Search

Clinical studies from WHO have demonstrated that only 50–70% of patients adhere properly to prescribed drug therapy. Such adherence failure can impact therapeutic efficacy for the patients in question and compromises data quality around the population-level efficacy of the drug for the indications targeted. In this study, we applied various ensemble learning and deep learning models to predict medication adherence among patients. Our contribution to this endeavour involves targeting the problem of adherence prediction for a particularly challenging class of patients who self-administer injectable medication at home. Our prediction pipeline, based on event history, comprises a connected sharps bin which aims to help patients better manage their condition and improve outcomes. In other words, the efficiency of interventions can be significantly improved by prioritizing the patients who are most likely to be non-adherent. The collected data comprising a rich event feature set may be exploited for the purposes of predicting the status of the next adherence state for individual patients. This paper reports on how this concept can be realized through an investigation using a wide range of ensemble learning and deep learning models on a real-world dataset collected from such a system. The dataset investigated comprises 342,174 historic injection disposal records collected over the course of more than 5 years. A comprehensive comparison of different models is given in this paper. Moreover, we demonstrate that the selected best performer, long short-term memory (LSTM), generalizes well by deploying it in a true future testing dataset. The proposed end-to-end pipeline is capable of predicting patient failure in adhering to their therapeutic regimen with 77.35 % accuracy (Specificity: 78.28 %, Sensitivity: 76.42%, Precision: 77.87%,F1 score: 0.7714, ROC AUC: 0.8390).

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

Artificial intelligence methods such as deep neural networks promise unprecedented capabilities in healthcare, from diagnosing diseases to prescribing treatments. While this can eventually produce a valuable suite of tools for automating clinical workflows, a critical step forward is to ensure that the predictive models are reliable and to enable a rigorous introspection of their behavior. This has led to the design of explainable AI techniques that are aimed at uncovering the relationships between discernible data signatures and decisions from machine-learned models, and characterizing strengths/weaknesses of models. In this context, the so-called counterfactual explanations that synthesize small, interpretable changes to a given query sample while producing desired changes in model predictions to support user-specified hypotheses (e.g., progressive change in disease severity) have become popular. When a model’s predictions are not well-calibrated (i.e., the prediction confidences are not indicative of the likelihood of the predictions being correct), the inverse problem of synthesizing counterfactuals can produce explanations with irrelevant feature manipulations. Hence, in this paper, we propose to leverage prediction uncertainties from the learned models to better guide this optimization. To this end, we present TraCE (Training Calibration-based Explainers), a counterfactual generation approach for deep models in medical image analysis, which utilizes pre-trained generative models and a novel uncertainty-based interval calibration strategy for synthesizing hypothesis-driven explanations. By leveraging uncertainty estimates in the optimization process, TraCE can consistently produce meaningful counterfactual evidences and elucidate complex decision boundaries learned by deep classifiers. Furthermore, we demonstrate the effectiveness of TraCE in revealing intricate relationships between different patient attributes and in detecting shortcuts, arising from unintended biases, in learned models. Given the widespread adoption of machine-learned solutions in radiology, our study focuses on deep models used for identifying anomalies in chest X-ray images. Using rigorous empirical studies, we demonstrate the superiority of TraCE explanations over several state-of-the-art baseline approaches, in terms of several widely adopted evaluation metrics in counterfactual reasoning. Our findings show that TraCE can be used to obtain a holistic understanding of deep models by enabling progressive exploration of decision boundaries, detecting shortcuts, and inferring relationships between patient attributes and disease severity.

Since the outbreak of COVID-19 in 2019, the rapid spread of the epidemic has brought huge challenges to medical institutions. If the pathological region in the COVID-19 CT image can be automatically segmented, it will help doctors quickly determine the patient’s infection, thereby speeding up the diagnosis process. To be able to automatically segment the infected area, we proposed a new network structure and named QC-HC U-Net. First, we combine residual connection and dense connection to form a new connection method and apply it to the encoder and the decoder. Second, we choose to add Hypercolumns in the decoder section. Compared with the benchmark 3D U-Net, the improved network can effectively avoid vanishing gradient while extracting more features. To improve the situation of insufficient data, resampling and data enhancement methods are selected in this paper to expand the datasets. We used 63 cases of MSD lung tumor data for training and testing, continuously verified to ensure the training effect of this model, and then selected 20 cases of public COVID-19 data for training and testing. Experimental results showed that in the segmentation of COVID-19, the specificity and sensitivity were 85.3% and 83.6%, respectively, and in the segmentation of MSD lung tumors, the specificity and sensitivity were 81.45% and 80.93%, respectively, without any fitting.

