A pyramidal neural network for visual pattern recognition

Optimizing clinico-genomic disease prediction across ancestries: a machine learning strategy with Pareto improvement

BACKGROUND

Accurate prediction of an individual's predisposition to diseases is vital for preventive medicine and early intervention. Various statistical and machine learning models have been developed for disease prediction using clinico-genomic data. However, the accuracy of clinico-genomic prediction of diseases may vary significantly across ancestry groups due to their unequal representation in clinical genomic datasets.

METHODS

We introduced a deep transfer learning approach to improve the performance of clinico-genomic prediction models for data-disadvantaged ancestry groups. We conducted machine learning experiments on multi-ancestral genomic datasets of lung cancer, prostate cancer, and Alzheimer's disease, as well as on synthetic datasets with built-in data inequality and distribution shifts across ancestry groups.

RESULTS

Deep transfer learning significantly improved disease prediction accuracy for data-disadvantaged populations in our multi-ancestral machine learning experiments. In contrast, transfer learning based on linear frameworks did not achieve comparable improvements for these data-disadvantaged populations.

CONCLUSIONS

This study shows that deep transfer learning can enhance fairness in multi-ancestral machine learning by improving prediction accuracy for data-disadvantaged populations without compromising prediction accuracy for other populations, thus providing a Pareto improvement towards equitable clinico-genomic prediction of diseases.

Reducing healthcare disparities using multiple multiethnic data distributions with fine-tuning of transfer learning

Healthcare disparities in multiethnic medical data is a major challenge; the main reason lies in the unequal data distribution of ethnic groups among data cohorts. Biomedical data collected from different cancer genome research projects may consist of mainly one ethnic group, such as people with European ancestry. In contrast, the data distribution of other ethnic races such as African, Asian, Hispanic, and Native Americans can be less visible than the counterpart. Data inequality in the biomedical field is an important research problem, resulting in the diverse performance of machine learning models while creating healthcare disparities. Previous researches have reduced the healthcare disparities only using limited data distributions. In our study, we work on fine-tuning of deep learning and transfer learning models with different multiethnic data distributions for the prognosis of 33 cancer types. In previous studies, to reduce the healthcare disparities, only a single ethnic cohort was used as the target domain with one major source domain. In contrast, we focused on multiple ethnic cohorts as the target domain in transfer learning using the TCGA and MMRF CoMMpass study datasets. After performance comparison for experiments with new data distributions, our proposed model shows promising performance for transfer learning schemes compared to the baseline approach for old and new data distributation experiments.

Structural and functional connectivity mapping of the human corpus callosum organization with white-matter functional networks

The corpus callosum serves as a crucial organization for understanding the information integration between the two hemispheres. Our previous study explored the functional connectivity between the corpus callosum and white-matter functional networks (WM-FNs), but the corresponding physical connectivity remains unknown. The current study uses the resting-state fMRI of Human Connectome Project data to identify ten WM-FNs in 108 healthy subjects, and then independently maps the structural and functional connectivity between the corpus callosum and above WM-FNs using the diffusion tensor images (DTI) tractography and resting-state functional connectivity (RSFC). Our results demonstrated that the structural and functional connectivity between the human corpus callosum and WM-FNs have the following high overall correspondence: orbitofrontal WM-FN, DTI map = 89% and RSFC map = 92%; sensorimotor middle WM-FN, DTI map = 47% and RSFC map = 77%; deep WM-FN, DTI map = 50% and RSFC map = 79%; posterior corona radiata WM-FN, DTI map = 82% and RSFC map = 73%. These findings reinforce the notion that the corpus callosum has unique spatial distribution patterns connecting to distinct WM-FNs. However, important differences between the structural and functional connectivity mapping results were also observed, which demonstrated a synergy between DTI tractography and RSFC toward better understanding the information integration of primary and higher-order functional systems in the human brain.

Deep transfer learning for reducing health care disparities arising from biomedical data inequality

As artificial intelligence (AI) is increasingly applied to biomedical research and clinical decisions, developing unbiased AI models that work equally well for all ethnic groups is of crucial importance to health disparity prevention and reduction. However, the biomedical data inequality between different ethnic groups is set to generate new health care disparities through data-driven, algorithm-based biomedical research and clinical decisions. Using an extensive set of machine learning experiments on cancer omics data, we find that current prevalent schemes of multiethnic machine learning are prone to generating significant model performance disparities between ethnic groups. We show that these performance disparities are caused by data inequality and data distribution discrepancies between ethnic groups. We also find that transfer learning can improve machine learning model performance for data-disadvantaged ethnic groups, and thus provides an effective approach to reduce health care disparities arising from data inequality among ethnic groups.

