# PD Cognition BioClinicalBERT Fine Tuned spaCy Model

This model is a spaCy pipeline for span categorization trained on clinical text related to Parkinson disease cognition. It uses BioClinicalBERT as the transformer backbone and predicts labeled spans stored under the key sc.

## Model Details

Base model emilyalsentzer Bio ClinicalBERT
Framework spaCy with spacy transformers
Task span categorization
Language English

## Labels

The label set is defined in labels.json included in the model directory.

## Input: 
Raw clinical text string

## Output: 
Predicted spans in doc.spans["sc"] with start end label and score

## Usage

```python
import spacy

nlp = spacy.load("path_to_model_best")
doc = nlp("Patient shows cognitive decline and memory impairment")

for span in doc.spans["sc"]:
    print(span.text, span.label_, span.score)
```

## Files

model best spaCy pipeline
config cfg training configuration
meta json pipeline metadata
labels json list of span labels

## Notes

This is a spaCy model and should be loaded with spacy.load not transformers
Performance depends on span alignment and threshold tuning
Intended for research use on clinical text

## Citation

```bibtex
@article{khanna2024cognitive,
  title={Toward Automated Cognitive Assessment in Parkinson’s Disease Using Pretrained Language Models},
  author={Khanna, Varada and Bhatt, Nilay and Shin, Ikgyu and Rosso, Mattia and Tinaz, Sule and Ren, Yang and Xu, Hua and Keloth, Vipina K},
  journal={arXiv preprint arXiv:2511.08806},
  year={2025}
}
```