Alzheimer_s Dataset

Alzheimer’s disease, a devastating neurodegenerative disorder, has no cure, and therefore demands early diagnosis to prevent its progression. However, limited and imbalanced MRI datasets have hindered progress in AI-driven solutions. This dataset is the result of pioneering research designed to solve two major challenges in Alzheimer’s Disease (AD) research: data scarcity and severe class imbalance. By harnessing Wasserstein GANs with Gradient Penalty (WGAN-GP), we generated high-fidelity DeepFake "axial" MRI scans that match real-world data quality while ensuring balanced representation across 4 distinct disease stages. Due to this high quality and balanced nature it resulted in in an overall increase of 11.77% in Balanced Accuracy, 15% increase in Matthew's Correlation Coefficient (MCC). The study also found a 91.4% improvement in the performance on minority classes. If you're interested in training your own GANs using our code, you can understand how its done in the published research, available here:https://docs.google.com/viewer?url=https://www.ijert.org/download/thesis/2023/IJERTTH0025.pdf阿尔茨海默病（Alzheimer’s disease, AD）是一种毁灭性神经退行性疾病，目前尚无治愈方案，因此亟需通过早期诊断阻断病情进展。然而，有限且类别失衡的磁共振成像（Magnetic Resonance Imaging, MRI）数据集，掣肘了人工智能驱动的AD诊疗方案研发进程。本数据集是为解决阿尔茨海默病研究领域两大核心挑战——数据稀缺与严重类别不平衡——而开展的开创性研究的成果。研究团队借助带梯度惩罚的沃瑟斯坦生成对抗网络（Wasserstein GANs with Gradient Penalty, WGAN-GP），生成了高保真的轴位深度伪造（DeepFake）磁共振成像扫描图像，此类图像既贴合真实临床数据的质量水准，又确保了4种不同疾病阶段的样本分布均衡。得益于该数据集的高质量与类别均衡特性，模型的平衡准确率提升11.77%，马修斯相关系数（Matthew's Correlation Coefficient, MCC）提升15%，少数类别的模型性能提升幅度达91.4%。若您希望基于本研究提供的代码自行训练生成对抗网络（Generative Adversarial Networks, GANs），可参阅已发表的研究论文以了解具体实现细节，论文链接为：https://docs.google.com/viewer?url=https://www.ijert.org/download/thesis/2023/IJERTTH0025.pdf