Densely Distilled Flow-Based Knowledge Transfer in Teacher-Student Framework for Image Classification

We propose a new teacherstudent framework (TSF)-based knowledge transfer method, in which knowledge in the form of dense flow across layers is distilled from a pre-trained "teacher" deep neural network (DNN) and transferred to another "student" DNN. In the case of distilled knowledge, multiple overlapped flow-based items of information from the pre-trained teacher DNN are densely extracted across layers. Transference of the densely extracted teacher information is then achieved in the TSF using repetitive sequential training from bottom to top between the teacher and student DNN models. In other words, to efficiently transmit extracted useful teacher information to the student DNN, we perform bottom-up step-by-step transfer of densely distilled knowledge. The performance of the proposed method in terms of image classification accuracy and fast optimization is compared with those of existing TSF-based knowledge transfer methods for application to reliable image datasets, including CIFAR-10, CIFAR-100, MNIST, and SVHN. When the dense flow-based sequential knowledge transfer scheme is employed in the TSF, the trained student ResNet more accurately reflects the rich information of the pre-trained teacher ResNet and exhibits superior accuracy to the existing TSF-based knowledge transfer methods for all benchmark datasets considered in this study.

A Novel Neural Network for Remote Sensing Image Matching

Rapid development of remote sensing (RS) imaging technology makes the acquired images have larger size, higher resolution, and more complex structure, which goes beyond the reach of classical hand-crafted feature-based matching. In this paper, we propose a feature learning approach based on two-branch networks to transform the image matching task into a two-class classification problem. To match two key points, two image patches centered at the key points are entered into the proposed network. The network aims to learn discriminative feature representations for patch matching, so that more matching pairs can be obtained on the premise of maintaining higher subpixel matching accuracy. The proposed network adopts a two-stage training mode to deal with the complex characteristics of RS images. An adaptive sample selection strategy is proposed to determine the size of each patch by the scale of its central key point. Thus, each patch can preserve the texture structure around its key point rather than all patches have a predetermined size. In the matching prediction stage, two strategies, namely, superpixel-based sample graded strategy and superpixel-based ordered spatial matching, are designed to improve the matching efficiency and matching accuracy, respectively. The experimental results and theoretical analysis demonstrate the feasibility, robustness, and effectiveness of the proposed method.

Optimization of CNN through Novel Training Strategy for Visual Classification Problems

The convolution neural network (CNN) has achieved state-of-the-art performance in many computer vision applications e.g., classification, recognition, detection, etc. However, the global optimization of CNN training is still a problem. Fast classification and training play a key role in the development of the CNN. We hypothesize that the smoother and optimized the training of a CNN goes, the more efficient the end result becomes. Therefore, in this paper, we implement a modified resilient backpropagation (MRPROP) algorithm to improve the convergence and efficiency of CNN training. Particularly, a tolerant band is introduced to avoid network overtraining, which is incorporated with the global best concept for weight updating criteria to allow the training algorithm of the CNN to optimize its weights more swiftly and precisely. For comparison, we present and analyze four different training algorithms for CNN along with MRPROP, i.e., resilient backpropagation (RPROP), Levenberg–Marquardt (LM), conjugate gradient (CG), and gradient descent with momentum (GDM). Experimental results showcase the merit of the proposed approach on a public face and skin dataset.

Structured Pyramidal Neural Networks

The Pyramidal Neural Networks (PNN) are an example of a successful recently proposed model inspired by the human visual system and deep learning theory. PNNs are applied to computer vision and based on the concept of receptive fields. This paper proposes a variation of PNN, named here as Structured Pyramidal Neural Network (SPNN). SPNN has self-adaptive variable receptive fields, while the original PNNs rely on the same size for the fields of all neurons, which limits the model since it is not possible to put more computing resources in a particular region of the image. Another limitation of the original approach is the need to define values for a reasonable number of parameters, which can turn difficult the application of PNNs in contexts in which the user does not have experience. On the other hand, SPNN has a fewer number of parameters. Its structure is determined using a novel method with Delaunay Triangulation and k-means clustering. SPNN achieved better results than PNNs and similar performance when compared to Convolutional Neural Network (CNN) and Support Vector Machine (SVM), but using lower memory capacity and processing time.

