Convolutional neural networks to predict brain tumor grades and Alzheimer's disease with MR spectroscopic imaging data.

PURPOSE: To evaluate the value of convolutional neural network (CNN) in the diagnosis of human brain tumor or Alzheimer's disease by MR spectroscopic imaging (MRSI) and to compare its Matthews correlation coefficient (MCC) score against that of other machine learning methods and previous evaluation of the same data. We address two challenges: 1) limited number of cases in MRSI datasets and 2) interpretability of results in the form of relevant spectral regions. METHODS: A shallow CNN with only one hidden layer and an ad-hoc loss function was constructed involving two branches for processing spectral and image features of a brain voxel respectively. Each branch consists of a single convolutional hidden layer. The output of the two convolutional layers is merged and fed to a classification layer that outputs class predictions for the given brain voxel. RESULTS: Our CNN method separated glioma grades 3 and 4 and identified Alzheimer's disease patients using MRSI and complementary MRI data with high MCC score (Area Under the Curve were 0.87 and 0.91 respectively). The results demonstrated superior effectiveness over other popular methods as Partial Least Squares or Support Vector Machines. Also, our method automatically identified the spectral regions most important in the diagnosis process and we show that these are in good agreement with existing biomarkers from the literature. CONCLUSION: Shallow CNNs models integrating image and spectral features improved quantitative and exploration and diagnosis of brain diseases for research and clinical purposes. Software is available at https://bitbucket.org/TeslaH2O/cnn_mrsi.

Encoder-decoder models for chest X-ray report generation perform no better than unconditioned baselines.

High quality radiology reporting of chest X-ray images is of core importance for high-quality patient diagnosis and care. Automatically generated reports can assist radiologists by reducing their workload and even may prevent errors. Machine Learning (ML) models for this task take an X-ray image as input and output a sequence of words. In this work, we show that ML models for this task based on the popular encoder-decoder approach, like 'Show, Attend and Tell' (SA&T) have similar or worse performance than models that do not use the input image, called unconditioned baseline. An unconditioned model achieved diagnostic accuracy of 0.91 on the IU chest X-ray dataset, and significantly outperformed SA&T (0.877) and other popular ML models (p-value < 0.001). This unconditioned model also outperformed SA&T and similar ML methods on the BLEU-4 and METEOR metrics. Also, an unconditioned version of SA&T obtained by permuting the reports generated from images of the test set, achieved diagnostic accuracy of 0.862, comparable to that of SA&T (p-value ≥ 0.05).

Evaluating diagnostic content of AI-generated radiology reports of chest X-rays.

Radiology reports are of core importance for the communication between the radiologist and clinician. A computer-aided radiology report system can assist radiologists in this task and reduce variation between reports thus facilitating communication with the medical doctor or clinician. Producing a well structured, clear, and clinically well-focused radiology report is essential for high-quality patient diagnosis and care. Despite recent advances in deep learning for image caption generation, this task remains highly challenging in a medical setting. Research has mainly focused on the design of tailored machine learning methods for this task, while little attention has been devoted to the development of evaluation metrics to assess the quality of AI-generated documents. Conventional quality metrics for natural language processing methods like the popular BLEU score, provide little information about the quality of the diagnostic content of AI-generated radiology reports. In particular, because radiology reports often use standardized sentences, BLEU scores of generated reports can be high while they lack diagnostically important information. We investigate this problem and propose a new measure that quantifies the diagnostic content of AI-generated radiology reports. In addition, we exploit the standardization of reports by generating a sequence of sentences. That is, instead of using a dictionary of words, as current image captioning methods do, we use a dictionary of sentences. The assumption underlying this choice is that radiologists use a well-focused vocabulary of 'standard' sentences, which should suffice for composing most reports. As a by-product, a significant training speed-up is achieved compared to models trained on a dictionary of words. Overall, results of our investigation indicate that standard validation metrics for AI-generated documents are weakly correlated with the diagnostic content of the reports. Therefore, these measures should be not used as only validation metrics, and measures evaluating diagnostic content should be preferred in such a medical context.

Generalized Convolution Spectral Mixture for Multitask Gaussian Processes.

Multitask Gaussian processes (MTGPs) are a powerful approach for modeling dependencies between multiple related tasks or functions for joint regression. Current kernels for MTGPs cannot fully model nonlinear task correlations and other types of dependencies. In this article, we address this limitation. We focus on spectral mixture (SM) kernels and propose an enhancement of this type of kernels, called multitask generalized convolution SM (MT-GCSM) kernel. The MT-GCSM kernel can model nonlinear task correlations and dependence between components, including time and phase delay dependence. Each task in MT-GCSM has its GCSM kernel with its number of convolution structures, and dependencies between all components from different tasks are considered. Another constraint of current kernels for MTGPs is that components from different tasks are aligned. Here, we lift this constraint by using inner and outer full cross convolution between a base component and the reversed complex conjugate of another base component. Extensive experiments on two synthetic and three real-life data sets illustrate the difference between MT-GCSM and previous SM kernels as well as the practical effectiveness of MT-GCSM.

