Initial upload of African Medical Multimodal Fracture Dataset

README.md

@@ -35,7 +35,7 @@ dataset_info:
   - name: patient_id
     dtype: string
   - name: image_path
-    dtype: string
+    dtype: image
   - name: fracture_type
     dtype: string
   - name: age
@@ -56,6 +56,8 @@ dataset_info:
     dtype: int32
   - name: blood_pressure_systolic
     dtype: int32
+  - name: blood_pressure_diastolic
+    dtype: int32
   - name: temperature
     dtype: float32
   - name: height
@@ -97,13 +99,13 @@ configs:
 
 ## Dataset Description
 
-This dataset is a comprehensive, multimodal bone break classification dataset designed explicitly for African healthcare contexts. It addresses critical gaps in medical AI for resource-constrained environments while ensuring cultural sensitivity and local relevance.
+This dataset transforms the PKDarabi bone break classification dataset into a comprehensive, multimodal system specifically designed for African healthcare contexts. It addresses critical gaps in medical AI for resource-constrained environments while ensuring cultural sensitivity and local relevance.
 
 ### Dataset Summary
 
 - **Total Records**: 1,129 multimodal medical cases
-- **Original Images**: 1,129 X-ray images (88% of dataset)
-- **Augmented Images**: 135 mildly augmented images (12% of dataset)
+- **Original Images**: 1,128 X-ray images (89% of dataset)
+- **Augmented Images**: 135 mildly augmented images (11% of dataset)
 - **Fracture Types**: 10 different bone fracture classifications
 - **Countries Represented**: 18 African countries
 - **Languages**: 11 African languages + English
@@ -123,8 +125,8 @@ This dataset is a comprehensive, multimodal bone break classification dataset de
 - Traditional medicine integration patterns
 
 🏥 **Realistic Equipment Simulation**
-- 88% high-quality images (representing modern African hospitals)
-- 12% mildly augmented images (rural/portable equipment variations)
+- 89% high-quality images (representing modern African hospitals)
+- 11% mildly augmented images (rural/portable equipment variations)
 - Quality scores: 6.2-9.5/10 (realistic for African healthcare)
 
 📱 **Mobile Health Integration**
@@ -132,79 +134,55 @@ This dataset is a comprehensive, multimodal bone break classification dataset de
 - GPS tracking and movement data
 - SMS/voice follow-up preferences
 
-## Dataset Structure
-
-### Modalities
-
-Each record contains 6 data modalities:
+## Usage Examples
 
-1. **Image Data**
-   - Original X-ray images (JPEG format)
-   - Optional mild augmentation for rural equipment simulation
-   - Quality scores and equipment metadata
+### Loading the Dataset
 
-2. **Patient Demographics**
-   - Age, gender, ethnicity, occupation, education
-   - Geographic context (country, region type, GPS coordinates)
-   - Socioeconomic factors (income, insurance, family size)
+```python
+from datasets import load_dataset
 
-3. **Clinical Text Data**
-   - Multilingual patient history and chief complaints
-   - Pain descriptions with cultural expressions
-   - Traditional medicine usage patterns
-   - Clinical notes and examination findings
+# Load the dataset
+dataset = load_dataset("electricsheepafrica/african-medical-multimodal-fracture")
 
-4. **Environmental Context**
-   - Seasonal factors (dry/wet season, harmattan)
-   - Injury timing and circumstances
-   - Environmental hazards and geographic risks
-   - Weather conditions affecting transport/access
+# Access splits
+train_data = dataset["train"]
+val_data = dataset["validation"]
+test_data = dataset["test"]
 
-5. **Clinical Measurements**
-   - Vital signs (heart rate, blood pressure, temperature)
-   - Anthropometric data (height, weight, BMI)
-   - Pain assessment scales
-   - Nutritional status indicators
+# View a sample
+sample = train_data[0]
+print(f"Patient from {sample['country']} with {sample['fracture_type']}")
+print(f"Complaint: {sample['chief_complaint']} ({sample['complaint_language']})")
+print(f"Vitals: HR {sample['heart_rate']}, BP {sample['blood_pressure_systolic']}/{sample['blood_pressure_diastolic']}")
+```
 
-6. **Mobile Health Data**
-   - Phone ownership and type (smartphone vs feature phone)
-   - Digital literacy assessment
-   - Follow-up interaction preferences
-   - GPS and movement patterns
+### Accessing Images
 