Gender recognition from facial images: two or three dimensions?

This paper seeks to compare encoded features from both two-dimensional (2D) and three-dimensional (3D) face images in order to achieve automatic gender recognition with high accuracy and robustness. The Fisher vector encoding method is employed to produce 2D, 3D, and fused features with escalated discriminative power. For 3D face analysis, a two-source photometric stereo (PS) method is introduced that enables 3D surface reconstructions with accurate details as well as desirable efficiency. Moreover, a 2D+3D imaging device, taking the two-source PS method as its core, has been developed that can simultaneously gather color images for 2D evaluations and PS images for 3D analysis. This system inherits the superior reconstruction accuracy from the standard (three or more light) PS method but simplifies the reconstruction algorithm as well as the hardware design by only requiring two light sources. It also offers great potential for facilitating human computer interaction by being accurate, cheap, efficient, and nonintrusive. Ten types of low-level 2D and 3D features have been experimented with and encoded for Fisher vector gender recognition. Evaluations of the Fisher vector encoding method have been performed on the FERET database, Color FERET database, LFW database, and FRGCv2 database, yielding 97.7%, 98.0%, 92.5%, and 96.7% accuracy, respectively. In addition, the comparison of 2D and 3D features has been drawn from a self-collected dataset, which is constructed with the aid of the 2D+3D imaging device in a series of data capture experiments. With a variety of experiments and evaluations, it can be proved that the Fisher vector encoding method outperforms most state-of-the-art gender recognition methods. It has also been observed that 3D features reconstructed by the two-source PS method are able to further boost the Fisher vector gender recognition performance, i.e., up to a 6% increase on the self-collected database.

Change Detection in Synthetic Aperture Radar Images Based on Deep Neural Networks

This paper presents a novel change detection approach for synthetic aperture radar images based on deep learning. The approach accomplishes the detection of the changed and unchanged areas by designing a deep neural network. The main guideline is to produce a change detection map directly from two images with the trained deep neural network. The method can omit the process of generating a difference image (DI) that shows difference degrees between multitemporal synthetic aperture radar images. Thus, it can avoid the effect of the DI on the change detection results. The learning algorithm for deep architectures includes unsupervised feature learning and supervised fine-tuning to complete classification. The unsupervised feature learning aims at learning the representation of the relationships between the two images. In addition, the supervised fine-tuning aims at learning the concepts of the changed and unchanged pixels. Experiments on real data sets and theoretical analysis indicate the advantages, feasibility, and potential of the proposed method. Moreover, based on the results achieved by various traditional algorithms, respectively, deep learning can further improve the detection performance.

Constructive autoassociative neural network for facial recognition

Autoassociative artificial neural networks have been used in many different computer vision applications. However, it is difficult to define the most suitable neural network architecture because this definition is based on previous knowledge and depends on the problem domain. To address this problem, we propose a constructive autoassociative neural network called CANet (Constructive Autoassociative Neural Network). CANet integrates the concepts of receptive fields and autoassociative memory in a dynamic architecture that changes the configuration of the receptive fields by adding new neurons in the hidden layer, while a pruning algorithm removes neurons from the output layer. Neurons in the CANet output layer present lateral inhibitory connections that improve the recognition rate. Experiments in face recognition and facial expression recognition show that the CANet outperforms other methods presented in the literature.

Lateral inhibition pyramidal neural network for image classification

The human visual system is one of the most fascinating and complex mechanisms of the central nervous system that enables our capacity to see. It is through the visual system that we are able to accomplish from the most simple task such as object recognition to the most complex visual interpretation, understanding and perception. Inspired by this sophisticated system, two models based on the properties of the human visual system are proposed. These models are designed based on the concepts of receptive and inhibitory fields. The first model is a pyramidal neural network with lateral inhibition, called lateral inhibition pyramidal neural network. The second proposed model is a supervised image segmentation system, called segmentation and classification based on receptive fields. This work shows that the combination of these two models is beneficial, and the results obtained are better than that of other state-of-the-art methods.