Deep neural networks (DNNs) have achieved state-of-the-art performance in many important domains, including medical diagnosis, security, and autonomous driving. In domains where safety is highly critical, an erroneous decision can result in serious consequences. While a perfect prediction accuracy is not always achievable, recent work on Bayesian deep networks shows that it is possible to know when DNNs are more likely to make mistakes. Knowing what DNNs do not know is desirable to increase the safety of deep learning technology in sensitive applications; Bayesian neural networks attempt to address this challenge. Traditional approaches are computationally intractable and do not scale well to large, complex neural network architectures. In this paper, we develop a theoretical framework to approximate Bayesian inference for DNNs by imposing a Bernoulli distribution on the model weights. This method called Monte Carlo DropConnect (MC-DropConnect) gives us a tool to represent the model uncertainty with little change in the overall model structure or computational cost. We extensively validate the proposed algorithm on multiple network architectures and datasets for classification and semantic segmentation tasks. We also propose new metrics to quantify uncertainty estimates. This enables an objective comparison between MC-DropConnect and prior approaches. Our empirical results demonstrate that the proposed framework yields significant improvement in both prediction accuracy and uncertainty estimation quality compared to the state of the art.

This paper proposes a convolutional neural network (CNN)-based deep learning model for predicting the difficulty of extracting a mandibular third molar using a panoramic radiographic image. The applied dataset includes a total of 1053 mandibular third molars from 600 preoperative panoramic radiographic images. The extraction difficulty was evaluated based on the consensus of three human observers using the Pederson difficulty score (PDS). The classification model used a ResNet-34 pretrained on the ImageNet dataset. The correlation between the PDS values determined by the proposed model and those measured by the experts was calculated. The prediction accuracies for C1 (depth), C2 (ramal relationship), and C3 (angulation) were 78.91%, 82.03%, and 90.23%, respectively. The results confirm that the proposed CNN-based deep learning model could be used to predict the difficulty of extracting a mandibular third molar using a panoramic radiographic image.

Texture features are designed to quantitatively evaluate patterns of spatial distribution of image pixels for purposes of image analysis and interpretation. Unexplained variations in the texture patterns often lead to misinterpretation and undesirable consequences in medical image analysis. In this paper we explore the ability of machine learning (ML) methods to design a radiology test of Osteoarthritis (OA) at early stage when the number of patients’ cases is small. In our experiments we use high-resolution X-ray images of knees in patients which were identified with Kellgren–Lawrence scores progressing from 1. The existing ML methods have provided a limited diagnostic accuracy, whilst the proposed Group Method of Data Handling strategy of Deep Learning has significantly extended the diagnostic test. The comparative experiments demonstrate that the proposed framework using the Zernike-based texture features has significantly improved the diagnostic accuracy on average by 11%. This allows us to conclude that the designed model for early diagnostic of OA will provide more accurate radiology tests, although new study is required when a large number of patients’ cases will be available.

This paper introduces a study on the classification of aortic stenosis (AS) based on cardio-mechanical signals collected using non-invasive wearable inertial sensors. Measurements were taken from 21 AS patients and 13 non-AS subjects. A feature analysis framework utilizing Elastic Net was implemented to reduce the features generated by continuous wavelet transform (CWT). Performance comparisons were conducted among several machine learning (ML) algorithms, including decision tree, random forest, multi-layer perceptron neural network, and extreme gradient boosting. In addition, a two-dimensional convolutional neural network (2D-CNN) was developed using the CWT coefficients as images. The 2D-CNN was made with a custom-built architecture and a CNN based on Mobile Net via transfer learning. After the reduction of features by 95.47%, the results obtained report 0.87 on accuracy by decision tree, 0.96 by random forest, 0.91 by simple neural network, and 0.95 by XGBoost. Via the 2D-CNN framework, the transfer learning of Mobile Net shows an accuracy of 0.91, while the custom-constructed classifier reveals an accuracy of 0.89. Our results validate the effectiveness of the feature selection and classification framework. They also show a promising potential for the implementation of deep learning tools on the classification of AS.

