A Hybrid of Deep Network and Hidden Markov Model for MCI Identification with Resting-State fMRI
In this paper, we propose a novel method for modelling functional dynamics in resting-state fMRI (rs-fMRI) for Mild Cognitive Impairment (MCI) identification. Specifically, we devise a hybrid architecture by combining Deep Auto-Encoder (DAE) and Hidden Ma
- PDF / 709,421 Bytes
- 8 Pages / 439.363 x 666.131 pts Page_size
- 80 Downloads / 191 Views
Department of Brain and Cognitive Engineering, Korea University, Republic of Korea Biomedical Research Imaging Center, University of North Carolina at Chapel Hill, USA [email protected]
Abstract. In this paper, we propose a novel method for modelling functional dynamics in resting-state fMRI (rs-fMRI) for Mild Cognitive Impairment (MCI) identification. Specifically, we devise a hybrid architecture by combining Deep Auto-Encoder (DAE) and Hidden Markov Model (HMM). The roles of DAE and HMM are, respectively, to discover hierarchical non-linear relations among features, by which we transform the original features into a lower dimension space, and to model dynamic characteristics inherent in rs-fMRI, i.e., internal state changes. By building a generative model with HMMs for each class individually, we estimate the data likelihood of a test subject as MCI or normal healthy control, based on which we identify the clinical label. In our experiments, we achieved the maximal accuracy of 81.08% with the proposed method, outperforming state-of-the-art methods in the literature.
1 Introduction Motivated by Biswal et al.’s study [1] that discovered different brain regions still actively interact while a subject lies at rest, i.e., not performing any cognitive task, restingstate fMRI (rs-fMRI) has been widely used as one of the major tools for investigation of brain networks. It provides insights to explore the brain’s functional organization and examine the altered functional networks possibly due to brain disorders such as Mild Cognitive Impairment (MCI). In this regard, functional connectivity analysis has played core roles for brain disease diagnosis or prognosis [4, 7, 11, 12, 15, 16]. While many existing methods for MCI diagnosis with rs-fMRI typically assumed stationarity on the functional networks over time [12, 16], recent studies in neuroscience have shown that the functional organization of a brain is dynamic rather than static, changing spontaneously over time [9]. Eavani et al. proposed to jointly model sparse dictionary learning within a state-space model framework [2]. Leonardi et al. devised a method to reveal hidden patterns of coherent functional connectivity dynamics based on principal component analysis [11]. In this paper, we propose a novel method that discovers non-linear relations among brain regions in a hierarchical manner and explicitly models the dynamic characteristics inherent in rs-fMRI. It is noteworthy that rather than computing correlation matrices and extracting graph-theoretic features [14] such as small-worldness and clustering coefficients as commonly performed in the literature,
Corresponding author.
c Springer International Publishing Switzerland 2015 N. Navab et al. (Eds.): MICCAI 2015, Part I, LNCS 9349, pp. 573–580, 2015. DOI: 10.1007/978-3-319-24553-9_70
574
H.-I. Suk, S.-W. Lee, and D. Shen
we directly model functional dynamics from regional mean time series of rs-fMRI. In a testing phase, our model estimates the data likelihood of a test subject as MCI and Normal healthy
Data Loading...
