NahianHasan/Cardiovascular_Disease_Classification_Employing_EMD

Cardiovascular Disease Classification Employing Empirical Mode Decomposition (EMD) of Modified ECG

Cardiovascular_Disease_Classification_Employing_EMD

Cardiovascular Disease Classification Employing Empirical Mode Decomposition (EMD) of Modified ECG

License

This repository is licensed under the MIT License - see the LICENSE file for details

Dependencies

• tensorflow
• teensorflow-gpu
• theano
• keras
• scikit-learn
• numpy
• matplotlib
• scipy
• wfdb package of PhysioNet
• GPUtil
• pickle
• pydot
• graphviz
• pyhht

Citation

If you are benefited by this repository or the research, please cite the work as follows.

• @article{HASAN2019128,
  title = "Deep Learning Approach to Cardiovascular Disease Classification Employing Modified ECG Signal from Empirical Mode Decomposition",
  journal = "Biomedical Signal Processing and Control",
  author = "Nahian Ibn Hasan and Arnab Bhattacharjee",
  volume = "52",
  pages = "128 - 140",
  year = "2019",
  issn = "1746-8094",
  doi = "https://doi.org/10.1016/j.bspc.2019.04.005",
  url = "http://www.sciencedirect.com/science/article/pii/S1746809419301028"}

Purpose and Features

• Cardiovascular disease classification based on Modified ECG employing Empirical Mode Decomposition.
• Different modes of training such as maximum voting strategy, parallel mode and etc.
• Compare modified ECG and raw ECG signal
• Visualization of Empirical Mode Decomposition
• Validate on multiple database such as MIT-BIH Arrhythmia database, PTB diagonstic database, St.-Petersburg Arrhythmia database etc. from PhysioNet.
• Different combinations of IMF testing
• Customizable to any custom built database
• Analyze the test result through confusion matrix
• custom learning rate scheduler implementation
• custom network implementation
• 1-D version of many standard DNN architectures such as Alexnet, ZFNet, Resnet150, GoogleNet and etc.
• provision for individual network structure for each of IMF signals

Corresponding Author and Principal Contributor

• Nahian Ibn Hasan
• Email: nahianhasan1994@gmail.com
• LinkedIn: www.linkedin.com/in/nahian-ibn-hasan-buet
• Bangladesh University of Engineering and Technology (BUET)
• Dhaka, Bangladesh

Co-Author of the article

• Arnab Bhattacharjee
• Email:pritamarnab@gmail.com
• Bangladesh University of Engineering and Technology (BUET)
• Dhaka, Bangladesh

Getting Started

This readme file will instruct you to run the codes in this repository and customize it to your own database.

Folder Files

• config.py--------------------------------- contains parameters.
  
• Folder_creation.py------------------------ creates separate folders to store the training outputs.
  
• EMD_Models.py----------------------------- contains different network architectures to train the model.
  
• EMD_Models_Parallel.py-------------------- contains parallel architectures for training. [for details refer to the original elsevier article.]
  
• EMD_data_prepare.py------ code for generating IMF signals from EMD analysis of signals.
  
• EMD_data_train_all_combinations_pc.py----- train on specified IMF combination in a CPU
  
• EMD_data_train_any_combinations.py-------- train on specified IMF combination in a GPU
  
• EMD_data_training_pc.py------------------- train on individual IMF in CPU
  
• EMD_data_training_server.py--------------- train on individual IMF in GPU
  
• EMD_data_training_server_ensembled.py----- train on IMF[1,2,3] combination only in GPU
  
• EMD_ensembled_data_prepare_Test_server.py- Modified ECG signal formation from IMF[1,2,3] for test purpose.
  
• EMD_ensembled_data_prepare_Train_server.py-Modified ECG signal formation from IMF[1,2,3] for train purpose.
  
• EMD_Test_all_combinations.py-------------- Test code on specified IMF combination
  
• EMD_Test.py------------------------------- Test code on individual IMF.
  
• EMD_Parallel_Test.py---------------------- Test code on parallel architecture
  
• EMD_Parallel_Test_3_IMF.py---------------- Test code on parallel architecture of 3 specified IMFs only.
  
