Extraction Module

The Extraction Module is divided into four components. The first three are pages dedicated to simple extraction processes on various data types (images, text notes, and time series) using pre-trained models and/or Python libraries. The fourth component is MEDimage, which is the implementation of a Python open-source package designed for medical image processing and radiomics features extraction.

In the video below, the "HAIM study" refers to the study of Soenksen et al.

Extraction Module Video Tutorial

Content

• 00:00 Overview

• 00:50 Extraction Images

• 04:24 Extraction Time Series

• 08:29 Extraction Text

• 10:48 Last Word

Data

• Physionet : https://physionet.org/

• MIMIC-IV Demo database : https://physionet.org/content/mimic-iv-demo/2.2/

• MIMIC-IV-Note : https://physionet.org/content/mimic-iv-note/2.2/

• MIMIC-IV-CXR : https://physionet.org/content/mimic-cxr-jpg/2.0.0/

• Study of Soenksen et al. : https://www.nature.com/articles/s41746-022-00689-4

Extraction tools

• TorchXRayVision : https://github.com/mlmed/torchxrayvision

• TSfresh : https://tsfresh.readthedocs.io

• BioBERT : https://arxiv.org/abs/1901.08746

• BioBERT weights : https://github.com/EmilyAlsentzer/clinicalBERT

---

Image Extraction PageText Extraction PageTime Series Extraction PageMEDimage