Master thesis: Detection and removal of texts burned in medical images using deep learning

• Author : Szelagowski Nicolas
• Promotor : Jodogne Sébastien
• Academic year : 2022-2023

Project Structure

MTC

This directory contains the source code of the MTC interface and the instructutions to compile the Vue.js project.

OrthancPlugin

This directory contains the necessary files to integrate MedTextCleaner (MTC) into Orthanc, a python plugin designed to help users remove texts burned in medical images.

SSD-TB

This directory contains the files used to train the models (Textboxes and SSD).

evaluation

This directory contains the files used in order to evaluate our models, including the detection evaluation protocol scripts (icdar, deteval, coco) and a google colab notebook inspired from this tutorial.

dataset_generator.ipynb

This notebook takes care of generating synthetic data on JPEG images and produces matching annotation files.

dicomTojpeg.py

This python script allows to convert DICOM instances to JPEG images using an Orthanc Server.

interactiveClasifier.ipynb

This notebook provides a user interface designed to simplify the manual classification of images into two categories.