swag-tag: A Streamlit-based webapp for annotation of medical images and reports

Swag-Tag is a web application built using Streamlit that enables efficient annotation of medical images and reports via tagging. The reports and scans as well as the added tags are all saved in a postgres database. We add a prefill option that uses Llama-2-70B (4bit/8bit quantization) to prefill the given template.

Features

Current features are:

• Displaying a report-scan combination from the database
  • If multiple scans are present, they are displayed in tabs
  • Reports are displayed divided into several categories like "Examination" and "Findings"
  • The user can select prefilled annotations (by an LLM) or
• Applying prefilled LL
• Setting min and max values for windowing X-Ray DICOMs
• Moving between slices along the z-axis for 3D-scans (e.g. CT)
• Moving between dataset cases and specifying a case to jump directly to it
• Previous annotations are loaded directly into the case.
• loading and saving user configurations
• Templates can be defined using JSON files.
  • The template allows nested structures with one-of-many and multi-select children nodes
  • In future we will provide a GUI to build the templates
• For planned future features and requests please have a look at the issues section!

Installation

Follow this step-by-step guide to install the project on your local system. We firmly recommend using conda on Linux since one of the required packages, psycopg2, has bugs regarding system-level libraries when installed via pip.

1. Clone this repository

   git clone https://github.com/Baessler-Lab/swag-tag.git
   cd swag-tag
   

Configuration

The configuration is stored under swagtag/config/config.yaml. Change the settings (e.g. directories) to your needs.

1. Make sure that the volumes (e.g. mimic files and templates) point to the right place on host system

2. Make sure that a postgres image runs in the docker container (or separate host) defined in docker/config/db_config.yaml

Run as a docker container

1. build and deploy docker container.

   cd docker
   docker compose up -d
   

(deprecated)

2. Create a new conda environment using the environment.yml provided in the repo and activate it

conda env create -f environment.yml
conda activate swag-tag

Configuration

The configuration is stored under swagtag/config/config.yaml. Change the settings (e.g. directories) to your needs.

Run (on host)

You can run the streamlit application by running

bash start-app.sh