Convolutional autoencoder based on latent subspace projection for anomaly detection.

Image anomaly detection (AD) is widely researched on many occasions in computer vision tasks. High-dimensional data, such as image data, with noise and complex background is still challenging to detect anomalies under the situation that imbalanced or incomplete data are available. Some deep learning methods can be trained in an unsupervised way and map the original input into low-dimensional manifolds to predict larger differences in anomalies according to normal ones by dimension reduction. However, training a single low-dimension latent space is limited to present the low-dimensional features due to the fact that the noise and irreverent features are mapped into this space, resulting in that the manifolds are not discriminative for detecting anomalies. To address this problem, a new autoencoder framework is proposed in this study with two trainable mutually orthogonal complementary subspaces in the latent space, by latent subspace projection (LSP) mechanism, which is named as LSP-CAE. Specifically, latent subspace projection is used to train the latent image subspace (LIS) and the latent kernel subspace (LKS) in the latent space of the autoencoder-like model respectively, which can enhance learning power of different features from the input instance. The features of normal data are projected into the latent image subspace, while the latent kernel subspace is trained to extract the irrelevant information according to normal features by end-to-end training. To verify the generality and effectiveness of the proposed method, we replace the convolutional network with the fully-connected network contucted in the real-world medical datasets. The anomaly score based on projection norms in two subspace is used to evaluate the anomalies in the testing. Consequently, our proposed method can achieve the best performance according to four public datasets in comparison of the state-of-the-art methods.

Diagnosis of osteoporosis from dental panoramic radiographs using the support vector machine method in a computer-aided system.

BACKGROUND: Early diagnosis of osteoporosis can potentially decrease the risk of fractures and improve the quality of life. Detection of thin inferior cortices of the mandible on dental panoramic radiographs could be useful for identifying postmenopausal women with low bone mineral density (BMD) or osteoporosis. The aim of our study was to assess the diagnostic efficacy of using kernel-based support vector machine (SVM) learning regarding the cortical width of the mandible on dental panoramic radiographs to identify postmenopausal women with low BMD. METHODS: We employed our newly adopted SVM method for continuous measurement of the cortical width of the mandible on dental panoramic radiographs to identify women with low BMD or osteoporosis. The original X-ray image was enhanced, cortical boundaries were determined, distances among the upper and lower boundaries were evaluated and discrimination was performed by a radial basis function. We evaluated the diagnostic efficacy of this newly developed method for identifying women with low BMD (BMD T-score of -1.0 or less) at the lumbar spine and femoral neck in 100 postmenopausal women (≥50 years old) with no previous diagnosis of osteoporosis. Sixty women were used for system training, and 40 were used in testing. RESULTS: The sensitivity and specificity using RBF kernel-SVM method for identifying women with low BMD were 90.9% [95% confidence interval (CI), 85.3-96.5] and 83.8% (95% CI, 76.6-91.0), respectively at the lumbar spine and 90.0% (95% CI, 84.1-95.9) and 69.1% (95% CI, 60.1-78.6), respectively at the femoral neck. The sensitivity and specificity for identifying women with low BMD at either the lumbar spine or femoral neck were 90.6% (95% CI, 92.0-100) and 80.9% (95% CI, 71.0-86.9), respectively. CONCLUSION: Our results suggest that the newly developed system with the SVM method would be useful for identifying postmenopausal women with low skeletal BMD.

Deep Neural Network Models for Colon Cancer Screening.

Early detection of colorectal cancer can significantly facilitate clinicians' decision-making and reduce their workload. This can be achieved using automatic systems with endoscopic and histological images. Recently, the success of deep learning has motivated the development of image- and video-based polyp identification and segmentation. Currently, most diagnostic colonoscopy rooms utilize artificial intelligence methods that are considered to perform well in predicting invasive cancer. Convolutional neural network-based architectures, together with image patches and preprocesses are often widely used. Furthermore, learning transfer and end-to-end learning techniques have been adopted for detection and localization tasks, which improve accuracy and reduce user dependence with limited datasets. However, explainable deep networks that provide transparency, interpretability, reliability, and fairness in clinical diagnostics are preferred. In this review, we summarize the latest advances in such models, with or without transparency, for the prediction of colorectal cancer and also address the knowledge gap in the upcoming technology.

Prediction of carcinogenicity for diverse chemicals based on substructure grouping and SVM modeling.

