Deep Learning Representation from Electroencephalography of Early-Stage Creutzfeldt-Jakob Disease and Features for Differentiation from Rapidly Progressive Dementia

Abstract

A novel technique of quantitative EEG for differentiating patients with early-stage Creutzfeldt-Jakob disease (CJD) from other forms of rapidly progressive dementia (RPD) is proposed. The discrimination is based on the extraction of suitable features from the time-frequency representation of the EEG signals through continuous wavelet transform (CWT). An average measure of complexity of the EEG signal obtained by permutation entropy (PE) is also included. The dimensionality of the feature space is reduced through a multilayer processing system based on the recently emerged deep learning (DL) concept. The DL processor includes a stacked auto-encoder, trained by unsupervised learning techniques, and a classifier whose parameters are determined in a supervised way by associating the known category labels to the reduced vector of high-level features generated by the previous processing blocks. The supervised learning step is carried out by using either support vector machines (SVM) or multilayer neural networks (MLP-NN). A subset of EEG from patients suffering from Alzheimer's Disease (AD) and healthy controls (HC) is considered for differentiating CJD patients. When fine-tuning the parameters of the global processing system by a supervised learning procedure, the proposed system is able to achieve an average accuracy of 89%, an average sensitivity of 92%, and an average specificity of 89% in differentiating CJD from RPD. Similar results are obtained for CJD versus AD and CJD versus HC.

Higher Sensitivity and Reproducibility of Wavelet-Based Amplitude of Resting-State fMRI

“…2 http://www.fil.ion.ucl.ac.uk/spm/software/spm8/ where x(t) denotes the time series, ψ k,s (t) denotes the mother wavelet function, s denotes wavelet scale (64 frequency bins in the current study, between 0 and 0.25 Hz at an interval of 0.0039 Hz), k denotes the localized time index (k ∈ [1, 170] and [1,230] for EOEC dataset and ADHD-200 dataset, respectively), and * denotes the complex conjugate (Poza et al, 2014;Morabito et al, 2017). We used five mother wavelets which have been used in previous fMRI literature, including db2 (Bullmore et al, 2004;Salomon et al, 2011;Zhang et al, 2016), bior4.4 (Laine, 2000;Van De Ville et al, 2003;Mutihac, 2006), morl (Chang and Glover, 2010;Bajaj et al, 2013;Omidvarnia et al, 2017;Yaesoubi et al, 2017), meyr (Behjat et al, 2015), and sym3 (Khullar et al, 2011).…”

“…As mentioned in a previous study, "a relatively high value of the coefficient is given in the product with the wavelet if there exists a spectral component of the signal corresponding to the value of s at a location k" (Morabito et al, 2017). Wavelet-ALFF was calculated by first adding up the wavelet coefficients at all time points for each frequency point, and the averaged coefficient across a given frequency band was then obtained as defined below:…”

Control methodologies for vibration control of smart civil and mechanical structures

“…Desirable characteristics of an ideal controller are simplicity, robustness, fault tolerance, providing an optimal solution, and implementation practicality. New research trends may include the following: A multi‐paradigm control strategy through adroit integration of CI techniques, advanced signal processing techniques such as EMD and synchrosqueezed WT (Li et al, ), and Bayesian probabilistic concepts (Chen, Jahanshahi, Wu, & Joffe, ; Kosgodagan et al, ). For large smart civil and mechanical structures, robust decentralized control algorithms are a promising control strategy. Application of machine learning techniques (Fernandez, Carmona, del Jesus, & Herrera, ; Guo, Wang, Cabrerizo, & Adjouadi, ; Palomo & Lopez‐Rubio, ) such as deep neural networks (Ortega‐Zamorano, Jerez, Gómez, & Franco, ; Koziarski & Cyganek, ; Morabito et al, ) learning for development of adaptive control algorithms with learning capability. Design for sustainability where the structural system performance is optimized and the controller energy consumption is minimized using multiobjective and many‐objective optimization algorithms. Another extension of the research is to unite the two fields of structural health monitoring (Cha, Choi, & Buyukozturk, ; Lin, Nie, & Ma, ; Tsogka, Daskalakis, Comanducci, & Ubertini, ) and vibration control where structural health is monitored based on measurements from sensors and control devices apply appropriate forces determined by an adaptive/intelligent control algorithm to minimize the structural response. …”

“…3. Application of machine learning techniques (Fernandez, Carmona, del Jesus, & Herrera, 2017;Guo, Wang, Cabrerizo, & Adjouadi, 2017;Palomo & Lopez-Rubio, 2016) such as deep neural networks (Ortega-Zamorano, Jerez, Gómez, & Franco, 2017;Koziarski & Cyganek, 2017;Morabito et al, 2017) learning for development of adaptive control algorithms with learning capability.…”

Drug-Drug Interaction Extraction via Recurrent Hybrid Convolutional Neural Networks with an Improved Focal Loss

“…The Convolutional Neural Network (CNN) is well-known deep learning approach. Morabito et al [16] employed CNNs to detect early-stage Creutzfeldt-Jakob disease (CJD) from other forms of rapidly progressive dementia (RPD) and achieved outstanding experimental results. Acharya et al [17] presented a novel computer model for EEG-based screening of depression using CNNs, which extended the diagnosis of different stages and levels of severity of depression.…”