Multioutput Convolution Spectral Mixture for Gaussian Processes.

Multioutput Gaussian processes (MOGPs) are an extension of Gaussian processes (GPs) for predicting multiple output variables (also called channels/tasks) simultaneously. In this article, we use the convolution theorem to design a new kernel for MOGPs by modeling cross-channel dependencies through cross convolution of time-and phase-delayed components in the spectral domain. The resulting kernel is called multioutput convolution spectral mixture (MOCSM) kernel. The results of extensive experiments on synthetic and real-life data sets demonstrate the advantages of the proposed kernel and its state-of-the-art performance. MOCSM enjoys the desirable property to reduce to the well-known spectral mixture (SM) kernel when a single channel is considered. A comparison with the recently introduced multioutput SM kernel reveals that this is not the case for the latter kernel, which contains quadratic terms that generate undesirable scale effects when the spectral densities of different channels are either very close or very far from each other in the frequency domain.

Novelty Detection using Deep Normative Modeling for IMU-Based Abnormal Movement Monitoring in Parkinson's Disease and Autism Spectrum Disorders.

Detecting and monitoring of abnormal movement behaviors in patients with Parkinson's Disease (PD) and individuals with Autism Spectrum Disorders (ASD) are beneficial for adjusting care and medical treatment in order to improve the patient's quality of life. Supervised methods commonly used in the literature need annotation of data, which is a time-consuming and costly process. In this paper, we propose deep normative modeling as a probabilistic novelty detection method, in which we model the distribution of normal human movements recorded by wearable sensors and try to detect abnormal movements in patients with PD and ASD in a novelty detection framework. In the proposed deep normative model, a movement disorder behavior is treated as an extreme of the normal range or, equivalently, as a deviation from the normal movements. Our experiments on three benchmark datasets indicate the effectiveness of the proposed method, which outperforms one-class SVM and the reconstruction-based novelty detection approaches. Our contribution opens the door toward modeling normal human movements during daily activities using wearable sensors and eventually real-time abnormal movement detection in neuro-developmental and neuro-degenerative disorders.

Generative models for local network community detection.

Local network community detection aims to find a single community in a large network, while inspecting only a small part of that network around a given seed node. This is much cheaper than finding all communities in a network. Most methods for local community detection are formulated as ad hoc optimization problems. In this paper, we instead start from a generative model for networks with a community structure. By assuming that the network is uniform, we can approximate the structure of the unobserved parts of the network to obtain a method for local community detection. We apply this local approximation technique to two variants of the stochastic block model. This results in local community detection methods based on probabilistic models. Interestingly, in the limit, one of the proposed approximations corresponds to conductance, a popular metric in this field. Experiments on real and synthetic data sets show comparable or improved results compared to state-of-the-art local community detection algorithms.

Convolutional neural networks for vibrational spectroscopic data analysis.

In this work we show that convolutional neural networks (CNNs) can be efficiently used to classify vibrational spectroscopic data and identify important spectral regions. CNNs are the current state-of-the-art in image classification and speech recognition and can learn interpretable representations of the data. These characteristics make CNNs a good candidate for reducing the need for preprocessing and for highlighting important spectral regions, both of which are crucial steps in the analysis of vibrational spectroscopic data. Chemometric analysis of vibrational spectroscopic data often relies on preprocessing methods involving baseline correction, scatter correction and noise removal, which are applied to the spectra prior to model building. Preprocessing is a critical step because even in simple problems using 'reasonable' preprocessing methods may decrease the performance of the final model. We develop a new CNN based method and provide an accompanying publicly available software. It is based on a simple CNN architecture with a single convolutional layer (a so-called shallow CNN). Our method outperforms standard classification algorithms used in chemometrics (e.g. PLS) in terms of accuracy when applied to non-preprocessed test data (86% average accuracy compared to the 62% achieved by PLS), and it achieves better performance even on preprocessed test data (96% average accuracy compared to the 89% achieved by PLS). For interpretability purposes, our method includes a procedure for finding important spectral regions, thereby facilitating qualitative interpretation of results.

Local quality functions for graph clustering with non-negative matrix factorization.