Deep neural networks (DNNs) have been widely applied for detecting COVID-19 in medical images. Existing studies mainly apply transfer learning and other data representation strategies to generate accurate point estimates. The generalization power of these networks is always questionable due to being developed using small datasets and failing to report their predictive confidence. Quantifying uncertainties associated with DNN predictions is a prerequisite for their trusted deployment in medical settings. Here we apply and evaluate three uncertainty quantification techniques for COVID-19 detection using chest X-Ray (CXR) images. The novel concept of uncertainty confusion matrix is proposed and new performance metrics for the objective evaluation of uncertainty estimates are introduced. Through comprehensive experiments, it is shown that networks pertained on CXR images outperform networks pretrained on natural image datasets such as ImageNet. Qualitatively and quantitatively evaluations also reveal that the predictive uncertainty estimates are statistically higher for erroneous predictions than correct predictions. Accordingly, uncertainty quantification methods are capable of flagging risky predictions with high uncertainty estimates. We also observe that ensemble methods more reliably capture uncertainties during the inference. DNN-based solutions for COVID-19 detection have been mainly proposed without any principled mechanism for risk mitigation. Previous studies have mainly focused on on generating single-valued predictions using pretrained DNNs. In this paper, we comprehensively apply and comparatively evaluate three uncertainty quantification techniques for COVID-19 detection using chest X-Ray images. The novel concept of uncertainty confusion matrix is proposed and new performance metrics for the objective evaluation of uncertainty estimates are introduced for the first time. Using these new uncertainty performance metrics, we quantitatively demonstrate where and when we could trust DNN predictions for COVID-19 detection from chest X-rays. It is important to note the proposed novel uncertainty evaluation metrics are generic and could be applied for evaluation of probabilistic forecasts in all classification problems.

Artificial intelligence provides the opportunity to reveal important information buried in large amounts of complex data. Electronic health records (eHRs) are a source of such big data that provide a multitude of health related clinical information about patients. However, text data from eHRs, e.g., discharge summary notes, are challenging in their analysis because these notes are free-form texts and the writing formats and styles vary considerably between different records. For this reason, in this paper we study deep learning neural networks in combination with natural language processing to analyze text data from clinical discharge summaries. We provide a detail analysis of patient phenotyping, i.e., the automatic prediction of ten patient disorders, by investigating the influence of network architectures, sample sizes and information content of tokens. Importantly, for patients suffering from Chronic Pain, the disorder that is the most difficult one to classify, we find the largest performance gain for a combined word- and sentence-level input convolutional neural network (ws-CNN). As a general result, we find that the combination of data quality and data quantity of the text data is playing a crucial role for using more complex network architectures that improve significantly beyond a word-level input CNN model. From our investigations of learning curves and token selection mechanisms, we conclude that for such a transition one requires larger sample sizes because the amount of information per sample is quite small and only carried by few tokens and token categories. Interestingly, we found that the token frequency in the eHRs follow a Zipf law and we utilized this behavior to investigate the information content of tokens by defining a token selection mechanism. The latter addresses also issues of explainable AI.

Automatic vertebrae localization and segmentation in computed tomography (CT) are fundamental for spinal image analysis and spine surgery with computer-assisted surgery systems. But they remain challenging due to high variation in spinal anatomy among patients. In this paper, we proposed a deep-learning approach for automatic CT vertebrae localization and segmentation with a two-stage Dense-U-Net. The first stage used a 2D-Dense-U-Net to localize vertebrae by detecting the vertebrae centroids with dense labels and 2D slices. The second stage segmented the specific vertebra within a region-of-interest identified based on the centroid using 3D-Dense-U-Net. Finally, each segmented vertebra was merged into a complete spine and resampled to original resolution. We evaluated our method on the dataset from the CSI 2014 Workshop with 6 metrics: location error (1.69 ± 0.78 mm), detection rate (100%) for vertebrae localization; the dice coefficient (0.953 ± 0.014), intersection over union (0.911 ± 0.025), Hausdorff distance (4.013 ± 2.128 mm), pixel accuracy (0.998 ± 0.001) for vertebrae segmentation. The experimental results demonstrated the efficiency of the proposed method. Furthermore, evaluation on the dataset from the xVertSeg challenge with location error (4.12 ± 2.31), detection rate (100%), dice coefficient (0.877 ± 0.035) shows the generalizability of our method. In summary, our solution localized the vertebrae successfully by detecting the centroids of vertebrae and implemented instance segmentation of vertebrae in the whole spine.

