Generative adversarial network constrained multiple loss autoencoder: A deep learning-based individual atrophy detection for Alzheimer's disease and mild cognitive impairment.

Exploring individual brain atrophy patterns is of great value in precision medicine for Alzheimer's disease (AD) and mild cognitive impairment (MCI). However, the current individual brain atrophy detection models are deficient. Here, we proposed a framework called generative adversarial network constrained multiple loss autoencoder (GANCMLAE) for precisely depicting individual atrophy patterns. The GANCMLAE model was trained using normal controls (NCs) from the Alzheimer's Disease Neuroimaging Initiative cohort, and the Xuanwu cohort was employed to validate the robustness of the model. The potential of the model for identifying different atrophy patterns of MCI subtypes was also assessed. Furthermore, the clinical application potential of the GANCMLAE model was investigated. The results showed that the model can achieve good image reconstruction performance on the structural similarity index measure (0.929 ± 0.003), peak signal-to-noise ratio (31.04 ± 0.09), and mean squared error (0.0014 ± 0.0001) with less latent loss in the Xuanwu cohort. The individual atrophy patterns extracted from this model are more precise in reflecting the clinical symptoms of MCI subtypes. The individual atrophy patterns exhibit a better discriminative power in identifying patients with AD and MCI from NCs than those of the t-test model, with areas under the receiver operating characteristic curve of 0.867 (95%: 0.837-0.897) and 0.752 (95%: 0.71-0.790), respectively. Similar findings are also reported in the AD and MCI subgroups. In conclusion, the GANCMLAE model can serve as an effective tool for individualised atrophy detection.

Application and prospect for generative adversarial networks in cross-modality reconstruction of medical images. / 生成对抗网络在医学图像跨模态重建中的应用及展望.

Cross-modality reconstruction of medical images refers to predicting the image from one modality to another so as to achieve more accurate personalized medicine. Generative adversarial networks is the most commonly used deep learning technique in cross-modality reconstruction. It can generate realistic images by learning features from implicit distributions that follow the distributions of real data and then reconstruct the image of another modality rapidly. With the sharp increase in clinical demand for multi-modality medical image, this technology has been widely used in the task of cross modal reconstruction between different medical image modalities, such as magnetic resonance imaging, computed tomography and positron emission computed tomography. It can achieve accurate and efficient cross-modality image reconstruction in different parts of the body, such as the brain, heart, etc. In addition, although GAN has achieved some success in cross-modality reconstruction, its stability, generalization ability, and accuracy still need further research and improvement.

Glucose metabolism patterns: A potential index to characterize brain ageing and predict high conversion risk into cognitive impairment.

Exploring individual hallmarks of brain ageing is important. Here, we propose the age-related glucose metabolism pattern (ARGMP) as a potential index to characterize brain ageing in cognitively normal (CN) elderly people. We collected 18F-fluorodeoxyglucose (18F-FDG) PET brain images from two independent cohorts: the Alzheimer's Disease Neuroimaging Initiative (ADNI, N = 127) and the Xuanwu Hospital of Capital Medical University, Beijing, China (N = 84). During follow-up (mean 80.60 months), 23 participants in the ADNI cohort converted to cognitive impairment. ARGMPs were identified using the scaled subprofile model/principal component analysis method, and cross-validations were conducted in both independent cohorts. A survival analysis was further conducted to calculate the predictive effect of conversion risk by using ARGMPs. The results showed that ARGMPs were characterized by hypometabolism with increasing age primarily in the bilateral medial superior frontal gyrus, anterior cingulate and paracingulate gyri, caudate nucleus, and left supplementary motor area and hypermetabolism in part of the left inferior cerebellum. The expression network scores of ARGMPs were significantly associated with chronological age (R = 0.808, p < 0.001), which was validated in both the ADNI and Xuanwu cohorts. Individuals with higher network scores exhibited a better predictive effect (HR: 0.30, 95% CI: 0.1340 ~ 0.6904, p = 0.0068). These findings indicate that ARGMPs derived from CN participants may represent a novel index for characterizing brain ageing and predicting high conversion risk into cognitive impairment.

Unsupervised detection of individual atrophy in Alzheimer's disease.

BACKGROUND: To realize precision medicine, it is important to realize the detection of the individual atrophy of Alzheimer's disease (AD) patients. Our objective is to find individual brain regions of interest (ROIs) in AD patients via an unsupervised deep learning network. METHODS: This study used structural Magnetic Resonance Imaging (sMRI) scans with the 732 healthy control (HC) subjects and 202 AD patients from the Alzheimer's disease Neuroimaging Initiative (ADNI), and the 105 HC subjects were collected at the Xuanwu Hospital. An unsupervised deep learning network based on Adversarial Autoencoders (AAE) was proposed to delineate the individual atrophy of AD patients. In the proposed model, Variational Autoencoders (VAE) and Generative Adversarial Networks (GAN) were combined to learn the potential distribution and train a generator. In this step, the 530 HCs from ADNI were applied as the training dataset and the 105 HCs from Xuanwu Hospital were applied as an external validation dataset. The structural similarity (SSIM) was used to judge the robustness of the proposed model. Then, ROIs of the 202 AD patients were detected. In order to verify the clinical performance of these ROIs, other 202 HCs were selected from ADNI and a multilayer perceptron (MLP) was used to classify AD versus HC by 5 folder cross-validation. In the comparative experiments, we compared our model with three other previous models. RESULTS: The SSIM reached 0.86 in both training and external validation datasets. Eventually, the classification accuracy of our model achieved 0.94±0.02. In the meanwhile, the classification accuracies were 0.89±0.01, 0.85±0.04 and 0.91±0.03 for the three previous methods. CONCLUSION: Our deep learning model could detect individual atrophy in AD patients. It may be a useful tool for AD diagnosis in clinics.