Dataset Overview
| Dataset Name | Mid-medicine Visual Intelligence Diagnostic Data Set (TCM-VISION) |
| Total Data Volume | 1 million + original images (with 50,000 + labels) |
| Data Format | JPEG / PNG Image + JSON / CSV / COCO Embedding + PNG Momented |
| Annotation Task | Face skin detection / tongue diagnosis / tongue bottom diagnosis / facial diagnosis / quality assessment |
| Breakdown category | 100+ China Medical Diagnosis Subcategory (18 T-tempo 8 / T-tempo 4 / Quality 8) |
| Data Source | Clinical collection (multicenter, multi-equipment, ethical clearance) at collaborating hospitals and Chinese medical clinics |
| Annotation Method | Independent representation of 2 Chinese attending physicians + 1 Director-General Medical Doctor |
| Privacy protection | Fuzzy facial characteristics, de-identification of patients, compliance with the Personal Information Protection Act and the HIPAA |
Detailed Description of Data Dimensions
| Field Name | Data Type | Value Range / Description |
|---|---|---|
| image_id | String(32) | Image unique identification, format: TCM{center}{seq} |
| sample_type | Enum | TONGUE_TOP (Tangue) ⁇ TONGUE_BOTTOM (Tang) ⁇ FACE (Face) ⁇ SKIN (Skin) |
| task_type | Enum | SKIN_DETECT (Face skin detection) ⁇ TONGUE_TOP (Tangure diagnosis) ⁇ TONGUE_BOTTOM (Top-diagnosis) ⁇ FACE_DiaG (Face-diagnosis) ⁇ QUALITY (Quality assessment) |
| label_category | Enum | Based on the tab_type dynamic value (18 tongues / bottom 8 / face 4 / mass 8) |
| label_value | String | Specific diagnostic labels, such as "Little Red Mouth", "Little White Moth", "Blue White Tongue" and "Qualified" |
| bbox_coords | Array[4] | Target detection frame coordinates: [x, y, width, head] (pixel coordinates) |
| mask_path | String | Semantic Split PNG Monument File Relative Path |
| annotator_id | String(16) | The only ID that points to is de-identification. |
| review_status | Enum | PENDING | REVIEWED | DISPUTED |
| quality_passed | Boolean | True ⁇ false (pass or pass) |
| quality_reason | String | Reasons for non-adoption (unscient/colour bias/exposure/scrution/cut/out-of-focus/reflect/other) |
| created_at | DateTime | ISO 8601 format, mark creation time |
De-identified JSON Data Sample
Annotation Specifications and Quality Control
Annotation Standards
- Following the standards of Chinese diagnostic teaching materials (V edition of Chinese Medicine Diagnostics, New Century) for the classification of the pictogram/face
- Telecommunications image pixel-by-pixel drawings of the pixels along the edge of the tirex, and separate tirade/trect
- Face skin detection is marked with a rectangle frame in COCO format, containing a bbox + class label
- Embedding tool: Self-study TCM-Annotator label platform, uniform output in JSON/CSV/COCO format
Quality Control Process (Three-Tier Inspection)
- Level I self-examination:The medical doctor is marked to perform a self-check of all samples, correct obvious errors and omissions
- Level 2 cross-checking:Another physician of the same grade cross-checked and Kappa & lt; 0.80 batch returns the re-mark
- Third-level expert check-ups:The chief physician randomly took 20% of the sample final, overall consistent rate ⁇ 0.92
- Label Consistency:Categorized Chinese Medicines Diagnostic Classification Kappa coefficient 0.85
- Telecommunications collection: natural light/standard light source, patient ' s tongue is relaxed, tongue is fully exposed, resolution ⁇ 1920 x 1080
- Face collection: face natural light, face relaxing, no makeup/coverage, background pure colour, resolution ⁇ 1920 x 1080
- Skin collection: micro-range/skin mirror pattern, disease-cantoned area is clear and light is not shadowed
- Quality screening: automatic detection of vague, colour-based, perceiving, shielding, tailoring, defolio, photometric 7
Quality Metrics
99.5%
Data integrity rate
0.85
Kappa coefficient
<2%
Denunciation rate
0.92
Convergence rate
AI Training Use Cases
Smart tongue diagnostics
Applying model: EfficientNet / ConvNeXt / Swin Transformer. Based on the tongue/tamp bottom partition mask and 18 Tampa-tampa-tampa-tastic classification labels, training-to-end MtT-AI model to achieve automatic recognition of the tongue, moss, tongue, and lower tongue arteries.
Face skin disease screening
Applying model: YOLOv8 / DETR / RT-DECR. Based on skin detection signs for 100+ subclass, training in high-precision skin detection models supports automatic identification and classification of multiple skin problems such as scabies, colours, wrinkles, etc.
Face analysis (seek color)
Applying model: CLIP / DINOV2 / VIT. Using the overall facial label (Class 4 for yellow and pale), training in visual profiling models for medical consultations in Chinese to achieve physical recognition and health status assessment based on facial images.
Automass control of image quality
Applying model: MobileNetV3/EfficientNet-B0. Based on quality labelling for eight unsatisfactory reasons, training in light mass image quality screening models, embedded at the collection end to achieve real-time quality feedback and reduce waste film rates.
Multimodular Chinese diagnostic integration
Applying model: Multimedia Transformer / ALBEF / ImageBind. Combining the multimodular characteristics of the hypothetical, facial, skin images, building a comprehensive diagnostic model for Chinese medicine, achieving the full-process alization of "visiting" and supporting Chinese physicians to improve their accuracy in identification.
Chinese medicine knowledge mapping
Applying model: GraphSAGE / RGCN / TransE. Based on the labeled classification system (18 + 8 + 4 + 4 ) a map of the knowledge of the Chinese medical clinic supports the application of intelligent consultations, forensic reasoning, etc.
Get This Dataset
The Chinese Doctors Visual Intelligence Diagnostic Data Set provides business authorization, containing complete documentation and de-identification images.
137-5502-0164 Online ConsultationQuick Navigation
Related Datasets
Commercial Application Value
China Medical Service Intelligence System
Training in the AIS diagnostic engine for medical and medical services in China, which enables primary health-care institutions to achieve standardized visits and reduce reliance on senior medical practitioners, is on a market of over $3 billion.
Internet Medical Platform
Embedding an online consultation app/show, providing ASP and VAT, increasing online access efficiency by over 50%, and reducing the risk of error and omission.
Science and teaching in Chinese medicine
Standardized labelling data are provided to secondary medical institutions and scientific institutions to support the objectiveisation of Chinese medical visits, digital research and the compilation of Chinese medical AI materials.