stevenah / mimir

Mimir Automatic Reporting

This is Mimir, an autoamtic reporting system aimed at medical doctors for endoscopic image analysis and reporting. May also be used to diagnose convolutional neural networks for potential imrpovements through the use of grad-CAM and guided grad-CAM visualizations.

Getting Started

The following instructions should get you up and running with a local instance of Mimir.

Before we start

It is recommended that Mimir is run on a computer using a GPU. Although Mimir will run on the CPU alone, it may be considerably less performant.

Prerequisites

Before we run Mimir, we must first make sure the following programs are installed.

• Git - Download & Install Git and git lfs to download the sample Keras models.
• Python 3.6+ - Download & Install Python 3.6 and the package manager pip.
• OpenCV 3.0+ - Download & Install OpenCV 3 and make sure it supports FFmpeg, this is required for video processing.
• SQLite - Download & Install SQLite.

Installation

Before we start, make sure that you clone this repository to your local computer by running git clone https://github.com/stevenah/mimir in the directory of your choice. With all prerequisites installed, running pip -r install requirements.txt should install all required python dependencies. This is all that is needed for a basic local instance of Mimir. To pull down the sample Keras models included with Mimir, run git lfs pull in the root directory.

Usage

To start a local instance of Mimir, it should be enough to run the app.py file located in the server directory.

Built With

• Flask - The web framework used
• React - Front-end view library used
• Keras - Deep learning framework used
• Tensorflow - Deep learning library used

Authors

• Steven Hicks - Work done as part of master thesis - Stevenah

License

This project is licensed under the GNU General Public License v3.0 - see the LICENSE.md file for details