• EMD_Test_original.py---------------------- Test code on original raw signal
  
• Original_signal_Data_prepare.py----------- Prepare original raw data for training
  
• confusion_mat_plot.py--------------------- Confusion matrix plot[used by Confusion_Matrix.py]
  
• Confusion_Matrix.py----------------------- Confusion matrix plot
  
• Confusion_Matrix_Parallel.py-------------- Confusion matrix for parallel architecture
  
• all_combinations_plot.py------------------ Plot the accuracy of all combinations used for modified ECG

FILES INSIDE 'Noise Analysis' FOLDER

• modified_ECG_analysis_MIT_BIH.py---------- Pearson correlation coefficient calculation for each of the classes considered from MIT-BIH Arrhythmia database
  
• modified_ECG_analysis_PTB_Petes.py-------- Pearson correlation coefficient calculation for each of the classes considered from PTB and ST.-Petersberg database
  

FILES INSIDE 'Visualization' FOLDER

• EMD_visualization_petersburg.py----------- IMFs visualization of Petersburg Database
  
• EMD_visualization_PTB.py------------------ IMFs visualization of PTB databse
  
• learning_rate_scheduler.py---------------- learning rate scheduler visualize used in training
  
• modified_EMD_visualization_Petersburg.py-- Modified EMD comparison with raw signal of Petersburg database
  
• modified_EMD_visualization_PTB.py--------- Modified EMD comparison with raw signal of PTB database
  

Code Run

1.EMD_data_train_any_combinations.py

python EMD_data_train_any_combinations.py -options

options
* means mandatory argument

• -p = Path to the main database | deafault = config.data_path
• -c = Path to the CSV Folder of EMD Data | default = config.IMF_csv_path
• -res = Resume Training | default='False'
• -inep = Initial Epoch after Resuming Training | default=config.initial_epoch
• -reim = Resumed IMF number after resuming | default=1
• -rc = Path to the Patient file RECORD.txt | default=config.patient_data_path
• -pd = Path to the text file where problematic data to be stored | default=config.preoblem_data_path
• -s = Number of samples to be taken by each record",type=int | default=config.samplenumber
• -imf = Number of IMFs to be extracted | default=config.number_of_IMFs | choices=[2,3,4,5,6]
• -spl = Splitting percentage of train and test(upper limit) | type=float | default=config.split_perc
• -fold* = Save training and testing results in folder

2.EMD_data_train_all_combinations_pc.py

python EMD_data_train_all_combinations_pc.py -options

options
* means mandatory argument

• -p = Path to the main database | deafault = config.data_path
• -c = Path to the CSV Folder of EMD Data | default = config.IMF_csv_path
• -res = Resume Training | default='False'
• -inep = Initial Epoch after Resuming Training | default=config.initial_epoch
• -reim = Resumed IMF number after resuming | default=1
• -rc = Path to the Patient file RECORD.txt | default=config.patient_data_path
• -pd = Path to the text file where problematic data to be stored | default=config.preoblem_data_path
• -s = Number of samples to be taken by each record",type=int | default=config.samplenumber
• -imf = Number of IMFs to be extracted | default=config.number_of_IMFs | choices=[2,3,4,5,6]
• -spl = Splitting percentage of train and test(upper limit) | type=float | default=config.split_perc
• -fold* = Save training and testing results in folder

3.EMD_data_training_server.py

python EMD_data_training_server.py -options

options
* means mandatory argument

• -p = Path to the main database | deafault = config.data_path
• -c = Path to the CSV Folder of EMD Data | default = config.IMF_csv_path
• -res = Resume Training | default='False'
• -inep = Initial Epoch after Resuming Training | default=config.initial_epoch
• -reim = Resumed IMF number after resuming | default=1
• -rc = Path to the Patient file RECORD.txt | default=config.patient_data_path
• -pd = Path to the text file where problematic data to be stored | default=config.preoblem_data_path
• -s = Number of samples to be taken by each record",type=int | default=config.samplenumber
• -imf = Number of IMFs to be extracted | default=config.number_of_IMFs | choices=[2,3,4,5,6]
• -spl = Splitting percentage of train and test(upper limit) | type=float | default=config.split_perc
• -fold* = Save training and testing results in folder

