HIVE阑尾炎预测模型与LLM数据提取代码

数据集获取（Figshare）

• 源站标题：Code for the HIVE Appendicitis prediction modelRepository with LLM_data_extractor_optuna for automated feature extraction
• DOI：10.6084/m9.figshare.28931030.v7
• 发布日期：2025-11-09
• 文件数：4 个
• 文件总体积：862.47 KB
• 源站页面：https://figshare.com/articles/dataset/Code_for_developing_and_validating_HIVE_model_for_Appendicitis_Prediction_XGBoost_/28931030

命令行下载

curl -L -o "HIVE_model.pkl" "https://ndownloader.figshare.com/files/54201830"
curl -L -o "05MAY2025 Appendicitis Prediction XGBoost LLM study.ipynb" "https://ndownloader.figshare.com/files/56185025"
curl -L -o "LLM_data_extractor_optuna_repo.zip" "https://ndownloader.figshare.com/files/59416175"
curl -L -o "README.md" "https://ndownloader.figshare.com/files/59416184"

文件清单

• HIVE_model.pkl（82.26 KB）
• 05MAY2025 Appendicitis Prediction XGBoost LLM study.ipynb（40.81 KB）
• LLM_data_extractor_optuna_repo.zip（733.73 KB）
• README.md（5.67 KB）

文件总体积 862.47 KB（源站 API 实测）。

数据加载示例（图像类）

from PIL import Image
import glob, os

files = (glob.glob(os.path.join(path, "**", "*.png"), recursive=True)
       + glob.glob(os.path.join(path, "**", "*.jpg"), recursive=True)
       + glob.glob(os.path.join(path, "**", "*.tif"), recursive=True))
print("图像文件数:", len(files))
img = Image.open(files[0]); print("尺寸/模式:", img.size, img.mode)

# torchvision Dataset 方式:
# from torchvision import datasets
# ds = datasets.ImageFolder(path)  # 要求 子目录=类别

目录组织与标注格式以源站说明和下载后实际文件为准。

数据说明

This repository contains the code and trained model for developing the HIVE (History, Intake, Vitals, Examination) machine learning model for predicting appendicitis in patients presenting with acute abdominal pain at the Emergency Department (ED).
The HIVE model was developed using structured data derived from ED intake forms, vital signs, and clinical signs and symptoms extracted from free-text ED reports. These clinical features were either manually annotated or automatically extracted using a large language model (LLM), depending on the experiment.
The codebase includes:

• Preprocessing scripts for merging structured and unstructured inputs
• Feature engineering and selection steps
• Model development using XGBoost
• Hyperparameter tuning via Optuna
• Evaluation procedures including AUROC calculation and bootstrapping for confidence intervals
  The repository also includes a pickled version of the final trained model (developed on 268 training cases), which can be used to generate appendicitis risk predictions on new patient data when provided with the appropriate input features.
  LLM Data Extractor optuna repo is a Python framework for generating and evaluating clinical text predictions using large language models (LLMs) like qwen2.5. It supports:
• Prediction task execution via a local Ollama server
• Hyperparameter tuning with Optuna across temperature, top-k, top-p, and min-p
• Designed to extract structured outputs from unstructured text data (e.g., ED reports)
• Fully configurable via CLI for automated or fine-tuned runs.
• Please read the attached readme.md for a brief walkthrough how to use the LLM JSON pipeline (with examples).
  The pipeline supports both direct prediction generation and structured evaluation with minimal setup. It is based on the llm_extractinator/README.md at main · DIAGNijmegen/llm_extractinator · GitHub.
  This repository is linked to the paper Large Language Model Automated Extraction of Clinical Signs and Symptoms From Emergency Department Reports for Machine Learning Prediction Models: A Development and Validation Study.

许可

CC BY 4.0

引用

学术使用请引用 DOI 10.6084/m9.figshare.28931030.v7（source: https://figshare.com/articles/dataset/Code_for_developing_and_validating_HIVE_model_for_Appendicitis_Prediction_XGBoost_/28931030）。