Many graph clustering quality functions suffer from a resolution limit, namely the inability to find small clusters in large graphs. So-called resolution-limit-free quality functions do not have this limit. This property was previously introduced for hard clustering, that is, graph partitioning. We investigate the resolution-limit-free property in the context of non-negative matrix factorization (NMF) for hard and soft graph clustering. To use NMF in the hard clustering setting, a common approach is to assign each node to its highest membership cluster. We show that in this case symmetric NMF is not resolution-limit free, but that it becomes so when hardness constraints are used as part of the optimization. The resulting function is strongly linked to the constant Potts model. In soft clustering, nodes can belong to more than one cluster, with varying degrees of membership. In this setting resolution-limit free turns out to be too strong a property. Therefore we introduce locality, which roughly states that changing one part of the graph does not affect the clustering of other parts of the graph. We argue that this is a desirable property, provide conditions under which NMF quality functions are local, and propose a novel class of local probabilistic NMF quality functions for soft graph clustering.

Predicting Drug-Target Interactions for New Drug Compounds Using a Weighted Nearest Neighbor Profile.

In silico discovery of interactions between drug compounds and target proteins is of core importance for improving the efficiency of the laborious and costly experimental determination of drug-target interaction. Drug-target interaction data are available for many classes of pharmaceutically useful target proteins including enzymes, ion channels, GPCRs and nuclear receptors. However, current drug-target interaction databases contain a small number of drug-target pairs which are experimentally validated interactions. In particular, for some drug compounds (or targets) there is no available interaction. This motivates the need for developing methods that predict interacting pairs with high accuracy also for these 'new' drug compounds (or targets). We show that a simple weighted nearest neighbor procedure is highly effective for this task. We integrate this procedure into a recent machine learning method for drug-target interaction we developed in previous work. Results of experiments indicate that the resulting method predicts true interactions with high accuracy also for new drug compounds and achieves results comparable or better than those of recent state-of-the-art algorithms. Software is publicly available at http://cs.ru.nl/~tvanlaarhoven/drugtarget2013/.

Graph clustering with local search optimization: the resolution bias of the objective function matters most.

Results of a recent comparative experimental assessment of methods for network community detection applied to benchmark graphs indicate that the two best methods use different objective functions but a similar local search-based optimization (LSO) procedure. This observation motivates the following research question: Given the LSO optimization procedure, how much does the choice of the objective function influence the results and in what way? We address this question empirically in a broad graph clustering context, that is, when graphs are either given as such or are k-nearest-neighbor graphs generated from a given data set. We consider normalized cut, modularity, and infomap, as well as two new objective functions. We show that all these objectives have a resolution bias, that is, they tend to prefer either small or large clusters. When removing this bias, by forcing the objective to generate a given number of clusters, LSO achieves similar performance across the considered objective functions on benchmark networks with built-in community structure. These results indicate that the resolution bias is the most important difference between objective functions in graph clustering with LSO. Spectral clustering is an alternative to LSO, which has been used to optimize the popular normalized cut and modularity objectives. We show experimentally that LSO often achieves superior performance than spectral clustering on various benchmark, real-life, and k-nearest-neighbor graphs. These results, the flexibility of LSO and its efficiency, provide arguments in favor of this optimization method.

Gaussian interaction profile kernels for predicting drug-target interaction.

MOTIVATION: The in silico prediction of potential interactions between drugs and target proteins is of core importance for the identification of new drugs or novel targets for existing drugs. However, only a tiny portion of all drug-target pairs in current datasets are experimentally validated interactions. This motivates the need for developing computational methods that predict true interaction pairs with high accuracy. RESULTS: We show that a simple machine learning method that uses the drug-target network as the only source of information is capable of predicting true interaction pairs with high accuracy. Specifically, we introduce interaction profiles of drugs (and of targets) in a network, which are binary vectors specifying the presence or absence of interaction with every target (drug) in that network. We define a kernel on these profiles, called the Gaussian Interaction Profile (GIP) kernel, and use a simple classifier, (kernel) Regularized Least Squares (RLS), for prediction drug-target interactions. We test comparatively the effectiveness of RLS with the GIP kernel on four drug-target interaction networks used in previous studies. The proposed algorithm achieves area under the precision-recall curve (AUPR) up to 92.7, significantly improving over results of state-of-the-art methods. Moreover, we show that using also kernels based on chemical and genomic information further increases accuracy, with a neat improvement on small datasets. These results substantiate the relevance of the network topology (in the form of interaction profiles) as source of information for predicting drug-target interactions. AVAILABILITY: Software and Supplementary Material are available at http://cs.ru.nl/~tvanlaarhoven/drugtarget2011/. CONTACT: tvanlaarhoven@cs.ru.nl; elenam@cs.ru.nl. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.