Copyright (C) 2022 Xu Tian (tianxu@csu.edu.cn), Jin Liu (liujin06@csu.edu.cn)
## Package Title:
3D MRI-based Multi-Task Learning for Alzheimer's Disease Diagnosis and MMSE Score Prediction: A Multi-site Validation
## Description:
This package is designed to enable automatic AD diagnosis and MMSE score prediction through multi-task interaction learning from structural magnetic resonance imaging (sMRI) brain scans.
![Framework of our proposed method](image/framework.png)
<p align="center">Fig. 1. Framework of our proposed method</p>
As is shown in Fig.1. , we propose an MRI-based multi-task interaction learning (MTIL) method for AD diagnosis and MMSE score prediction. Firstly, we implement both tasks using the same backbone. We insert three multi-task interaction layers between these two backbones. Each multi-task interaction layer consists of two feature decoupling modules and one feature sharing module. It takes interactive features of each task and interacts them to obtain shared representations for both tasks. Secondly, in order to further improve the consistency of the features selected by the feature decoupling module, we design a feature consistency loss to further constrain the module. Finally, in order to exploit the specific distribution information of MMSE scores in AD group and NC group, we design a distribution loss.
In this code, "config.json" places the model's hyperparameters and other information. "model.py" places the main CNN model. "model_wrapper.py" encapsulates the training, validation and testing process of the model. "model.py" encapsulates the main CNN model. "dataloader.py" encapsulates the data usage process required by the model. "loss.py" contains forward and backward computations for part of the model's loss.
## How to run this project:
This project must run in python>=3.6, The following steps should be taken to run this project:
1. Data preparation: Before the code can run, the data needs to be prepared. You need to put a file named "dataset.csv" in the "opencsv" folder. "dataset" is the name of the dataset in the configuration file. In this csv files, subjects' AD labels and MMSE scores should be placed in the second and third columns respectively.
To demonstrate this code, we provide data for 10 subjects from the ADNI1 dataset. These subjects were randomly selected from the test set of the experiments for the provided model files. These data are placed in the "data" folder and their label information is in the "opencsv" folder.
2. Environment building:
(1) Software: Information about the packages required by the code is at "requirements.txt".
(2) Hardware: This code has been tested with NVIDIA GTX2080.
3. Code running:
(1) Information about code running in "config.json" should be modified.
(2) You can use
python main.py
to complete the training and testing of the model, or use
python main.py train
python main.py test
to complete them separately.
Also, we put our trained model file in this code, it can be directly used for AD detection and MMSE score prediction. If you want to use it, execute the following command
python main.py valid
基于深度学习的磁共振成像脑部扫描实现自动阿尔茨海默病内含数据集和运行环境说明.zip
版权申诉
120 浏览量
2024-05-06
10:56:09
上传
评论
收藏 27.45MB ZIP 举报
小码蚁.
- 粉丝: 2534
- 资源: 4146
最新资源
- 基于Python的新能源承载力计算及界面设计源码 - HAINING-DG
- 基于Java的本科探索学习项目设计源码 - 本科探索
- 基于Javascript和Python的微商城项目设计源码 - MicroMall
- 基于Java的网上订餐系统设计源码 - online ordering system
- 基于Javascript的超级美眉网络资源管理应用模块设计源码
- 基于Typescript和PHP的编程知识储备库设计源码 - study-php
- Screenshot_2024-05-28-11-40-58-177_com.tencent.mm.jpg
- 基于Dart的Flutter小提琴调音器APP设计源码 - violinhelper
- 基于JavaScript和CSS的随寻订购网页设计源码 - web-order
- 基于MATLAB的声纹识别系统设计源码 - VoiceprintRecognition
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