The Carcinogenicity Reliability Database (CRDB) was constructed by collecting experimental carcinogenicity data on about 1,500 chemicals from six sources, including IARC, and NTP databases, and then by ranking their reliabilities into six unified categories. A wide variety of 911 organic chemicals were selected from the database for QSAR modeling, and 1,504 kinds of different molecular descriptors were calculated, based on their 3D molecular structures as modeled by the Dragon software. Positive (carcinogenic) and negative (non-carcinogenic) chemicals containing various substructures were counted using atom and functional group count descriptors, and the statistical significance of ratios of positives to negatives was tested for those substructures. Very few were judged to be strongly related to carcinogenicity, among substructures known to be responsible for carcinogens as revealed from biomedical studies. In order to develop QSAR models for the prediction of the carcinogenicities of a wide variety of chemicals with a satisfactory performance level, the relationship between the carcinogenicity data with improved reliability and a subset of significant descriptors selected from 1,504 Dragon descriptors was analyzed with a support vector machine (SVM) method: the classification function (SVC) for weighted data in LIBSVM program was used to classify chemicals into two carcinogenic categories (positive or negative), where weights were set depending on the reliabilities of the carcinogenicity data. The quality and stability of the models presented were tested by performing a dual cross-validation procedure. A single SVM model as the first step was developed for all the 911 chemicals using 250 selected descriptors, achieving an overall accuracy level, i.e., positive and negative correct estimate, of about 70%. In order to improve the accuracy of the final model, the 911 chemicals were classified into 20 mutually overlapping subgroups according to contained substructures, a specific SVM model was optimized for each subgroup, and the predicted carcinogenicities of the 911 chemicals were determined by the majorities of the outputs of the corresponding SVM models. The model developed on the basis of grouping of chemicals into 20 substructures predicts the carcinogenicities of a wide variety of chemicals with a satisfactory overall accuracy of approximately 80%.

Deep learning enables automated localization of the metastatic lymph node for thyroid cancer on 131I post-ablation whole-body planar scans.

The accurate detection of radioactive iodine-avid lymph node (LN) metastasis on 131I post-ablation whole-body planar scans (RxWBSs) is important in tracking the progression of the metastatic lymph nodes (mLNs) of patients with papillary thyroid cancer (PTC). However, severe noise artifacts and the indiscernible location of the mLN from adjacent tissues with similar gray-scale values make clinical decisions extremely challenging. This study aims (i) to develop a multilayer fully connected deep network (MFDN) for the automatic recognition of mLNs from thyroid remnant tissue by utilizing the dataset of RxWBSs and (ii) to evaluate its diagnostic performance using post-ablation single-photon emission computed tomography. Image patches focused on the mLN and remnant tissues along with their variations of probability of pixel positions were fed as inputs to the network. With this efficient automatic approach, we achieved a high F1-score and outperformed the physician score (P < 0.001) in detecting mLNs. Competitive segmentation networks on RxWBS displayed moderate performance for the mLN but remained robust for the remnant tissue. Our results demonstrated that the generalization performance with the multiple layers by replicating signal transmission overcome the constraint of local minimum optimization, it can be suitable to localize the unstable location of mLN region on RxWBS and therefore MFDN can be useful in clinical decision-making to track mLN progression for PTC.

Author Correction: Deep learning enables automated localization of the metastatic lymph node for thyroid cancer on 131I post-ablation whole-body planar scans.

An amendment to this paper has been published and can be accessed via a link at the top of the paper.

Prediction of Sequential Organelles Localization under Imbalance using A Balanced Deep U-Net.

Assessing the structure and function of organelles in living organisms of the primitive unicellular red algae Cyanidioschyzon merolae on three-dimensional sequential images demands a reliable automated technique in the class imbalance among various cellular structures during mitosis. Existing classification networks with commonly used loss functions were focused on larger numbers of cellular structures that lead to the unreliability of the system. Hence, we proposed a balanced deep regularized weighted compound dice loss (RWCDL) network for better localization of cell organelles. Specifically, we introduced two new loss functions, namely compound dice (CD) and RWCD by implementing multi-class variant dice and weighting mechanism, respectively for maximizing weights of peroxisome and nucleus among five classes as the main contribution of this study. We extended the Unet-like convolution neural network (CNN) architecture for evaluating the ability of our proposed loss functions for improved segmentation. The feasibility of the proposed approach is confirmed with three different large scale mitotic cycle data set with different number of occurrences of cell organelles. In addition, we compared the training behavior of our designed architectures with the ground truth segmentation using various performance measures. The proposed balanced RWCDL network generated the highest area under the curve (AUC) value in elevating the small and obscure peroxisome and nucleus, which is 30% higher than the network with commonly used mean square error (MSE) and dice loss (DL) functions. The experimental results indicated that the proposed approach can efficiently identify the cellular structures, even when the contour between the cells is obscure and thus convinced that the balanced deep RWCDL approach is reliable and can be helpful for biologist to accurately identify the relationship between the cell behavior and structures of cell organelles during mitosis.

Critical texture pattern feature assessment for characterizing colonies of induced pluripotent stem cells through machine learning techniques.

