Creator: Changsha Langhui Information Technology Co., Ltd.
Keywords: ,3D,,,ACL,,,NIfTI

Dataset Overview

Dataset Name	1 million-grade bone image label dataset
Total Data Volume	1 Million+ 3D image marked slices
Core Anatomical Structures	Femur, Tibia, front cross-fault (ACL)
Annotation Task	3D split labels, footprint labels, bone track planning point labels, bone-model distance calculations
Data Format	NiffTI body data (.ni/.ni.gz), with original images and partition masks
Annotation Platform	3D Slicer 5.8, extended to ModelToModelDistance
	3D re-establishment visualization, footprint shape consistency check, extraction of 0-0.5 mm threshold from bone distance

Anatomy structure and labelling tasks

Core Anatomical Structures

FemurTibiaFront cross-flex (ACL)

3D Split Tasks

The femur is divided into three dimensions.The gillbone is divided into three dimensions.ACL 3D Division

Footprint Area Marker

bobble footprint (ellip)Skull footprint (weeds)Bones-model distance calculation

Bonest planning

The entrance to the femur.The femur exit.The entrance point to the cheekbones.The exit point for the cheekbones.

Imaging Data Specifications

Specification Item	Detailed Description
Data Format	NiffTI body data (.ni/.ni.gz), with original images and partition masks
Annotation Format	3D split mask (security mark)+ keypoint label (Markups Point List)
Annotation Platform	3D Slicer 5.8, extended to ModelToModelDistance
Footprint Region Authentication	Bones-model distance calculation, 0-0.5 mm threshold extraction footprint model (STL export)
Bone-tracking approach	The entry point for the femur is based on IDEAL anatomy, with the rest being marked by clinical actuality.
	3D Reconstruct visualization, recyclable rectangular consistency check (guilium osteoporus/steoporus wedge)

Data Security and De-identification

Processing Type	Detailed Description
De-identified Fields	All personal identification information, including name, ID number, medical record number, medical institution identification
Reserved Fields	NiffTI image data, partition mask, footprint model STL, coordinates of key points for bone-channel planning

Field Data Dimensions

Field Name	Data Type	Dimension Description
image_id	String(32)	Image Slice Exclusive Identification
patient_anon_id	String(24)	De-sensitive patient unique identifiers supporting multiple video association of the same patient
study_uid	String(64)	DICOM Study Instance UID
modality	Enum	XRAY \| CT \| MRI \| CTA \| MRA \| DEXA
anatomy_region	Enum	KNEE \|SPUMBAR \| SHOULDER \| HIP \| ANKLE \| WRIST \| ELBOW \| PELVIS
view_position	String(20)	AP / PA / LAT / SAGITTAL / CORONAL / AXIAL
pathology_label	Array[Enum]	Bone fracture / osteoporosis / intervertebrate retreat / constriction / osteoporosis / tumor / infection / post-operative modification
fracture_type	Enum	NO_FRACTURE \| AO_TYPE_A \| AO_TYPE_B \| AO_TYPE_C
disc_degeneration	Enum	PFIRRMANN_I \| II \| III \| IV \| V
spinal_stenosis	Enum	NONE \| MILD \| MODERATE \| SEVERE
implant_present	Boolean	Existence of internal fixtures (steel plate/snail/manual joint)
postop_status	Enum	NO \| POST_ACUTE \| POST_SUBACUTE \| POST_CHRONIC \| REVISION
age_group	Enum	0-18 \| 18-40 \| 40-60 \| 60-75 \| 75+
gender	Enum	MALE \| FEMALE
quality_score	Enum	S ⁇ A ⁇ B ⁇ C (image mass slotting)

De-identified JSON Data Sample

Other Organiser
"image id": "ORT-IMG-20240912-06741",
"Patient anon id": "POR-0084721",
"Study uid": "1.2.840.113619.2.xx.xx."
"modality": "MRI,"
"anatomy region": "SPINE LUMBAR",
"view position": "SAGITAL T2",
"Pathology label":
"fracture type": "NO FRACTURE",
"Facture migration":
"Osteophyte gade": "Moderade",
"disc degeneration": "PFIRRMANN IV",
"spinal stenosis": "MODERATE",
"bone denity tscore":
"implant president."
"postop status": "NO",
"age group": "40-60,"
"gender": "FEMALE",
"quality score": "A",
"Report summary": "L4-L5 intervertebral disc Pfirrmann IV retreat, medium narrow vertebrae. L5-S1 intervertebrate disc slightly retreated. No acute fracture."
♪ I'm sorry ♪