With the presence of novel coronavirus disease at the end of 2019, several approaches were proposed to help physicians detect the disease, such as using deep learning to recognize lung involvement based on the pattern of pneumonia. These approaches rely on analyzing the CT images and exploring the COVID-19 pathologies in the lung. Most of the successful methods are based on the deep learning technique, which is state-of-the-art. Nevertheless, the big drawback of the deep approaches is their need for many samples, which is not always possible. This work proposes a combined deep architecture that benefits both employed architectures of DenseNet and CapsNet. To more generalize the deep model, we propose a regularization term with much fewer parameters. The network convergence significantly improved, especially when the number of training data is small. We also propose a novel Cost-sensitive loss function for imbalanced data that makes our model feasible for the condition with a limited number of positive data. Our novelties make our approach more intelligent and potent in real-world situations with imbalanced data, popular in hospitals. We analyzed our approach on two publicly available datasets, HUST and COVID-CT, with different protocols. In the first protocol of HUST, we followed the original paper setup and outperformed it. With the second protocol of HUST, we show our approach superiority concerning imbalanced data. Finally, with three different validations of the COVID-CT, we provide evaluations in the presence of a low number of data along with a comparison with state-of-the-art.

Coronavirus 2019 (COVID-19) is a new acute respiratory disease that has spread rapidly throughout the world. In this paper, a lightweight convolutional neural network (CNN) model named multi-scale gated multi-head attention depthwise separable CNN (MGMADS-CNN) is proposed, which is based on attention mechanism and depthwise separable convolution. A multi-scale gated multi-head attention mechanism is designed to extract effective feature information from the COVID-19 X-ray and CT images for classification. Moreover, the depthwise separable convolution layers are adopted as MGMADS-CNN’s backbone to reduce the model size and parameters. The LeNet-5, AlexNet, GoogLeNet, ResNet, VGGNet-16, and three MGMADS-CNN models are trained, validated and tested with tenfold cross-validation on X-ray and CT images. The results show that MGMADS-CNN with three attention layers (MGMADS-3) has achieved accuracy of 96.75% on X-ray images and 98.25% on CT images. The specificity and sensitivity are 98.06% and 96.6% on X-ray images, and 98.17% and 98.05% on CT images. The size of MGMADS-3 model is only 43.6 M bytes. In addition, the detection speed of MGMADS-3 on X-ray images and CT images are 6.09 ms and 4.23 ms for per image, respectively. It is proved that the MGMADS-3 can detect and classify COVID-19 faster with higher accuracy and efficiency.

Among the most leading causes of mortality across the globe are infectious diseases which have cost tremendous lives with the latest being coronavirus (COVID-19) that has become the most recent challenging issue. The extreme nature of this infectious virus and its ability to spread without control has made it mandatory to find an efficient auto-diagnosis system to assist the people who work in touch with the patients. As fuzzy logic is considered a powerful technique for modeling vagueness in medical practice, an Adaptive Neuro-Fuzzy Inference System (ANFIS) was proposed in this paper as a key rule for automatic COVID-19 detection from chest X-ray images based on the characteristics derived by texture analysis using gray level co-occurrence matrix (GLCM) technique. Unlike the proposed method, especially deep learning-based approaches, the proposed ANFIS-based method can work on small datasets. The results were promising performance accuracy, and compared with the other state-of-the-art techniques, the proposed method gives the same performance as the deep learning with complex architectures using many backbone.

