计算机科学
机器学习
正规化(语言学)
人工智能
神经影像学
一致性(知识库)
深度学习
监督学习
疾病
人工神经网络
阿尔茨海默病
阿尔茨海默病神经影像学倡议
医学
病理
精神科
作者
Xiaobo Zhang,Zhimin Li,Qian Zhang,Zegang Yin,Zhijie Lu,Yang Li
标识
DOI:10.1016/j.compbiomed.2023.107079
摘要
Alzheimer's disease (AD) is a chronic neurodegenerative disease that mainly affects older adults, causing memory loss and decline in thinking skills. In recent years, many traditional machine learning and deep learning methods have been used to assist in the diagnosis of AD, and most existing methods focus on early prediction of disease on a supervised basis. In reality, there is a massive amount of medical data available. However, some of those data have problems with the low-quality or lack of labels, and the cost of labeling them will be too high. To solve above problem, a new Weakly Supervised Deep Learning model (WSDL) is proposed, which adds attention mechanisms and consistency regularization to the EfficientNet framework and uses data augmentation techniques on the original data that can take full advantage of this unlabeled data. Validation of the proposed WSDL method on the brain MRI datasets of the Alzheimer's Disease Neuroimaging Program by setting five different unlabeled ratios to complete weakly supervised training showed better performance according to the compared experimental results with others baselines.
科研通智能强力驱动
Strongly Powered by AbleSci AI