The objectives of this study are to assess various automated texture features obtained from the segmented colony regions of induced pluripotent stem cells (iPSCs) and confirm their potential for characterizing the colonies using different machine learning techniques. One hundred and fifty-one features quantified using shape-based, moment-based, statistical and spectral texture feature groups are extracted from phase-contrast microscopic colony images of iPSCs. The forward stepwise regression model is implemented to select the most appropriate features required for categorizing the colonies. Support vector machine (SVM), random forest (RF), multilayer perceptron (MLP), decision tree (DT), and adaptive boosting (Adaboost) classifiers are used with ten-fold cross-validation to evaluate the texture features within each texture feature group and fused-features group to characterize healthy and unhealthy colonies of iPSCs. Overall, based on the classification performances of the four texture feature groups using the five classifier models, statistical features always exhibit a high predictive capacity (>87.5%). However, the classification performance using fused texture patterns with statistical, shape-based, and moment-based features was found to be robust and reliable with fewer false positive and false negative values compared to the features when either one is used for the classification of colonies of iPSCs. Furthermore, the results showcase that the SVM, RF and Adaboost classifiers deliver better classification performances than DT and MLP. Our findings suggest that the proposed automated fused statistical, shape-based, and moment-based texture pattern features trained with machine learning techniques are potentially more appropriate and helpful to biologists for characterizing colonies of stem cells.

Large-scale clustering of CAGE tag expression data.

BACKGROUND: Recent analyses have suggested that many genes possess multiple transcription start sites (TSSs) that are differentially utilized in different tissues and cell lines. We have identified a huge number of TSSs mapped onto the mouse genome using the cap analysis of gene expression (CAGE) method. The standard hierarchical clustering algorithm, which gives us easily understandable graphical tree images, has difficulties in processing such huge amounts of TSS data and a better method to calculate and display the results is needed. RESULTS: We use a combination of hierarchical and non-hierarchical clustering to cluster expression profiles of TSSs based on a large amount of CAGE data to profit from the best of both methods. We processed the genome-wide expression data, including 159,075 TSSs derived from 127 RNA samples of various organs of mouse, and succeeded in categorizing them into 70-100 clusters. The clusters exhibited intriguing biological features: a cluster supergroup with a ubiquitous expression profile, tissue-specific patterns, a distinct distribution of non-coding RNA and functional TSS groups. CONCLUSION: Our approach succeeded in greatly reducing the calculation cost, and is an appropriate solution for analyzing large-scale TSS usage data.

Deep vector-based convolutional neural network approach for automatic recognition of colonies of induced pluripotent stem cells.

Pluripotent stem cells can potentially be used in clinical applications as a model for studying disease progress. This tracking of disease-causing events in cells requires constant assessment of the quality of stem cells. Existing approaches are inadequate for robust and automated differentiation of stem cell colonies. In this study, we developed a new model of vector-based convolutional neural network (V-CNN) with respect to extracted features of the induced pluripotent stem cell (iPSC) colony for distinguishing colony characteristics. A transfer function from the feature vectors to the virtual image was generated at the front of the CNN in order for classification of feature vectors of healthy and unhealthy colonies. The robustness of the proposed V-CNN model in distinguishing colonies was compared with that of the competitive support vector machine (SVM) classifier based on morphological, textural, and combined features. Additionally, five-fold cross-validation was used to investigate the performance of the V-CNN model. The precision, recall, and F-measure values of the V-CNN model were comparatively higher than those of the SVM classifier, with a range of 87-93%, indicating fewer false positives and false negative rates. Furthermore, for determining the quality of colonies, the V-CNN model showed higher accuracy values based on morphological (95.5%), textural (91.0%), and combined (93.2%) features than those estimated with the SVM classifier (86.7, 83.3, and 83.4%, respectively). Similarly, the accuracy of the feature sets using five-fold cross-validation was above 90% for the V-CNN model, whereas that yielded by the SVM model was in the range of 75-77%. We thus concluded that the proposed V-CNN model outperforms the conventional SVM classifier, which strongly suggests that it as a reliable framework for robust colony classification of iPSCs. It can also serve as a cost-effective quality recognition tool during culture and other experimental procedures.

Automatic detection of osteoporosis based on hybrid genetic swarm fuzzy classifier approaches.

