Heung Il Suk

Deep Learning in Medical Image Analysis

This review covers computer-assisted analysis of images in the field of medical imaging. Recent advances in machine learning, especially with regard to deep learning, are helping to identify, classify, and quantify patterns in medical images. At the core of these advances is the ability to exploit hierarchical feature representations learned solely from data, instead of features designed by hand according to domain-specific knowledge. Deep learning is rapidly becoming the state of the art, leading to enhanced performance in various medical applications. We introduce the fundamentals of deep learning methods and review their successes in image registration, detection of anatomical and cellular structures, tissue segmentation, computer-aided disease diagnosis and prognosis, and so on. We conclude by discussing research issues and suggesting future directions for further improvement.

A Novel Bayesian Framework for Discriminative Feature Extraction in Brain-Computer Interfaces

As there has been a paradigm shift in the learning load from a human subject to a computer, machine learning has been considered as a useful tool for Brain-Computer Interfaces (BCIs). In this paper, we propose a novel Bayesian framework for discriminative feature extraction for motor imagery classification in an EEG-based BCI in which the class-discriminative frequency bands and the corresponding spatial filters are optimized by means of the probabilistic and information-theoretic approaches. In our framework, the problem of simultaneous spatiospectral filter optimization is formulated as the estimation of an unknown posterior probability density function (pdf) that represents the probability that a single-trial EEG of predefined mental tasks can be discriminated in a state. In order to estimate the posterior pdf, we propose a particle-based approximation method by extending a factored-sampling technique with a diffusion process. An information-theoretic observation model is also devised to measure discriminative power of features between classes. From the viewpoint of classifier design, the proposed method naturally allows us to construct a spectrally weighted label decision rule by linearly combining the outputs from multiple classifiers. We demonstrate the feasibility and effectiveness of the proposed method by analyzing the results and its success on three public databases.

A novel matrix-similarity based loss function for joint regression and classification in AD diagnosis

Recent studies on AD/MCI diagnosis have shown that the tasks of identifying brain disease and predicting clinical scores are highly related to each other. Furthermore, it has been shown that feature selection with a manifold learning or a sparse model can handle the problems of high feature dimensionality and small sample size. However, the tasks of clinical score regression and clinical label classification were often conducted separately in the previous studies. Regarding the feature selection, to our best knowledge, most of the previous work considered a loss function defined as an element-wise difference between the target values and the predicted ones. In this paper, we consider the problems of joint regression and classification for AD/MCI diagnosis and propose a novel matrix-similarity based loss function that uses high-level information inherent in the target response matrix and imposes the information to be preserved in the predicted response matrix. The newly devised loss function is combined with a group lasso method for joint feature selection across tasks, i.e., predictions of clinical scores and a class label. In order to validate the effectiveness of the proposed method, we conducted experiments on the Alzheimers Disease Neuroimaging Initiative (ADNI) dataset, and showed that the newly devised loss function helped enhance the performances of both clinical score prediction and disease status identification, outperforming the state-of-the-art methods.

Deep learning based imaging data completion for improved brain disease diagnosis.

Combining multi-modality brain data for disease diagnosis commonly leads to improved performance. A challenge in using multimodality data is that the data are commonly incomplete; namely, some modality might be missing for some subjects. In this work, we proposed a deep learning based framework for estimating multi-modality imaging data. Our method takes the form of convolutional neural networks, where the input and output are two volumetric modalities. The network contains a large number of trainable parameters that capture the relationship between input and output modalities. When trained on subjects with all modalities, the network can estimate the output modality given the input modality. We evaluated our method on the Alzheimers Disease Neuroimaging Initiative (ADNI) database, where the input and output modalities are MRI and PET images, respectively. Results showed that our method significantly outperformed prior methods.

Deep Learning-Based Feature Representation for AD/MCI Classification

In recent years, there has been a great interest in computer-aided diagnosis of Alzheimers Disease (AD) and its prodromal stage, Mild Cognitive Impairment (MCI). Unlike the previous methods that consider simple low-level features such as gray matter tissue volumes from MRI, mean signal intensities from PET, in this paper, we propose a deep learning-based feature representation with a stacked auto-encoder. We believe that there exist latent complicated patterns, e.g., non-linear relations, inherent in the low-level features. Combining latent information with the original low-level features helps build a robust model for AD/MCI classification with high diagnostic accuracy. Using the ADNI dataset, we conducted experiments showing that the proposed method is 95.9%, 85.0%, and 75.8% accurate for AD, MCI, and MCI-converter diagnosis, respectively.

Subspace Regularized Sparse Multitask Learning for Multiclass Neurodegenerative Disease Identification

The high feature-dimension and low sample-size problem is one of the major challenges in the study of computer-aided Alzheimers disease (AD) diagnosis. To circumvent this problem, feature selection and subspace learning have been playing core roles in the literature. Generally, feature selection methods are preferable in clinical applications due to their ease for interpretation, but subspace learning methods can usually achieve more promising results. In this paper, we combine two different methodological approaches to discriminative feature selection in a unified framework. Specifically, we utilize two subspace learning methods, namely, linear discriminant analysis and locality preserving projection, which have proven their effectiveness in a variety of fields, to select class-discriminative and noise-resistant features. Unlike previous methods in neuroimaging studies that mostly focused on a binary classification, the proposed feature selection method is further applicable for multiclass classification in AD diagnosis. Extensive experiments on the Alzheimers disease neuroimaging initiative dataset showed the effectiveness of the proposed method over other state-of-the-art methods.