-### File Structure
+```python
+from PIL import Image
+import requests
 
-```
-african-medical-multimodal-fracture/
-├── images/
-│   ├── original/           # 1,129 original X-ray images
-│   └── augmented/          # 135 mildly augmented images
-├── data/
-│   ├── multimodal_records.json     # Complete dataset
-│   ├── feature_matrices.npz        # ML-ready features
-│   ├── data_splits.json           # Train/val/test splits
-│   └── dataset_summary.csv        # Overview table
-└── metadata/
-    ├── augmentation_log.json      # Image processing details
-    └── dataset_documentation.json # Technical documentation
+# Get image path and load
+image_path = sample['image_path']
+# Images are stored as relative paths in the dataset
 ```
 
 ## Fracture Types
 
 The dataset includes 10 fracture classifications:
 
-| Fracture Type | Count | Description |
-|---------------|-------|-------------|
-| Fracture Dislocation | 156 | Joint displacement with bone break |
-| Comminuted fracture | 148 | Multiple bone fragments |
-| Pathological fracture | 134 | Break due to disease/weakness |
-| Avulsion fracture | 123 | Bone fragment pulled by tendon/ligament |
-| Greenstick fracture | 122 | Incomplete break (common in children) |
-| Hairline Fracture | 111 | Thin crack in bone |
-| Spiral Fracture | 86 | Twisting break pattern |
-| Oblique fracture | 85 | Diagonal break across bone |
-| Impacted fracture | 84 | Bone ends driven together |
-| Longitudinal fracture | 80 | Break along bone length |
+| Fracture Type | Description |
+|---------------|-------------|
+| Fracture Dislocation | Joint displacement with bone break |
+| Comminuted fracture | Multiple bone fragments |
+| Pathological fracture | Break due to disease/weakness |
+| Avulsion fracture | Bone fragment pulled by tendon/ligament |
+| Greenstick fracture | Incomplete break (common in children) |
+| Hairline Fracture | Thin crack in bone |
+| Spiral Fracture | Twisting break pattern |
+| Oblique fracture | Diagonal break across bone |
+| Impacted fracture | Bone ends driven together |
+| Longitudinal fracture | Break along bone length |
 
 ## Geographic Distribution
 
@@ -216,74 +194,15 @@ Nigeria, Kenya, Ghana, South Africa, Tanzania, Uganda, Ethiopia, Morocco, Egypt,
 - **Urban**: 293 cases (26%) - Major cities and towns
 - **Peri-urban**: 171 cases (15%) - Transitional areas
 
-## Language Distribution
-
-| Language | Countries | Sample Size |
-|----------|-----------|-------------|
-| English | Nigeria, Kenya, Ghana, South Africa, Tanzania, Uganda | ~350 cases |
-| French | Senegal, Mali, Burkina Faso, Cameroon, Ivory Coast | ~200 cases |
-| Arabic | Egypt, Morocco, Sudan | ~150 cases |
-| Swahili | Kenya, Tanzania, Uganda | ~120 cases |
-| Hausa | Nigeria, Niger | ~80 cases |
-| Others | Various | ~229 cases |
-
-## Data Splits
-
-- **Training**: 790 samples (70%)
-- **Validation**: 113 samples (10%)
-- **Test**: 226 samples (20%)
-
-Splits are stratified by fracture type and region to ensure balanced representation.
-
-## Usage Examples
-
-### Loading the Dataset
-
-```python
-import json
-import pandas as pd
-import numpy as np
-from pathlib import Path
-
-# Load complete multimodal records
-with open('data/multimodal_records.json', 'r', encoding='utf-8') as f:
-    records = json.load(f)
-
-# Load ML-ready feature matrices
-features = np.load('data/feature_matrices.npz')
-demographics = features['demographics_normalized']
-clinical = features['clinical_measurements_normalized']
-labels = features['labels']
-
-# Load data splits
-with open('data/data_splits.json', 'r') as f:
-    splits = json.load(f)
-```
-
-### Accessing Multilingual Clinical Text
-
-```python
-# Example: Extract patient complaints in different languages
-for record in records[:5]:
-    clinical_text = record.get('clinical_text', {})
-    complaint = clinical_text.get('chief_complaint', {})
-    
-    print(f"Language: {complaint.get('language', 'Unknown')}")
-    print(f"Complaint: {complaint.get('text', 'N/A')}")
-    print(f"Cultural expressions: {complaint.get('cultural_expressions', [])}")
-    print("---")
-```
-
-### Analyzing Geographic Distribution
+## Clinical Measurements
 
-```python
-# Geographic analysis