Cancer tumor classification based on morphological characteristics alone has been shown to have serious limitations. Breast, lung, colorectal, thyroid, and ovarian are the most commonly diagnosed cancers among women. Precise classification of cancers into their types is considered a vital problem for cancer diagnosis and therapy. In this paper, we proposed a stacking ensemble deep learning model based on one-dimensional convolutional neural network (1D-CNN) to perform a multi-class classification on the five common cancers among women based on RNASeq data. The RNASeq gene expression data was downloaded from Pan-Cancer Atlas using GDCquery function of the TCGAbiolinks package in the R software. We used least absolute shrinkage and selection operator (LASSO) as feature selection method. We compared the results of the new proposed model with and without LASSO with the results of the single 1D-CNN and machine learning methods which include support vector machines with radial basis function, linear, and polynomial kernels; artificial neural networks; k-nearest neighbors; bagging trees. The results show that the proposed model with and without LASSO has a better performance compared to other classifiers. Also, the results show that the machine learning methods (SVM-R, SVM-L, SVM-P, ANN, KNN, and bagging trees) with under-sampling have better performance than with over-sampling techniques. This is supported by the statistical significance test of accuracy where the p-values for differences between the SVM-R and SVM-P, SVM-R and ANN, SVM-R and KNN are found to be p = 0.003, p = p = p = 0.009. Moreover, SVM-P and ANN, SVM-P and KNN are found to be significantly different with p-values p = p = p-values p = p = 0.004, respectively. Thus, the proposed model can help in the early detection and diagnosis of cancer in women, and hence aid in designing early treatment strategies to improve survival.

This work develops a robust classifier for a COVID-19 pre-screening model from crowdsourced cough sound data. The crowdsourced cough recordings contain a variable number of coughs, with some input sound files more informative than the others. Accurate detection of COVID-19 from the sound datasets requires overcoming two main challenges (i) the variable number of coughs in each recording and (ii) the low number of COVID-positive cases compared to healthy coughs in the data. We use two open datasets of crowdsourced cough recordings and segment each cough recording into non-overlapping coughs. The segmentation enriches the original data without oversampling by splitting the original cough sound files into non-overlapping segments. Splitting the sound files enables us to increase the samples of the minority class (COVID-19) without changing the feature distribution of the COVID-19 samples resulted from applying oversampling techniques. Each cough sound segment is transformed into six image representations for further analyses. We conduct extensive experiments with shallow machine learning, Convolutional Neural Network (CNN), and pre-trained CNN models. The results of our models were compared to other recently published papers that apply machine learning to cough sound data for COVID-19 detection. Our method demonstrated a high performance using an ensemble model on the testing dataset with area under receiver operating characteristics curve = 0.77, precision = 0.80, recall = 0.71, F1 measure = 0.75, and Kappa = 0.53. The results show an improvement in the prediction accuracy of our COVID-19 pre-screening model compared to the other models.

COVID-19 has crippled the world’s healthcare systems, setting back the economy and taking the lives of several people. Although potential vaccines are being tested and supplied around the world, it will take a long time to reach every human being, more so with new variants of the virus emerging, enforcing a lockdown-like situation on parts of the world. Thus, there is a dire need for early and accurate detection of COVID-19 to prevent the spread of the disease, even more. The current gold-standard RT-PCR test is only 71% sensitive and is a laborious test to perform, leading to the incapability of conducting the population-wide screening. To this end, in this paper, we propose an automated COVID-19 detection system that uses CT-scan images of the lungs for classifying the same into COVID and Non-COVID cases. The proposed method applies an ensemble strategy that generates fuzzy ranks of the base classification models using the Gompertz function and fuses the decision scores of the base models adaptively to make the final predictions on the test cases. Three transfer learning-based convolutional neural network models are used, namely VGG-11, Wide ResNet-50-2, and Inception v3, to generate the decision scores to be fused by the proposed ensemble model. The framework has been evaluated on two publicly available chest CT scan datasets achieving state-of-the-art performance, justifying the reliability of the model. The relevant source codes related to the present work is available in: GitHub.

