数据集获取

git clone https://github.com/AndrewRomitti/Evaluating-The-Effectiveness-of-Synthetic-Datasets-for-Dementia-Diagnosis-Using-Deep-Learning.git

curl -L -o repo.zip https://github.com/AndrewRomitti/Evaluating-The-Effectiveness-of-Synthetic-Datasets-for-Dementia-Diagnosis-Using-Deep-Learning/archive/refs/heads/main.zip
unzip repo.zip

源站 README 摘录（使用方式）

Evaluating-The-Effectiveness-of-Synthetic-Datasets-for-Dementia-Diagnosis-Using-Deep-Learning

This is the code for the paper: Evaluating the Effectiveness of Synthetic Datasets for Dementia Diagnosis for Dementia Diagnosis Using Deep Learning. This paper was presented and published at the 2023 IEEE AIPR Workshop as one of the 15 minute presentations.
Citation: Romitti, Andrew, Jiya Shetty, and Praveen Rao. “Evaluating the Effectiveness of Synthetic Datasets for Dementia Diagnosis Using Deep Learning.” 2023 IEEE Applied Imagery Pattern Recognition Workshop (AIPR). IEEE, 2023.
Link: https://www.computer.org/csdl/proceedings-article/aipr/2023/10440719/1UKgJwpsg7u

Abstract

Early and accurate diagnosis of dementia can lead to better treatment of the disease and improve patients’’ quality of
life. Advanced neuroimaging technologies such as magnetic resonance imaging (MRI) and deep learning hold promise for early and accurate dementia diagnosis. However, there is limited number of real-world MRI datasets for training deep-learning
models to classify a patient’‘s degree of dementia. Generative adversarial networks (GANs) are deep learning-based generative models that can generate synthetic data samples based on a real dataset’'s data distribution. They have been successfully used in clinical neuroimaging studies. In this work, we investigate how synthetic MRI images generated by GANs can improve the performance of deep learning models for accurately classifying the level of dementia (i.e., very mildly demented, mildly demented, moderately demented, and no dementia.) We trained a state-ofthe-art deep learning model for image classification, namely, the Data-Efficient Image Transformer (DeiT) using a real-world MRI dataset along with synthetic MRI images generated by GANs. We combined real and synthetic images during training by varying the proportion of synthetic images in the training set. We evaluated the accuracy and F1-score of the trained DeiT models on real MRI
images. Our results showed that DeiT can achieve good performance even with synthetic images in the training set. Hence, GANs can offer a promising solution to improving dementia diagnosis via deep learning especially when real data are scarce.
Real Images
Synthetic Image Examples

Code

The code was run on CloudLab, an experimental testbed for cloud computing research. The code was run on

数据加载示例（图像类）

from PIL import Image
import glob, os

files = (glob.glob(os.path.join(path, "**", "*.png"), recursive=True)
       + glob.glob(os.path.join(path, "**", "*.jpg"), recursive=True)
       + glob.glob(os.path.join(path, "**", "*.tif"), recursive=True))
print("图像文件数:", len(files))
img = Image.open(files[0]); print("尺寸/模式:", img.size, img.mode)

# torchvision Dataset 方式:
# from torchvision import datasets
# ds = datasets.ImageFolder(path)  # 要求 子目录=类别

目录组织与标注格式以源站说明和下载后实际文件为准。

完整仓库：github.com/AndrewRomitti/Evaluating-The-Effectiveness-of-Synthetic-Datasets-for-Dementia-Diagnosis-Using-Deep-Learning