You can deploy this model and start earning money today!

cja5553/
Bio_ClinicalBERT_MIMIC_IV_discharge_within_12_hours_prediction_lora_ti

Name: Bio_ClinicalBERT_MIMIC_IV_discharge_within_12_hours_prediction_lora_ti
Availability: InStock

peft

safetensors

emilyalsentzer/Bio_ClinicalBERT

adapter

Views

No views yet

Bio_ClinicalBERT_MIMIC_IV_discharge_within_12_hours_prediction_lora_ti

This model is designed to predict if a patient will get discharged within 12-hours based on the prior hospital records. It is trained on clinical notes from prior hospitalizations on MIMIC-IV. Model was trained on a novel tabular-infused LoRA, whereby the pre-operative tabular features (e.g., patient demographics and insurance information) were used to initialize the newly introduced LoRA parameters.

Model Details

Data: MIMIC-IV (available here)
Outcome: Discharge in 12-hours (True/False)
Base Model: emilyalsentzer/Bio ClinicalBERT

How to use model

python

from transformers import AutoTokenizer, AutoModelForSequenceClassification
tokenizer = AutoTokenizer.from_pretrained("cja5553/Bio_ClinicalBERT_MIMIC_IV_discharge_within_12_hours_prediction_lora_ti")
model = AutoModelForSequenceClassification.from_pretrained("cja5553/Bio_ClinicalBERT_MIMIC_IV_discharge_within_12_hours_prediction_lora_ti")

Then you can use this function below to get one test point

python

import torch

def get_outcome(tokenizer, model, text, device="cuda:0", max_length=512):

    device = torch.device(device)
    model = model.to(device)
    model.eval()

    inputs = tokenizer(
        text,
        return_tensors="pt",
        max_length=max_length,
        truncation=True,
        padding="max_length"
    ).to(device)

    with torch.no_grad():
        outputs = model(**inputs)
        probs = torch.softmax(outputs.logits, dim=-1)[0]  # (2,)

    probs = probs.detach().cpu().numpy()
    result = {
        "False": float(probs[0]),
        "True": float(probs[1])
    }

    return result

Questions?

Contact me at alba@wustl.edu