Abstract
18F-fluorodeoxyglucose (FDG)-positron emission tomography (PET) reveals altered brain metabolism in individuals with mild cognitive impairment (MCI) and Alzheimer’s disease (AD). Some biomarkers derived from FDG-PET by computer-aided-diagnosis (CAD) technologies have been proved that they can accurately diagnosis normal control (NC), MCI, and AD. However, existing FDG-PET-based researches are still insufficient for the identification of early MCI (EMCI) and late MCI (LMCI). Compared with methods based other modalities, current methods with FDG-PET are also inadequate in using the inter-region-based features for the diagnosis of early AD. Moreover, considering the variability in different individuals, some hard samples which are very similar with both two classes limit the classification performance. To tackle these problems, in this paper, we propose a novel bilinear pooling and metric learning network (BMNet), which can extract the inter-region representation features and distinguish hard samples by constructing the embedding space. To validate the proposed method, we collect 898 FDG-PET images from Alzheimer’s disease neuroimaging initiative (ADNI) including 263 normal control (NC) patients, 290 EMCI patients, 147 LMCI patients, and 198 AD patients. Following the common preprocessing steps, 90 features are extracted from each FDG-PET image according to the automatic anatomical landmark (AAL) template and then sent into the proposed network. Extensive fivefold cross-validation experiments are performed for multiple two-class classifications. Experiments show that most metrics are improved after adding the bilinear pooling module and metric losses to the Baseline model respectively. Specifically, in the classification task between EMCI and LMCI, the specificity improves 6.38% after adding the triple metric loss, and the negative predictive value (NPV) improves 3.45% after using the bilinear pooling module. In addition, the accuracy of classification between EMCI and LMCI achieves 79.64% using imbalanced FDG-PET images, which illustrates that the proposed method yields a state-of-the-art result of the classification accuracy between EMCI and LMCI based on PET images.
Highlights
18F-fluorodeoxyglucose (FDG)-positron emission tomography (PET) can reveal altered brain metabolism in individuals with mild cognitive impairment (MCI) and Alzheimer’s disease (AD) (Sörensen et al, 2019; Zhou et al, 2019; Wang et al, 2020)
The bigger dataset might increase the variety of individuals and the probability of special samples which are hard to distinguish, causing complexity of the problem for classification tasks. Considering these two limitations, we propose a novel bilinear pooling and metric learning network (BMNet) for early Alzheimer’s disease identification with FDG-PET images, especially for the classification task between early MCI (EMCI) and late MCI (LMCI)
Our main contributions are as follows: (1) We propose a shallow convolutional neural network model to achieve the classification; (2) We introduce a bilinear pooling module into the model for exploring the inter-region representation features in the whole brain; (3) We introduce the deep metric learning to help model learn the hard samples in the embedding feature space; (4) We conduct our method on the dataset collected from the publicly released Alzheimer’s disease neuroimaging initiative (ADNI) database and obtain a state-of-the-art result of the classification between EMCI and LMCI based on PET images
Summary
18F-fluorodeoxyglucose (FDG)-positron emission tomography (PET) can reveal altered brain metabolism in individuals with MCI and AD (Sörensen et al, 2019; Zhou et al, 2019; Wang et al, 2020). Liu et al (2018) proposed a new classification framework for AD diagnosis with 3D PET images. They decomposed 3D images into 2D slices to learn the intra-slice and inter-slice features and achieved a promising classification performance of AUC of 83.9% for MCI vs NC classification. Zhou et al (2021) developed a new deep belief network model for AD diagnosis based on sparse-response theory, which identified a better classification result than that of other models. Pan et al (2021) developed a disease-imagespecific deep learning (DSDL) framework which can achieve neuroimage synthesis and disease diagnosis simultaneously using incomplete multi-modality neuroimages
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