The main purpose of this work is to investigate and compare several deep learning enhanced techniques applied to X-ray and CT-scan medical images for the detection of COVID-19. In this paper, we used four powerful pre-trained CNN models, VGG16, DenseNet121, ResNet50,and ResNet152, for the COVID-19 CT-scan binary classification task. The proposed Fast.AI ResNet framework was designed to find out the best architecture, pre-processing, and training parameters for the models largely automatically. The accuracy and F1-score were both above 96% in the diagnosis of COVID-19 using CT-scan images. In addition, we applied transfer learning techniques to overcome the insufficient data and to improve the training time. The binary and multi-class classification of X-ray images tasks were performed by utilizing enhanced VGG16 deep transfer learning architecture. High accuracy of 99% was achieved by enhanced VGG16 in the detection of X-ray images from COVID-19 and pneumonia. The accuracy and validity of the algorithms were assessed on X-ray and CT-scan well-known public datasets. The proposed methods have better results for COVID-19 diagnosis than other related in literature. In our opinion, our work can help virologists and radiologists to make a better and faster diagnosis in the struggle against the outbreak of COVID-19. The research leading to these results received funding from the Innovative Medicines Innitiative 2 Joint Undertaking (JU) under grant agreement No 853989. The JU receives support from the European Union's Horizon 2020 research and innovation programme and EFPIA and Global Alliance for TB Drug Development non profit organisation, Bill & Melinda Gates Foundation and University of Dundee.

By emulating biological features in brain, Spiking Neural Networks (SNNs) offer an energy-efficient alternative to conventional deep learning. To make SNNs ubiquitous, a ‘visual explanation’ technique for analysing and explaining the internal spike behavior of such temporal deep SNNs is crucial. Explaining SNNs visually will make the network more transparent giving the end-user a tool to understand how SNNs make temporal predictions and why they make a certain decision. In this paper, we propose a bio-plausible visual explanation tool for SNNs, called Spike Activation Map (SAM). SAM yields a heatmap (i.e., localization map) corresponding to each time-step of input data by highlighting neurons with short inter-spike interval activity. Interestingly, without the use of gradients and ground truth, SAM produces a temporal localization map highlighting the region of interest in an image attributed to an SNN’s prediction at each time-step. Overall, SAM outsets the beginning of a new research area ‘explainable neuromorphic computing’ that will ultimately allow end-users to establish appropriate trust in predictions from SNNs.

Protein is an essential component of the living organism. The prediction of protein-protein interactions (PPIs) has important implications for understanding the behavioral processes of life, preventing diseases, and developing new drugs. Although the development of high-throughput technology makes it possible to identify PPIs in large-scale biological experiments, it restricts the extensive use of experimental methods due to the constraints of time, cost, false positive rate and other conditions. Therefore, there is an urgent need for computational methods as a supplement to experimental methods to predict PPIs rapidly and accurately. In this paper, we propose a novel approach, namely CNN-FSRF, for predicting PPIs based on protein sequence by combining deep learning Convolution Neural Network (CNN) with Feature-Selective Rotation Forest (FSRF). The proposed method firstly converts the protein sequence into the Position-Specific Scoring Matrix (PSSM) containing biological evolution information, then uses CNN to objectively and efficiently extracts the deeply hidden features of the protein, and finally removes the redundant noise information by FSRF and gives the accurate prediction results. When performed on the PPIs datasets Yeast and Helicobacter pylori, CNN-FSRF achieved a prediction accuracy of 97.75% and 88.96%. To further evaluate the prediction performance, we compared CNN-FSRF with SVM and other existing methods. In addition, we also verified the performance of CNN-FSRF on independent datasets. Excellent experimental results indicate that CNN-FSRF can be used as a useful complement to biological experiments to identify protein interactions.

Robotic weed control has seen increased research of late with its potential for boosting productivity in agriculture. Majority of works focus on developing robotics for croplands, ignoring the weed management problems facing rangeland stock farmers. Perhaps the greatest obstacle to widespread uptake of robotic weed control is the robust classification of weed species in their natural environment. The unparalleled successes of deep learning make it an ideal candidate for recognising various weed species in the complex rangeland environment. This work contributes the first large, public, multiclass image dataset of weed species from the Australian rangelands; allowing for the development of robust classification methods to make robotic weed control viable. The DeepWeeds dataset consists of 17,509 labelled images of eight nationally significant weed species native to eight locations across northern Australia. This paper presents a baseline for classification performance on the dataset using the benchmark deep learning models, Inception-v3 and ResNet-50. These models achieved an average classification accuracy of 95.1% and 95.7%, respectively. We also demonstrate real time performance of the ResNet-50 architecture, with an average inference time of 53.4 ms per image. These strong results bode well for future field implementation of robotic weed control methods in the Australian rangelands., Comment: 14 pages, 8 figures, 4 tables

