, abdomen, agry, agry, agglomerations, agglomerations, arrhythms, arrhythmia, arrhythmia, arrhythm, arrhythmia, arrhythm, arrhythmia, arrhythm, arrhythmia, arrhythm, arrhythmia, arrhythmia, arrhythmia, arrhythmia, arrhythmia, arrhythmia, arrhythmia, arrh, arrhythmia, arrh, arrh, arrh, arrhythmia, arrhythm, arrhythmia, arrhythm, arrance, arrhythm, arrhythm, arrity.
organ_system	String(30)	Hepatic pancreas, cystals, urology, hysteria, thyroid mammary, hyena, hysteria, blood vessels, hysteria, cytology, musculos, other
probe_type	Enum	CONVEX \| LINEAR \| PHASED_ARRAY \| ENDOCAVITY \| MICROCONVEX
probe_frequency	String(10)	Spectrum frequency "3.5 MHz" / "7.5 MHz" / "10 MHz" etc.
image_mode	Enum	B_MODE \| M_MODE \| COLOR_DOPPLER \| POWER_DOPPLER \| SPECTRAL_DOPPLER \| ELASTOGRAPHY \| CONTRAST_ENHANCED
finding_category	Array[Enum]	Normal, cystals, cysts, physicals, calcification, fluids, cylindricals, quartz, cystals, hysteria, vascular anomalies, cystals, growth abnormalities.
birads_tirads	Enum	NONE \| BI-RADS_0 ~ _6 \| TI-RADS_1 ~ _5
lesion_size	String(20)	3D measurement of the pathology "X-wide x mm" or "mm in diameter"
vascularity	Enum	NONE \| HYPOVASCULAR \| ISOVASCULAR \| HYPERVASCULAR
elastography_score	Enum	NONE \| SOFT \| INTERMEDIATE \| HARD
calcification	Enum	NONE \| MICROCALCIFICATION \| MACROCALCIFICATION \| BOTH
acoustic_shadowing	Boolean	Any rear sound?
border_definition	Enum	CIRCUMSCRIBED \| IRREGULAR \| SPICULATED \| INFILTRATIVE
echogenicity	Enum	ANECHOIC \| HYPOECHOIC \| ISOECHOIC \| HYPERECHOIC \| MIXED
age_group	Enum	0-18 \| 18-40 \| 40-60 \| 60-75 \| 75+
quality_score	Enum	S ⁇ A ⁇ B ⁇ C (ultrasound image mass slotting)

De-identified JSON Data Sample

The structured JSON model for de-identified treatment is presented below, real patient information has been removed and core medical data is maintained.

Other Organiser
"image id": "US-IG-20241108-00924,
"video id": "US-VID-20241108-0012",
"Pus-0076319,"
"exam type": "The subsurface organ ultrasound,"
"organ system": "the thyroid breast,"
"Probe type": "LINEAR",
"Probe frequency": "10 MHz",
"image mode": "COLOR DOPPLER",
"finding category": ["Performance", "Calcification"],
"Birads tirads": "TI-RADS 4A",
"Lesion size": "7.2x5.8x6.4 mm",
"vascularity": "HYPERVASCULAR",
"Elastography score": "Hald",
"Calcification": "MiccocalCIFCATION,"
"acoustic shading":
IRREGULAR,
"Echogenity": "HYPOECHOIC,"
"age group": "40-60,"
"quality score": "A",
"report summary": "Low echoes in the right-hand thyroid leaf, irregular boundaries, back-to-back growth, microcalcified points, CDFI blood flow, high elastic rating, T-RADS 4A class"
♪ I'm sorry ♪

Data Distribution Statistics

32%

Abdominal ultrasound.

28%

Light organ

22%

Maternity ultrasound

18%

Cardiovascular/skeletal/involvement

Dimension	Distribution Description
Image Model Distribution	Model B 58% ⁇ Colour Doppler 28% ⁇ Spectrum Doppler 8% ⁇ Resilient imaging 4% ⁇ M 2%
Age Distribution	0-18 years 5% ⁇ 18-40 years 25% ⁇ 40-60 years 45% ⁇ 60-75 years 20% ⁇ 75 years and over 5%
BI-RADS/TI-RADS distribution	2 and below 35% 3 32% 4 28% 5%
Size distribution of the stove	<1cm 22% \| 1-3cm 45% \| 3-5cm 22% \| >5cm 11%
Quality Grading	S 15% A 45% B 30% C 10%

AI Training Use Cases

Oversonics test.

Applying model: YOLOv8-Sono / Master R-CNN/ DETR. The Locator of Disease + Multi-Case Cook Test supports real-time ultrasound diagnostics.

T-RADS/BI-RADS classification

Applicable model: ResNet-50 / EfficientNet / Vit. ACR-TIRADS + K-TIRADS multi-characterization integration automatic rating.

Categorization of the disease ' s ill and bad

Applying model: Multi-characterization integration model. Border feature + microcalcification test + elasticity characterization learning, combining good and bad.

Hypervisit video analysis

Applying model: SlowFast / TimeSformer. Key frame detection + head position estimation + auto-scanning navigation.

Auto-generated ultrasound reports

Applying model: Multimodular LLM / CLIP-Sono. Image-text multimodular training, structured report generation.

UltrasoundLarge Models

Applying model: US-FM Ultrasound Basic Model / Self-supervised Pre-Training / Multi-mission Learning / RAG Knowledge Questions and Answers.

Data Acquisition and Quality Control System

DICOM collection and dissemination

Collect DICOM data from various types of ultrasound devices (GE/Philips/Siemens/Merry, etc.) and maintain a unified archive after automatic de-identification

Image Mass Control Filter

Assesses the quality of the image ' s audio window, the visibility of the anatomical structure, the quality of the probe ' s exposure and removes the unsatisfactory images

AI pre-reference + expert review

The pathology model pre-marks position and boundary frame, and ultrasound expert amends and points to the BI-RADS/TI-RADS classification

Stored by Quality Grade

Comprehensive image quality, marking integrity, pathological control recognition rating S/A/B/C, step IV

Frequently asked questions

Do the data cover different brands of ultrasound devices?

Coverage. Includes the GE Logiq/Vivid series, the Philips EPIQ/Affiniti series, the Siemens Sequoia/Acuson series, the Mary Resona/DC series, the metadata recording equipment brand and model.

How is consistency ensured?

The BI-RADS/TI-RADS label is performed by a ultrasound attending physician and the Kappa conformity test > 0.80 is qualified.

Does it contain video data?

Contains. About 35% of the checks include DICOM video series (line loop) to support time-series analysis and training in key frame detection models.

Is there a pathological confirmation?

About 22% of the malignant/suspicious stoves are recognized by the pathological gold standard (propellation/surgery) and labelled as "identified". The rest are clinical diagnosis confirmed by video follow-up visits.

Are the data supporting multi-modular studies?

Support. The same patient_anon_id can relate to model B, CDFI, elastic imaging and spectrum Doppler multiple-modular images to support multimodular characterization integration studies.

Need a customized data solution?

Changsha Langhui Information Technology Co., Ltd. The DataAssetsAPI platform is dedicated to providing high-quality, compliant medical data assets to AI enterprises and scientific institutions.

Data Sample Preview · Customized Data Solutions · Professional Technical Support

Data Sample Preview Customized Discipline-Specific Solutions Professional Technical Support

Million-Scale Ultrasound Imaging Dataset