OBJECTIVES: This study proposed a new automated screening system based on a hybrid genetic swarm fuzzy (GSF) classifier using digital dental panoramic radiographs to diagnose females with a low bone mineral density (BMD) or osteoporosis. METHODS: The geometrical attributes of both the mandibular cortical bone and trabecular bone were acquired using previously developed software. Designing an automated system for osteoporosis screening involved partitioning of the input attributes to generate an initial membership function (MF) and a rule set (RS), classification using a fuzzy inference system and optimization of the generated MF and RS using the genetic swarm algorithm. Fivefold cross-validation (5-FCV) was used to estimate the classification accuracy of the hybrid GSF classifier. The performance of the hybrid GSF classifier has been further compared with that of individual genetic algorithm and particle swarm optimization fuzzy classifiers. RESULTS: Proposed hybrid GSF classifier in identifying low BMD or osteoporosis at the lumbar spine and femoral neck BMD was evaluated. The sensitivity, specificity and accuracy of the hybrid GSF with optimized MF and RS in identifying females with a low BMD were 95.3%, 94.7% and 96.01%, respectively, at the lumbar spine and 99.1%, 98.4% and 98.9%, respectively, at the femoral neck BMD. The diagnostic performance of the proposed system with femoral neck BMD was 0.986 with a confidence interval of 0.942-0.998. The highest mean accuracy using 5-FCV was 97.9% with femoral neck BMD. CONCLUSIONS: The combination of high accuracy along with its interpretation ability makes this proposed automatic system using hybrid GSF classifier capable of identifying a large proportion of undetected low BMD or osteoporosis at its early stage.

A robust classifier combined with an auto-associative network for completing partly occluded images.

This paper describes an approach for constructing a classifier which is unaffected by occlusions in images. We propose a method for integrating an auto-associative network into a simple classifier. As the auto-associative network can recall the original image from a partly occluded input image, we can employ it to detect occluded regions and complete the input image by replacing those regions with recalled pixels. By iterating this reconstruction process, the integrated network is able to classify target objects with occlusions robustly. To confirm the effectiveness of this method, we performed experiments involving face image classification. It is shown that the classification performance is not decreased, even if about 30% of the face image is occluded.

Mixture of Subspaces Image Representation and Compact Coding for Large-Scale Image Retrieval.

There are two major approaches to content-based image retrieval using local image descriptors. One is descriptor-by-descriptor matching and the other is based on comparison of global image representation that describes the set of local descriptors of each image. In large-scale problems, the latter is preferred due to its smaller memory requirements; however, it tends to be inferior to the former in terms of retrieval accuracy. To achieve both low memory cost and high accuracy, we investigate an asymmetric approach in which the probability distribution of local descriptors is modeled for each individual database image while the local descriptors of a query are used as is. We adopt a mixture model of probabilistic principal component analysis. The model parameters constitute a global image representation to be stored in database. Then the likelihood function is employed to compute a matching score between each database image and a query. We also propose an algorithm to encode our image representation into more compact codes. Experimental results demonstrate that our method can represent each database image in less than several hundred bytes achieving higher retrieval accuracy than the state-of-the-art method using Fisher vectors.

Transfer learning for Bag-of-Visual words approach to NBI endoscopic image classification.

We address a problem of endoscopic image classification taken by different (e.g., old and new) endoscopies. Our proposed method formulates the problem as a constraint optimization that estimates a linear transformation between feature vectors (or Bag-of-Visual words histograms) in a framework of transfer learning. Experimental results show that the proposed method works much better than the case without feature transformation.

Automatic analysis of composite physical signals using non-negative factorization and information criterion.

In time-resolved spectroscopy, composite signal sequences representing energy transfer in fluorescence materials are measured, and the physical characteristics of the materials are analyzed. Each signal sequence is represented by a sum of non-negative signal components, which are expressed by model functions. For analyzing the physical characteristics of a measured signal sequence, the parameters of the model functions are estimated. Furthermore, in order to quantitatively analyze real measurement data and to reduce the risk of improper decisions, it is necessary to obtain the statistical characteristics from several sequences rather than just a single sequence. In the present paper, we propose an automatic method by which to analyze composite signals using non-negative factorization and an information criterion. The proposed method decomposes the composite signal sequences using non-negative factorization subjected to parametric base functions. The number of components (i.e., rank) is also estimated using Akaike's information criterion. Experiments using simulated and real data reveal that the proposed method automatically estimates the acceptable ranks and parameters.

An evolutionary approach for gene selection and classification of microarray data based on SVM error-bound theories.

Microarrays have thousands to tens-of-thousands of gene features, but only a few hundred patient samples are available. The fundamental problem in microarray data analysis is identifying genes whose disruption causes congenital or acquired disease in humans. In this paper, we propose a new evolutionary method that can efficiently select a subset of potentially informative genes for support vector machine (SVM) classifiers. The proposed evolutionary method uses SVM with a given subset of gene features to evaluate the fitness function, and new subsets of features are selected based on the estimates of generalization error of SVMs and frequency of occurrence of the features in the evolutionary approach. Thus, in theory, selected genes reflect to some extent the generalization performance of SVM classifiers. We compare our proposed method with several existing methods and find that the proposed method can obtain better classification accuracy with a smaller number of selected genes than the existing methods.