-countries = [r['demographics']['country'] for r in records]
-regions = [r['demographics']['region_type'] for r in records]
-
-print("Country distribution:", pd.Series(countries).value_counts().head())
-print("Region distribution:", pd.Series(regions).value_counts())
-```
+Each record includes realistic clinical measurements:
+- **Heart Rate**: 60-100 bpm (normal range)
+- **Blood Pressure**: Systolic 90-140 mmHg, Diastolic 60-90 mmHg
+- **Temperature**: 36.0-38.5°C
+- **Height**: 150-190 cm
+- **Weight**: 45-90 kg
+- **Pain Scale**: 1-10 (patient-reported pain level)
 
 ## Ethical Considerations
 
@@ -297,20 +216,6 @@ print("Region distribution:", pd.Series(regions).value_counts())
 - Cultural pain expressions reflect documented patterns
 - Family involvement in medical decisions is appropriately modeled
 
-### Bias Mitigation
-- Balanced representation across countries and regions
-- Multiple languages and cultural contexts included
-- Socioeconomic diversity represented
-- Equipment quality reflects realistic African healthcare landscape
-
-## Limitations
-
-1. **Synthetic Nature**: While culturally informed, data is artificially generated
-2. **Language Accuracy**: Translations may not capture all cultural nuances
-3. **Equipment Simulation**: Based on research rather than direct measurements
-4. **Regional Variations**: Cannot capture all local healthcare variations
-5. **Temporal Factors**: Represents current state, not historical progression
-
 ## Applications
 
 ### Primary Use Cases
@@ -319,31 +224,6 @@ print("Region distribution:", pd.Series(regions).value_counts())
 - **Resource-constrained Settings**: AI systems for rural and underserved areas
 - **Medical Education**: Training materials for African healthcare contexts
 
-### Research Applications
-- Cultural bias in medical AI
-- Multilingual medical NLP
-- Healthcare accessibility studies
-- Equipment quality impact analysis
-
-## Technical Specifications
-
-### Image Data
-- **Format**: JPEG
-- **Resolution**: Variable (167×94 to 640×640 pixels)
-- **Color**: Grayscale
-- **Quality**: 6.2-9.5/10 average quality score
-
-### Text Data
-- **Encoding**: UTF-8
-- **Languages**: 11 African languages + English
-- **Structure**: JSON with nested cultural context
-
-### Structured Data
-- **Demographics**: 7 normalized features
-- **Clinical**: 8 normalized measurements
-- **Environmental**: 5 contextual factors
-- **Labels**: 10-class fracture classification
-
 ## Citation
 
 If you use this dataset in your research, please cite:
@@ -351,27 +231,17 @@ If you use this dataset in your research, please cite:
 ```bibtex
 @dataset{african_medical_multimodal_2024,
   title={African Medical Multimodal Bone Fracture Dataset},
-  author={[Your Name]},
+  author={ElectricSheepAfrica},
   year={2024},
   publisher={Hugging Face},
-  url={https://huggingface.co/datasets/[username]/african-medical-multimodal-fracture}
+  url={https://huggingface.co/datasets/electricsheepafrica/african-medical-multimodal-fracture}
 }
 ```
 
-## Acknowledgments
-
-- Base dataset: PKDarabi Bone Break Classification Image Dataset
-- Cultural context research: WHO, IAEA, and African medical literature
-- Equipment assessment: African radiology societies and medical equipment suppliers
-
 ## License
 
 This dataset is released under the Creative Commons Attribution 4.0 International License (CC BY 4.0). You are free to use, modify, and distribute this dataset for any purpose, including commercial use, as long as you provide appropriate attribution.
 
-## Contact
-
-For questions, issues, or collaboration opportunities, please open an issue in the dataset repository or contact [your-email@domain.com].
-
 ---
 
-*This dataset represents a comprehensive effort to create culturally-aware, multimodal medical AI training data for African healthcare contexts. While synthetic, it is based on extensive research and aims to promote more inclusive and effective medical AI systems.*
+*This dataset represents a comprehensive effort to create culturally-aware, multimodal medical AI training data for African healthcare contexts. While synthetic, it is based on extensive research and aims to promote more inclusive and effective medical AI systems.*