The application of machine learning (ML) techniques in materials science has attracted significant attention in recent years, due to their impressive ability to efficiently extract data-driven linkages from various input materials representations to their output properties. While the application of traditional ML techniques has become quite ubiquitous, there have been limited applications of more advanced deep learning (DL) techniques, primarily because big materials datasets are relatively rare. Given the demonstrated potential and advantages of DL and the increasing availability of big materials datasets, it is attractive to go for deeper neural networks in a bid to boost model performance, but in reality, it leads to performance degradation due to the vanishing gradient problem. In this paper, we address the question of how to enable deeper learning for cases where big materials data is available. Here, we present a general deep learning framework based on Individual Residual learning (IRNet) composed of very deep neural networks that can work with any vector-based materials representation as input to build accurate property prediction models. We find that the proposed IRNet models can not only successfully alleviate the vanishing gradient problem and enable deeper learning, but also lead to significantly (up to 47%) better model accuracy as compared to plain deep neural networks and traditional ML techniques for a given input materials representation in the presence of big data.

Deep convolutional neural networks, assisted by architectural design strategies, make extensive use of data augmentation techniques and layers with a high number of feature maps to embed object transformations. That is highly inefficient and for large datasets implies a massive redundancy of features detectors. Even though capsules networks are still in their infancy, they constitute a promising solution to extend current convolutional networks and endow artificial visual perception with a process to encode more efficiently all feature affine transformations. Indeed, a properly working capsule network should theoretically achieve higher results with a considerably lower number of parameters count due to intrinsic capability to generalize to novel viewpoints. Nevertheless, little attention has been given to this relevant aspect. In this paper, we investigate the efficiency of capsule networks and, pushing their capacity to the limits with an extreme architecture with barely 160K parameters, we prove that the proposed architecture is still able to achieve state-of-the-art results on three different datasets with only 2% of the original CapsNet parameters. Moreover, we replace dynamic routing with a novel non-iterative, highly parallelizable routing algorithm that can easily cope with a reduced number of capsules. Extensive experimentation with other capsule implementations has proved the effectiveness of our methodology and the capability of capsule networks to efficiently embed visual representations more prone to generalization., Accepted by Scientific Reports

Early detection and appropriate medical treatment are of great use for ear disease. However, a new diagnostic strategy is necessary for the absence of experts and relatively low diagnostic accuracy, in which deep learning plays an important role. This paper puts forward a mechanic learning model which uses abundant otoscope image data gained in clinical cases to achieve an automatic diagnosis of ear diseases in real time. A total of 20,542 endoscopic images were employed to train nine common deep convolution neural networks. According to the characteristics of the eardrum and external auditory canal, eight kinds of ear diseases were classified, involving the majority of ear diseases, such as normal, Cholestestoma of the middle ear, Chronic suppurative otitis media, External auditory cana bleeding, Impacted cerumen, Otomycosis external, Secretory otitis media, Tympanic membrane calcification. After we evaluate these optimization schemes, two best performance models are selected to combine the ensemble classifiers with real-time automatic classification. Based on accuracy and training time, we choose a transferring learning model based on DensNet-BC169 and DensNet-BC1615, getting a result that each model has obvious improvement by using these two ensemble classifiers, and has an average accuracy of 95.59%. Considering the dependence of classifier performance on data size in transfer learning, we evaluate the high accuracy of the current model that can be attributed to large databases. Current studies are unparalleled regarding disease diversity and diagnostic precision. The real-time classifier trains the data under different acquisition conditions, which is suitable for real cases. According to this study, in the clinical case, the deep learning model is of great use in the early detection and remedy of ear diseases.

