heart-murmur-detection

KO | EN

gitlite — search

#ai #llm #golang #go #deep-learning #docker #macos #nodejs #java #awesome #awesome-list #pytorch #security #ios

heart-murmur-detection

(★ 23)

Dual Bayesian ResNet: A Deep Learning Approach to Heart Murmur Detection (Physionet Challenge 2022)

heart-murmur-detection 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)

## Heart Murmur Detection using Bayesian Residual Neural Networks

### This is the extension of our entry to the George B. Moody PhysioNet Challenge 2022: Heart Murmur Detection from Phonocardiogram Recordings

When using this code, please cite [the challenge paper](https://cinc.org/archives/2022/pdf/CinC2022-355.pdf) as well as [the journal paper]().

This repository contains our solution to the first task from the PhysioNet 
Challenge 2022: Design an algorithm to classify the present, absent, or unknown cases 
of heart murmurs from heart sound recordings [1,2].

Two models are implemented:
* A Dual Bayesian ResNet (DBRes), where overlapping log mel spectrograms of the 
heart sound recordings undergo two binary classifications simultaneously: present 
versus unknown or absent, and unknown versus present or absent.
* The output from DBRes integrated with demographic data and signal features 
using XGBoost.

### Challenge performance for murmur task on held-out subset

Using this code, we reached the following performance:

| Model                 | Accuracy      | Murmur  |
| -------------         | ------------- | ------------- |
| DBRes                 | 0.762         | 0.780 |
| DBRes with XGBoost    | 0.820         | 0.749 |

## Data

The challenge dataset can be downloaded via this 
[link](https://physionet.org/static/published-projects/circor-heart-sound/the-circor-digiscope-phonocardiogram-dataset-1.0.3.zip)
or using the following command.
```angular2html
wget -r -N -c -np https://physionet.org/files/circor-heart-sound/1.0.3/
```

## Dependencies

- At least Python 3.9;
- [PyTorch (torch, torchvision)](https://github.com/pytorch/pytorch/) for neural network architecture and training;
- [XGBoost](https://github.com/dmlc/xgboost) for xgboost;
- [Librosa](https://github.com/librosa/librosa) for audio processing and feature extraction;

### Install using conda on cuda device

```
conda create -n myenv python=3.9
conda activate myenv
conda install pytorch==1.11.0 torchvision==0.12.0 cudatoolkit=11.3 -c pytorch
pip install tqdm==4.64.1 scipy==1.9.2 resampy==0.4.2 scikit-learn==1.1.2 pandas==1.5.0 xgboost==1.6.2 librosa==0.9.2
```

### Install using pipenv (mps enabled)

```
pipenv install
```

## Running Experiments
You can run all steps using the template files in `run_files`.`

A full experiment consists of four steps:
1. Splitting the data into stratified train, test, and validation sets (`data_split.py`).
2. Training the Bayesian ResNets on their respective binary classification tasks (`train_resnet.py`).
3. Calculating and evaluating the output from DBRes (`dbres.py`).
4. Calculating and evaluating the output from DBRes with XGBoost integration (`xgboost_integration.py`).

These steps can be run independently using the relevant script, or sequentially using `main.py`,
```angular2html
CUDA_VISIBLE_DEVICES=0 python main.py --full_data_directory physionet.org/files/circor-heart-sound/1.0.3/training_data --stratified_directory data/stratified_data --vali_size 0.2 --test_size 0.2 --random_state 14 --recalc_features --spectrogram_directory data/spectrograms --model_name resnet50dropout --recalc_output --dbres_output_directory outputs
```

## Affiliations

Ben Walker1, Felix Krones2, Ivan Kiskin3,4, 
Guy Parsons5, Terence Lyons1, Adam Mahdi2,3

1. The Mathematical Institute, University of Oxford, Oxford, UK, 
2. The Oxford Internet Institute, University of Oxford, Oxford, UK, 
3. The People-Centred AI Institute, University of Surrey, Surrey, UK, 
4. The Surrey Sleep Research Centre, University of Surrey, Surrey, UK, 
5. The Intensive Care Registrar, Thames Valley Deanery, NIHR Academic Clinical Fellow at University of Oxford, Oxford, UK.

## Acknowledgment

This work was supported by the Hong Kong Innovation and Technology Commission (InnoHK Project CIMDA) and the Friedrich Naumann Foundation.

## Bibtex Citation

```bibtex
@article{walker2022DBResNet,
    title={Dual Bayesian ResNet: A Deep Learning Approach to Heart Murmur Detection},
    author={Benjamin Walker and Felix Krones and Ivan Kiskin and Guy Parsons and Terry Lyons and Adam Mahdi},
    journal={Computing in Cardiology},
    volume={49},
    year={2022}
}
```

heart-murmur-detection

heart-murmur-detection 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

heart-murmur-detection

heart-murmur-detection 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README