Data Distribution Statistics

28%

Knees Image

22%

Spinal Image

18%

Shoulder joint image

32%

Other

Dimension	Distribution Description
Image Model Distribution	X-ray 58% MRI 28% CT 10% DEXA 3% CTA/MRA 1%
Age Distribution	0-18 years 8% ⁇ 18-40 years 22% ⁇ 40-60 years 42% ⁇ 60-75 years 23% ⁇ 75 years and over 5%
Pathology distribution	Osteoporosis 35% thirties 28% thiram vertebrate retreat 22% thiram osteoporosis 8% thiram 4% tumours 3% thiram thiram
Quality Grading	S, 18% A, 42% B, 30% C, 10%
Post-operative image ratio	After the internal surgery 12% after the joint replacement 8% after the renovation 2% after the total 22%

AI Training Use Cases

Bone detection and classification

Applying model: YOLOv8-Ortho / Fast R-CNN. AO-specified automatic determination + multi-perspective integration to support rapid screening for emergency fractures.

Osteoporosis grade

Applicable model: ResNet-50 / EfficientNet. Kellgren-Lawrence Level + Fracture + Burscular Coat Test.

Cylinder and spinal analysis

Applying model: 3D CNN / Swin Transformer. Pfirrmann grade + narrow vertebrae measurement + Cobb angle to the side of the spine automatically calculates.

Osteoporosis screening

Applying model: DEXA T-value regression/X-line texture analysis. Bone risk prediction + FRAX model integration.

Post-operative assessment and prediction

Applying model: Internal fixed position assessment / Prosthetic loose detection / Renovation risk projection.

Multimodular Osteotypical Large Model

Applying model: Image-Report Multimodular / Osteo-Astetric Knowledge Questions / Surgery Programs Recommendations.

Data Acquisition and Quality Control System

DICOM Original Collection

Collecting raw DICOM data from the Cooperative Image Centre and removing PHI from the automatic de-identification stream

Image Mass Control Filter

Automatically detect mass issues such as motion hypothymics, inappropriate position, under-exposure, and remove unsatisfactory images

Expert marking and auditing

Pathological indications by bone-image specialists, double-checking + arbitration mechanisms to ensure accuracy of marking

Stored by Quality Grade

S/A/B/C-4 based on image quality, marking integrity, pathological recognition

Frequently asked questions

Does the data cover the child ' s bone?

Including. The under-bone population aged 0-18 years, which covers 8 per cent of the total, covers childhood-specific diseases such as fractures of the bone (Salter-Harris spectrotype), developmental hips (DDH).

How does post-operative images fit?

Post-operative images are singled out for "postop_status" and distinguish between post-operative acute periods (PACs)<3）、（3-12）（>12 months).

How does DEXA data relate to images?

The same patient ' s DEXA examination was linked to X-ray/MRI through the Patient_anon_id to support a joint bone density-image expression modelling.

Does support 3D rehabilitation training?

Support. Part of the CT data provides 3D reconstruction labels (STL format) and supports 3D training in depth learning models, such as 3D CNN and PointNet++.

Is training/certification/testing divided?

Provide standard classification (7:1.5:1.5) to avoid data leakage by level of patient, and provide a cross-centre panoramic test subset.

A customized bone image data program?

Changsha Langhui Information Technology Co., Ltd. DataAssetsAPI platform, which aims to provide high-quality, compliant data assets for AI enterprises and scientific institutions. Supports flexible combinations by anatomical structure, job description, quality level.

Data Sample Preview · Customized Data Solutions · Professional Technical Support

Data Sample Preview Customized Discipline-Specific Solutions Professional Technical Support

Million-Scale Orthopedic Imaging Dataset