schema · 13 fields
A database of artificial intelligence applications in healthcare
Automate extraction for "A database of artificial intelligence applications in healthcare". Extract structured, evidence-grounded data points directly from hundreds of papers with source page coordinates.
Schema Field Definitions (13)
| Field Name | Data Type | Description |
|---|---|---|
| medical_specialty | string | The medical specialty or clinical domain addressed by the AI application, such as Radiology, Pathology, Cardiology, or Oncology. |
| target_disease_or_condition | string | The specific disease, pathology, or clinical condition targeted by the AI system. Return null if the application is not disease-specific. |
| clinical_task | string | The specific clinical or operational task performed by the AI system, such as lesion segmentation, mortality risk prediction, automated diagnosis, or triage. |
| data_modality | string | The primary input data type used to train or evaluate the model, such as Chest X-ray, MRI, EHR tabular records, Whole Slide Images (WSI), or ECG signals. |
| model_architecture | string | The specific machine learning or deep learning architecture, algorithm, or framework used, such as U-Net, ResNet-50, XGBoost, or Vision Transformer. |
| total_dataset_size | number | The total numerical count of data instances (e.g., patients, scans, records) included in the study's training, validation, and test datasets combined. |
| total_dataset_size_unit | string | The unit of measurement corresponding to the total dataset size, such as patients, images, slides, visits, or records, exactly as reported in the study. |
| validation_method | string | The validation strategy used to evaluate the AI model. Must be one of: Internal split, K-fold cross-validation, External independent cohort, Prospective clinical evaluation. Return null when the paper does not provide enough information to select one. |
| primary_performance_metric_name | string | The name of the primary metric reported for evaluating model performance on the test/validation cohort, such as AUROC, Sensitivity, Specificity, F1-score, or Dice Similarity Coefficient. |
| primary_performance_metric_value | number | The numerical score of the primary performance metric reported for the best-performing model on the test or validation set. |
| primary_performance_metric_unit | string | The unit associated with the extracted primary performance metric value (e.g., %, fraction 0-1), exactly as reported in the paper. |
| prospective_validation_performed | boolean | True if the AI application was evaluated in a real-time prospective clinical setting or clinical trial within the study; false if validation was solely retrospective. |
| code_publicly_available | boolean | True if the study authors state that the model code, architecture weights, or implementation scripts are publicly accessible in a repository. |
Data Verification Status Taxonomy
To maintain rigorous scientific standards, Sci-database differentiates between automated signals and human audit:
Extracted
Value located and extracted from the full-text PDF by the Kateeb engine.
Model-Agreed
An independent second model produced a compatible value during automated second-pass review.
Human-Reviewed
A named researcher audited the value against the visual page coordinates.
Benchmark-Validated
Performance evaluated against a standardized, published evaluation protocol.
Replicated
An independent external research group reproduced the extraction and experimental result.