A neural network-based analysis of acoustic courtship signals and female responses in Chorthippus biguttulus grasshoppers

In many animal species, male acoustic courtship signals are evaluated by females for mate choice. At the behavioural level, this phenomenon has been well studied. However, although several song characteristics have been determined to affect the attractiveness of a given song, the mechanisms of the evaluation process remain largely unclear. Here, we present a simple neural network model for analysing and evaluating courtship songs of Chorthippus biguttulus males in real-time. The model achieves a high predictive power of the attractiveness of artificial songs as assigned by real Chorthippus biguttulus females: about 87% of the variance can be explained. It also allows us to determine the relative contribution of different song characteristics to overall attractiveness and how each of the song components influences female responsiveness. In general, the obtained results closely match those of empirical studies. Therefore, our model may be used to obtain a first estimate of male song attractiveness and may thus complement actual testing of female responsiveness in the laboratory. In addition, the model allows including and testing novel song parameters to generate new hypotheses for further experimental studies. The supplemental material of this article contains the article's data in an active, re-usable format.

Adaptive hierarchical architecture for visual recognition

We propose a new hierarchical architecture for visual pattern classification. The new architecture consists of a set of fixed, directional filters and a set of adaptive filters arranged in a cascade structure. The fixed filters are used to extract primitive features such as orientations and edges that are present in a wide range of objects, whereas the adaptive filters can be trained to find complex features that are specific to a given object. Both types of filter are based on the biological mechanism of shunting inhibition. The proposed architecture is applied to two problems: pedestrian detection and car detection. Evaluation results on benchmark data sets demonstrate that the proposed architecture outperforms several existing ones.

MicroRNA signatures predict oestrogen receptor, progesterone receptor and HER2/neu receptor status in breast cancer

INTRODUCTION

Breast cancer is a heterogeneous disease encompassing a number of phenotypically diverse tumours. Expression levels of the oestrogen, progesterone and HER2/neu receptors which characterize clinically distinct breast tumours have been shown to change during disease progression and in response to systemic therapies. Mi(cro)RNAs play critical roles in diverse biological processes and are aberrantly expressed in several human neoplasms including breast cancer, where they function as regulators of tumour behaviour and progression. The aims of this study were to identify miRNA signatures that accurately predict the oestrogen receptor (ER), progesterone receptor (PR) and HER2/neu receptor status of breast cancer patients to provide insight into the regulation of breast cancer phenotypes and progression.

METHODS

Expression profiling of 453 miRNAs was performed in 29 early-stage breast cancer specimens. miRNA signatures associated with ER, PR and HER2/neu status were generated using artificial neural networks (ANN), and expression of specific miRNAs was validated using RQ-PCR.

RESULTS

Stepwise ANN analysis identified predictive miRNA signatures corresponding with oestrogen (miR-342, miR-299, miR-217, miR-190, miR-135b, miR-218), progesterone (miR-520g, miR-377, miR-527-518a, miR-520f-520c) and HER2/neu (miR-520d, miR-181c, miR-302c, miR-376b, miR-30e) receptor status. MiR-342 and miR-520g expression was further analysed in 95 breast tumours. MiR-342 expression was highest in ER and HER2/neu-positive luminal B tumours and lowest in triple-negative tumours. MiR-520g expression was elevated in ER and PR-negative tumours.

CONCLUSIONS

This study demonstrates that ANN analysis reliably identifies biologically relevant miRNAs associated with specific breast cancer phenotypes. The association of specific miRNAs with ER, PR and HER2/neu status indicates a role for these miRNAs in disease classification of breast cancer. Decreased expression of miR-342 in the therapeutically challenging triple-negative breast tumours, increased miR-342 expression in the luminal B tumours, and downregulated miR-520g in ER and PR-positive tumours indicates that not only is dysregulated miRNA expression a marker for poorer prognosis breast cancer, but that it could also present an attractive target for therapeutic intervention.

