PINet: Privileged Information Improve the Interpretablity and generalization of structural MRI in Alzheimer's Disease.
PINet: Privileged Information Improve the Interpretablity and generalization of structural MRI in Alzheimer's Disease.
复制标题
PINet：特权信息提高阿尔茨海默病结构 MRI 的可解释性和概括性。
DOI：
10.1145/3584371.3613000
复制
发表时间：
2023
期刊：
影响因子：
--
通讯作者：
Yang,Baijian
中科院分区：
文献类型：
--
作者：
Tang,Zijia;Zhang,Tonglin;Song,Qianqian;Su,Jing;Yang,Baijian
阿尔茨海默病导致的不可逆转的进行性萎缩导致思维和行为技能的不断下降。到目前为止，CNN分类器被广泛应用于AD及其相关异常结构的早期诊断。然而，大多数现有的黑盒CNN分类器严重依赖有限的MRI扫描，并且很少使用以前临床发现的领域知识。在本研究中，我们提出了一个称为asPINet的框架，将先前的领域知识视为特权信息(PI)，并在预测过程中打开黑盒。输入领域知识指导神经网络学习具有代表性的特征，并引入可理解性进行进一步分析；PI网络使用变形金刚类融合模块PIF(Privileged Information Fusion)迭代计算图像特征与PI特征之间的相关性，并将特征投影到潜在空间进行分类。金字塔特征可视化(PFV)模块作为验证，突出输入图像上的重要特征。PInet适用于神经成像任务，我们使用ADNI数据集的结构MRI扫描演示了其在阿尔茨海默病中的应用。在实验过程中，我们使用异常的大脑结构如海马体作为PI，用1.5T扫描仪的数据训练模型，并用3T扫描仪的数据进行测试。F1-Score显示，PINet在转移到新的数据集方面更稳健，下降了大约2%(从0.9471到0.9231)，而基线CNN方法下降了29%(从0.8679到0.6154)。PINet的性能依赖于对领域知识的选择，我们的最佳模型是在12个ROI的指导下训练的，主要是临时LOB和枕部LOB的结构。综上所述，PINET将领域知识作为PI来训练CNN模型，所选择的PI在黑盒CNN分类器中引入了可解释性和泛化能力。
The irreversible and progressive atrophy by Alzheimer's Disease resulted in continuous decline in thinking and behavioral skills. To date, CNN classifiers were widely applied to assist the early diagnosis of AD and its associated abnormal structures. However, most existing black-box CNN classifiers relied heavily on the limited MRI scans, and used little domain knowledge from the previous clinical findings. In this study, we proposed a framework, named asPINet, to consider the previous domain knowledge as aPrivileged Information (PI), and open the black-box in the prediction process. The input domain knowledge guides the neural network to learn representative features and introduced intepretability for further analysis.PINetused a Transformer-like fusion modulePrivileged Information Fusion (PIF)to iteratively calculate the correlation of the features between image features and PI features, and project the features into a latent space for classification. ThePyramid Feature Visualization (PFV)module served as a verification to highlight the significant features on the input images.PINetwas suitable for neuro-imaging tasks and we demonstrated its application in Alzheimer's Disease using structural MRI scans from ADNI dataset. During the experiments, we employed the abnormal brain structures such as the Hippocampus as thePI, trained the model with the data from 1.5T scanners and tested from 3T scanners. The F1-score showed thatPINetwas more robust in transferring to a new dataset, with approximatedly 2% drop (from 0.9471 to 0.9231), while the baseline CNN methods had a 29% drop (from 0.8679 to 0.6154). The performance ofPINetwas relied on the selection of the domain knowledge as thePI.Our best model was trained under the guidance of 12 selected ROIs, major in the structures ofTemporal LobeandOccipital Lobe.In summary, PINet considered the domain knowledge as thePIto train the CNN model, and the selectedPIintroduced both interpretability and generalization ability to the black box CNN classifiers.