Dataset Overview
| Attribute | Description |
|---|---|
| Dataset Name | Medical Image Training Data Set |
| Total Scale | 20 million+ medical image training sample |
| CT MRI X-ray DR . Ultrasound . endoscopy PT-CT SPEC . Pathological slice . DSA OCT | |
| Task Types | Image classification: Target detection; Semantic division; Examples split; Keypoint detection; Image report generation; Image alignment; Denomination. |
| Annotation Standards | Double-person independent label + Director Medical Review + Pathological fund standard certification (when pathological) |
| Source | Self-study/collection (de-identification image data, authorized to operate at MNA Hospital) |
| Use Cases | Model training Model fine-tuning Model assessment |
| Supply Format | Hard Drive · Cloud Drive · API · Data Infrastructure |
Dataset fields
| Category | Field | Description |
|---|---|---|
| Image Identification | Image ID.StudyUID.SeriesUID.SOPInstanceUID | DICOM Standard Single Identification System |
| Model Properties | Model type . Anatomy part . Image size . Pixels spacing . Layer thickness | Image Physics Properties and Scan Parameters |
| Type of notation · Box/scene • Tag • Marker ID = Audit status | Medically labeled results and qualitative information | |
| ClinicalReference | The pathology gold standard clinical diagnosis follow-up results ICD code | Objective diagnostic reference criteria |
| Data Source | Collection agency • Equipment model • Time of collection • Data authorization modalities | Data traceability and compliance information |
Image Modour and Mark Task Distribution
| Imaging Modality | Typical anatomy. | Annotation Task | Sample size |
|---|---|---|---|
| CT | The chest, the abdominal, the skull, the spinal cord. | Pulmonary hysterectomy detection/division, liver tumour partition, intracircle haemorrhage classification | 800+ |
| MRI | The skull, the knee joint, the spinal cord, the prostate. | Brain tumor partition, ACL torn detection, intervertebrae highlighting classification | 5 Million+ |
| X-ray (DR) | The chest, the bones, the abdomen. | Pneumonia detection/classification, fracture detection, aerobic chest recognition | 400+ |
| Ultrasound | The abdominal, the thyroid, the breast, the heart. | Thyroid T-RADS classification, breast BI-RADS classification, heart function assessment | 120+ |
| Indigestion. Respiration. | Spectrum testing, early cancer screening, disease classification | 80+ | |
| Pathological slices | Breasts, lungs, gastrointestinals, cervicals. | Cell nucleus separation, tumour area classification, immunisation rating | 1M+ |
De-identified JSON Sample
Other Organiser "image id": "IMG CT CHEST 2024Q3 0582134", "modality": "CT", "body part": "chest," "image size": [512, 512, 321], "voxel spacing": [625, 0.625, 1.0], "Annotations": "type": "bbox", "label": "solid nodule", "coordinates": "x": 234, "y":187, "z": 145, "w":18, "h":18, "d":3" , "confidence":0.96, "type": "bbox", "label": "ground glass opacity", "coordinates": "x": 410, "y": 298, "z":152, "w":24, "h":32, "d":5" , "confidence":0.9 I don't know. "clinical diagnosis": "Lung-RADS 4A on the right lung, recommended three months of follow-up", "Annotator id": "ANNO RAD 0113", "review status": "approved," "reviewer id": "REVIEW RAD CHIEF 04" ♪ I'm sorry ♪
Other Organiser
"image id": "IMG MRI BRAIN 2024Q3 0128976",
"modality": "MRI,"
"body part": "head,"
"sequences": ["T1", "T1ce", "T2", "FLAIR"]
"image size": [240, 240, 155],
"Annotations":
"type": "security mask",
"labels":
"enhancing tumor":
"peritumoral edema":
"necrotic core":
♪ I'm sorry ♪
{\cHFFFFFF}{\cH00FFFF}
"Pathology report": "Glastic mother-cell tumor, WHO class IV, IDH wild, MGMT-negative,"
"good standard source": "post-operative pathology,"
"review status": "approved"
♪ I'm sorry ♪Other Organiser
"image id": "IMG Xray CHEST 2024Q3 0321456,
"modality": "X RAY,"
"body part": "chest,"
"view position": "PA",
"image size": [2160, 2560],
"Annotations":
"type": "multiilabel classification",
"labels":
"Finding": "cardiomegaly", "president": false,
"Finding": "prevent": True, "severity": "moderate"
"Finding": "pneumonia", "prevent":false,
"Finding": "pneumothorax", "prevent":false
{\cHFFFFFF}{\cH00FFFF}
"Report summary": "The right-hand medium chest cavity, recommending ultrasound piercing flow,"
"Annotator id": "ANNO RAD 0207",
"kappa score": 0.92
♪ I'm sorry ♪Other Organiser "image id": "IMG US THYROID 2024Q3 0087142", "modality": "interrasund", "body part": "theroid," "image size": [512, 512], "Annotations": "type": "contour", "label": "thyroid nodule", "points": [187,132], [195,128]... [182,136], "area px":324, "ype": "classiation", "label": "TI-RADS 4b", "criteria": ["hypoechoic": True" ], "irregular margin": True" I don't know. "Pathology report": "The thyroid breast cancer" (FNA diagnosis, Brasilian Bethesda VI), "gold standard source": "Putting work." "review status": "approved" ♪ I'm sorry ♪
Other Organiser
"image id": "IMG PATH BREAST 2024Q3 0015248",
"modality": "pathology,"
"body part": "breast,"
"Stain": "H E",
"magnification": "40x,"
"image size": [2048, 2048],
"Annotations":
"type": "instance security",
"nuclei": Chuckles
"instance id": 1, "label": "epithal", "mask rle": "...", "area": 452, "circurality": 0.87).