Intelligent visual recognition and classification of cork tiles with neural networks

An intelligent machine vision system is investigated and used for pattern recognition and classification of seven different types of cork tiles. The system includes image acquisition with a charge-coupled device (CCD) camera, texture feature generation (co-occurrence matrices and Laws' masks), analysis and processing of the feature vectors [linear discriminant analysis (LDA) and principal component analysis (PCA)], and cork tiles classification with feedforward neural networks (NN), employing our GLP(tau) S (genetic low-discrepancy search) hybrid global optimization method. In addition, the same NN are trained with backpropagation (BP) and the obtained results are compared with the ones from GLP(tau) S . The NN generalization abilities are discussed and assessed with respect to the NN architectures and the texture feature sets. The reported results are very encouraging with testing rate reaching up to 95%.

Comparison between analog and digital neural network implementations for range-finding applications

A neural network (NN) was developed in order to increase the distance range of a phase-shift laser range finder and to achieve surface recognition, by using two photoelectrical signals issued from the measurement system. The NN architecture consists of a multilayer perceptron (MLP) with two inputs, three neurons in the hidden layer, and one output. Depending on the application, the NN output has to resolve the ambiguity due to phase-shift measurement by linearizing the inverse of the square law, or to indicate an output voltage corresponding to the tested surface. This embedded system dedicated to optoelectronic measurements was successfully tested with an analog NN, implemented in 0.35- microm complimentary metal-oxide-semiconductor (CMOS) technology, resulting in a threefold increase in the distance range with respect to the one limited by the phase-shift measurement, and by discriminating four types of surfaces (a plastic surface, glossy paper, a painted wall, and a porous surface), at a remote distance between the range finder and the target varying from 0.5 m up to 1.25 m and with a laser beam angle varying between -pi/6 and pi/6 with respect to the target. In this type of application, NN analog implementation provides many advantages, notably use of a small silicon area, low power consumption and no analog-to-digital conversions (ADCs). Nevertheless, digital implementation allows ease of conception and reconfigurability and an embedded weight and bias update. This paper presents the complete measurement system and a comparison between both types of implementation, by developing the advantages and drawbacks relative to each method. An optimized mixed architecture, using both techniques, is then proposed and discussed at the end of the paper.

Efficient object recognition using boundary representation and wavelet neural network

Wavelet neural networks combine the functions of time-frequency localization from the wavelet transform and of self-studying from the neural network, which make them particularly suitable for the classification of complex patterns. In this paper, an efficient object recognition method using boundary representation and the wavelet neural network is proposed. The method employs a wavelet neural network (WNN) to characterize the singularities of the object curvature representation and to perform the object classification at the same time and in an automatic way. The local time-frequency attributes of the singularities on the object boundary are detected by making a preliminary wavelet analysis of the curvature representation. Then, the discriminative scale-translation features of the singularities are stored as the initial scale-translation parameters of the wavelet nodes in the WNN. These parameters are trained to their optimum status during the learning stage. With our approach, as opposed to matching features by convolving the signal with wavelet functions at a large number of scales, the computational burden is significantly reduced. Only a few convolutions are performed at the optimum scale-translation grids during the classification, which makes it suitable for real-time recognition tasks. Compared with the artificial-neural-network-based approaches preceded by wavelet filter banks with fixed scale-translation parameters, the support vector machine (SVM) using traditional Fourier descriptors and K-nearest-neighbor ( K-NN) classifier based on the state-of-the-art shape descriptors, our scheme demonstrates superior and stable discrimination performance under various noisy and affine conditions.

A hierarchical graph neuron scheme for real-time pattern recognition

The hierarchical graph neuron (HGN) implements a single cycle memorization and recall operation through a novel algorithmic design. The HGN is an improvement on the already published original graph neuron (GN) algorithm. In this improved approach, it recognizes incomplete/noisy patterns. It also resolves the crosstalk problem, which is identified in the previous publications, within closely matched patterns. To accomplish this, the HGN links multiple GN networks for filtering noise and crosstalk out of pattern data inputs. Intrinsically, the HGN is a lightweight in-network processing algorithm which does not require expensive floating point computations; hence, it is very suitable for real-time applications and tiny devices such as the wireless sensor networks. This paper describes that the HGN's pattern matching capability and the small response time remain insensitive to the increases in the number of stored patterns. Moreover, the HGN does not require definition of rules or setting of thresholds by the operator to achieve the desired results nor does it require heuristics entailing iterative operations for memorization and recall of patterns.