A novel relational regularization feature selection method for joint regression and classification in AD diagnosis

HIGHLIGHTSA novel graph feature selection method for the AD/MCI diagnosis.A novel regularization exploiting the relational information inherent in the observations.First work considering three relationships for joint classification and regression.High accuracy of 95.7% for AD classification and 79.9% for MCI classification. ABSTRACT In this paper, we focus on joint regression and classification for Alzheimers disease diagnosis and propose a new feature selection method by embedding the relational information inherent in the observations into a sparse multi‐task learning framework. Specifically, the relational information includes three kinds of relationships (such as feature‐feature relation, response–response relation, and sample‐sample relation), for preserving three kinds of the similarity, such as for the features, the response variables, and the samples, respectively. To conduct feature selection, we first formulate the objective function by imposing these three relational characteristics along with an ∂2,1‐norm regularization term, and further propose a computationally efficient algorithm to optimize the proposed objective function. With the dimension‐reduced data, we train two support vector regression models to predict the clinical scores of ADAS‐Cog and MMSE, respectively, and also a support vector classification model to determine the clinical label. We conducted extensive experiments on the Alzheimers Disease Neuroimaging Initiative (ADNI) dataset to validate the effectiveness of the proposed method. Our experimental results showed the efficacy of the proposed method in enhancing the performances of both clinical scores prediction and disease status identification, compared to the state‐of‐the‐art methods.

Person authentication from neural activity of face-specific visual self-representation

In this paper, we propose a new biometric system based on the neurophysiological features of face-specific visual self representation in a human brain, which can be measured by ElectroEncephaloGraphy (EEG). First, we devise a novel stimulus presentation paradigm, using self-face and non-self-face images as stimuli for a person authentication system that can validate a persons identity by comparing the observed trait with those stored in the database (one-to-one matching). Unlike previous methods that considered the brain activities of the resting state, motor imagery, or visual evoked potentials, there are evidences that the proposed paradigm generates unique subject-specific brain-wave patterns in response to self- and non-self-face images from psychology and neurophysiology studies. Second, we devise a method for adaptive selection of EEG channels and time intervals for each subject in a discriminative manner. This makes the system immune to forgery since the selected EEG channels and time intervals for a client may not be consistent with those of imposters in terms of the latency and amplitude of the brain-waves. Based on our experimental results and analysis, it is believed that the proposed person authentication system can be considered as a new biometric authentication system.

State-space model with deep learning for functional dynamics estimation in resting-state fMRI

Studies on resting-state functional Magnetic Resonance Imaging (rs-fMRI) have shown that different brain regions still actively interact with each other while a subject is at rest, and such functional interaction is not stationary but changes over time. In terms of a large-scale brain network, in this paper, we focus on time-varying patterns of functional networks, i.e., functional dynamics, inherent in rs-fMRI, which is one of the emerging issues along with the network modelling. Specifically, we propose a novel methodological architecture that combines deep learning and state-space modelling, and apply it to rs-fMRI based Mild Cognitive Impairment (MCI) diagnosis. We first devise a Deep Auto-Encoder (DAE) to discover hierarchical non-linear functional relations among regions, by which we transform the regional features into an embedding space, whose bases are complex functional networks. Given the embedded functional features, we then use a Hidden Markov Model (HMM) to estimate dynamic characteristics of functional networks inherent in rs-fMRI via internal states, which are unobservable but can be inferred from observations statistically. By building a generative model with an HMM, we estimate the likelihood of the input features of rs-fMRI as belonging to the corresponding status, i.e., MCI or normal healthy control, based on which we identify the clinical label of a testing subject. In order to validate the effectiveness of the proposed method, we performed experiments on two different datasets and compared with state-of-the-art methods in the literature. We also analyzed the functional networks learned by DAE, estimated the functional connectivities by decoding hidden states in HMM, and investigated the estimated functional connectivities by means of a graph-theoretic approach.

Matrix-Similarity Based Loss Function and Feature Selection for Alzheimer's Disease Diagnosis

Recent studies on Alzheimers Disease (AD) or its prodromal stage, Mild Cognitive Impairment (MCI), diagnosis presented that the tasks of identifying brain disease status and predicting clinical scores based on neuroimaging features were highly related to each other. However, these tasks were often conducted independently in the previous studies. Regarding the feature selection, to our best knowledge, most of the previous work considered a loss function defined as an element-wise difference between the target values and the predicted ones. In this paper, we consider the problems of joint regression and classification for AD/MCI diagnosis and propose a novel matrix-similarity based loss function that uses high-level information inherent in the target response matrix and imposes the information to be preserved in the predicted response matrix. The newly devised loss function is combined with a group lasso method for joint feature selection across tasks, i.e., clinical scores prediction and disease status identification. We conducted experiments on the Alzheimers Disease Neuroimaging Initiative (ADNI) dataset, and showed that the newly devised loss function was effective to enhance the performances of both clinical score prediction and disease status identification, outperforming the state-of-the-art methods.