"instance id" :2, "label" : "inflammatory", "mask rle" : "...", "area" : 321, "circulity" : 0.9.
I don't know.
"Total nuclei": 1247,
"tumor stroma ratio": 0.38
{\cHFFFFFF}{\cH00FFFF}
"Diagnosis": "Insurgent catheter cancer, tissue grade II."
"gold standard source": "two-person independent label + pathology expert audit,"
"review status": "approved"
♪ I'm sorry ♪Annotation Specifications
Independent Annotation by Two Annotators
Each case is marked by two video doctors, and the results are stored independently.
PathologyGold Standard
Post-operative pathology/prospect cases are treated as pathologically sound to ensure objective and reliable marking
The chief physician's clearance.
The sample of the dispute is finally decided by the chief physician and the quality of the mark is reviewed periodically by a sample sample of the physician
D.I.COM dissembling.
Complete separation of the patient ' s identity information from the DICOM header, with only video data and clinical diagnosis maintained
AI Application Scenarios
The disease is smart.
Automatic detection and positioning of common cookstoves such as pulmonary dysentery, liver tumours, and skull haemorrhage, to improve the effectiveness of video readings
Organ/tumour division
The organ contours of multimodular images are precisely separated from the tumor boundary, supporting automatic mapping of the therapeutic target area
Automatic Imaging Report Generation
Automatic structured reports based on image characteristics are generated, reducing the paperwork of video physicians
I'm not sure.
Standardized intelligence assessment of Lung-RaddS, thyroid TTI-RADS, mammary BI-RADS, etc.
Multimodular Image Integration
Cross-modular image alignment and integration training CT+PET-CT, MRI+T1+T2+FLAIR
Assessment of efficacy and follow-up
Comparison of pre- and post-treatment images, vertical follow-up models such as REDD assessment, automatic quantification of tumor retreats, etc.
Commercial Value
Reduce AI R & D costs
The model label data set replaces the one-million-scale image label outsourcing, reducing the medical image AI development cycle by 50 per cent +.
Standard security for pathogen
The pathological standard for confirmed cases is marked by a pathological standard, and model training has been significantly better than video-only indications.
Compliance.
DICOM complete de-identification + authorized operation + data security law compliance, free of data compliance concerns.
Insert
Provides a double format for DICOM+NIfTI, indicating that multiple labeling formats such as COCO/YOLO/ITK-SNAP are exported.
Frequently asked questions
Did you do the de-identification of the video data?
Yes. All images are fully de-identified by DICOM, stripping out the patient ' s name, ID, date of birth, number of examination, and only keep anonymous signs, image pixel data and clinical diagnosis.
How is the pathological merit standard confirmed?
Cases with surgical ectoplasmization or perforation are considered as a gold criterion for post-operative pathology reports; cases without pathology are independently marked and cross-checked by two video doctors.
Which label formats to export?
The video file is in a double format of DCOM/NIFTI, marked as supporting the generic formats of COCO JSON, YOLO txt, ITK-SNAP seg.nrrd, PASCAL VOC, which provide seamless access to the mainstream training framework.
Get a complete catalogue of medical images
Contact data experts to get detailed video lists and commercial offers.