MTFD-Net

MTFD-Net: Left Atrium Segmentation in CT Images through Fractal Dimension Estimation

This repository consists of the implementation of our network MTFD-Net, proposed for left atrium (LA) segmentation in computed tomography (CT) images directly utilizing fractal dimension (FD) maps calculated from the CT slices. MTFD-Net is a novel multi-task network that learns to simultaneously predict LA regions as the main task and estimate the FD maps as an auxiliary task.

The source codes are provided here.

The following source codes are provided for research purposes only. The authors have no responsibility for any consequences of the use of these source codes. If you use any part of these codes, please cite the following paper.

  • A. Saber Jabdaragh, M. Firouznia, K. Faez, F. Alikhani, J. Alikhani Koupaei, and C. Gunduz-Demir, “MTFD-Net: Left Atrium Segmentation in CT Images through Fractal Dimension Estimation,” Pattern Recognition Letters, 173:108-114, 2023.

Please contact Aziza Saber for further questions.

Source code

The provided zip file contains the necessary Python files for training and testing our proposed MTFD-Net model. The structure of the files is as follows:

  • fnet.sh: This file is used to start training the model. The model inputs and configuration can be specified in this file. Running instruction of the bash script file is included in the file.

  • MTFD_Net.py: This file consists of the model architecture for training.

  • data_generator.py: This is a custom data generator for loading the dataset as network input.

  • calculate_loss.py: The function needed to calculate class weight maps for weighted categorical cross-entropy loss is used from this file. It also consists of different loss functions that can be used if needed.

  • fnet_test.ipynb: The Jupyter Notebook file to test the model.

  • This code has been tested using the following requirements: Python 3.7.3, TensorFlow 1.13.1, Keras 2.2.4

  • Please see README.txt for further detail.