4.EMD_data_training_pc.py

python EMD_data_training_pc.py -options

options
* means mandatory argument

• -p = Path to the main database | deafault = config.data_path
• -c = Path to the CSV Folder of EMD Data | default = config.IMF_csv_path
• -res = Resume Training | default='False'
• -inep = Initial Epoch after Resuming Training | default=config.initial_epoch
• -reim = Resumed IMF number after resuming | default=1
• -rc = Path to the Patient file RECORD.txt | default=config.patient_data_path
• -pd = Path to the text file where problematic data to be stored | default=config.preoblem_data_path
• -s = Number of samples to be taken by each record",type=int | default=config.samplenumber
• -imf = Number of IMFs to be extracted | default=config.number_of_IMFs | choices=[2,3,4,5,6]
• -spl = Splitting percentage of train and test(upper limit) | type=float | default=config.split_perc
• -fold* = Save training and testing results in folder

5.EMD_data_training_server_ensembled.py

python EMD_data_training_server_ensembled.py -options

options
* means mandatory argument

• -p = Path to the main database | deafault = config.data_path
• -c = Path to the CSV Folder of EMD Data | default = config.IMF_csv_path
• -res = Resume Training | default='False'
• -inep = Initial Epoch after Resuming Training | default=config.initial_epoch
• -reim = Resumed IMF number after resuming | default=1
• -rc = Path to the Patient file RECORD.txt | default=config.patient_data_path
• -pd = Path to the text file where problematic data to be stored | default=config.preoblem_data_path
• -s = Number of samples to be taken by each record",type=int | default=config.samplenumber
• -imf = Number of IMFs to be extracted | default=config.number_of_IMFs | choices=[2,3,4,5,6]
• -spl = Splitting percentage of train and test(upper limit) | type=float | default=config.split_perc
• -fold* = Save training and testing results in folder

6.EMD_data_prepare.py

python EMD_data_prepare.py -options

options
* means mandatory argument

• -p = Path to the main database | deafault = config.data_path
• -c = Path to the CSV Folder of EMD Data | default = config.IMF_csv_path
• -rc = Path to the Patient file RECORD.txt | default=config.patient_data_path
• -pd = Path to the text file where problematic data to be stored | default=config.preoblem_data_path
• -s = Number of samples to be taken by each record",type=int | default=config.samplenumber
• -imf = Number of IMFs to be extracted | default=config.number_of_IMFs | choices=[2,3,4,5,6]
• -spl = Splitting percentage of train and test(upper limit) | type=float | default=config.split_perc

7.EMD_ensembled_data_prepare_Train_server.py

python EMD_ensembled_data_prepare_Train_server.py -options

options
* means mandatory argument

• -p = Path to the main database | deafault = config.data_path
• -c = Path to the CSV Folder of EMD Data | default = config.IMF_csv_path
• -rc = Path to the Patient file RECORD.txt | default=config.patient_data_path
• -pd = Path to the text file where problematic data to be stored | default=config.preoblem_data_path
• -s = Number of samples to be taken by each record",type=int | default=config.samplenumber
• -imf = Number of IMFs to be extracted | default=config.number_of_IMFs | choices=[2,3,4,5,6]
• -trl* = training file list as a csv file

8.EMD_ensembled_data_prepare_Test_server.py

python EMD_ensembled_data_prepare_Test_server.py -options

options
* means mandatory argument

• -p = Path to the main database | deafault = config.data_path
• -c = Path to the CSV Folder of EMD Data | default = config.IMF_csv_path
• -rc = Path to the Patient file RECORD.txt | default=config.patient_data_path
• -pd = Path to the text file where problematic data to be stored | default=config.preoblem_data_path
• -s = Number of samples to be taken by each record",type=int | default=config.samplenumber
• -imf = Number of IMFs to be extracted | default=config.number_of_IMFs | choices=[2,3,4,5,6]
• -tel* = test file list as a csv file

Stay tuned for further tutorials