The DNA replication influences the inheritance of genetic information in the DNA life cycle. As the distribution of replication origins (ORIs) is the major determinant to precisely regulate the replication process, the correct identification of ORIs is significant in giving an insightful understanding of DNA replication mechanisms and the regulatory mechanisms of genetic expressions. For eukaryotes in particular, multiple ORIs exist in each of their gene sequences to complete the replication in a reasonable period of time. To simplify the identification process of eukaryote’s ORIs, most of existing methods are developed by traditional machine learning algorithms, and target to the gene sequences with a fixed length. Consequently, the identification results are not satisfying, i.e. there is still great room for improvement. To break through the limitations in previous studies, this paper develops sequence segmentation methods, and employs the word embedding technique, ‘Word2vec’, to convert gene sequences into word vectors, thereby grasping the inner correlations of gene sequences with different lengths. Then, a deep learning framework to perform the ORI identification task is constructed by a convolutional neural network with an embedding layer. On the basis of the analysis of similarity reduction dimensionality diagram, Word2vec can effectively transform the inner relationship among words into numerical feature. For four species in this study, the best models are obtained with the overall accuracy of 0.975, 0.765, 0.885, 0.967, the Matthew’s correlation coefficient of 0.940, 0.530, 0.771, 0.934, and the AUC of 0.975, 0.800, 0.888, 0.981, which indicate that the proposed predictor has a stable ability and provide a high confidence coefficient to classify both of ORIs and non-ORIs. Compared with state-of-the-art methods, the proposed predictor can achieve ORI identification with significant improvement. It is therefore reasonable to anticipate that the proposed method will make a useful high throughput tool for genome analysis.

Coronary calcifications are an obstacle for successful percutaneous treatment of coronary artery disease patients. The optimal method for delineating calcifications extent is coronary optical coherence tomography (OCT). To identify calcification on OCT and subsequently tailor the appropriate treatment, requires expertise in both image acquisition and interpretation. Image acquisition consists from system calibration, blood clearance by a contrast agent along with synchronization of the pullback process. Accurate interpretation demands careful review by the operator of a segment of 50–75 mm of the coronary vessel at steps of 5–10 frames per mm accounting for 375–540 images in each OCT run, which is time consuming and necessitates some expertise in OCT analysis. In this paper we developed a new deep learning algorithm to assist the physician to identify and quantify coronary calcifications promptly, efficiently and accurately. Our algorithm achieves an accuracy of 0.9903 ± 0.009 over the test set at size of 1500 frames and even managed to find calcifications that were not recognized manually by the physician. For the best knowledge of the authors our algorithm achieves high accuracy which was never achieved in the past.

Transfer learning has been shown to be effective in many applications in which training data for the target problem are limited but data for a related (source) problem are abundant. In this paper, we apply transfer learning to the prediction of anti-cancer drug response. Previous transfer learning studies for drug response prediction focused on building models that predict the response of tumor cells to a specific drug treatment. We target the more challenging task of building general prediction models that can make predictions for both new tumor cells and new drugs. We apply the classic transfer learning framework that trains a prediction model on the source dataset and refines it on the target dataset, and extends the framework through ensemble. The ensemble transfer learning pipeline is implemented using LightGBM and two deep neural network (DNN) models with different architectures. Uniquely, we investigate its power for three application settings including drug repurposing, precision oncology, and new drug development, through different data partition schemes in cross-validation. We test the proposed ensemble transfer learning on benchmark in vitro drug screening datasets, taking one dataset as the source domain and another dataset as the target domain. The analysis results demonstrate the benefit of applying ensemble transfer learning for predicting anti-cancer drug response in all three applications with both LightGBM and DNN models. Compared between the different prediction models, a DNN model with two subnetworks for the inputs of tumor features and drug features separately outperforms LightGBM and the other DNN model that concatenates tumor features and drug features for input in the drug repurposing and precision oncology applications. In the more challenging application of new drug development, LightGBM performs better than the other two DNN models.

Diabetic macular edema (DME) and retina vein occlusion (RVO) are macular diseases in which central photoreceptors are affected due to pathological accumulation of fluid. Optical coherence tomography allows to visually assess and evaluate photoreceptor integrity, whose alteration has been observed as an important biomarker of both diseases. However, the manual quantification of this layered structure is challenging, tedious and time-consuming. In this paper we introduce a deep learning approach for automatically segmenting and characterising photoreceptor alteration. The photoreceptor layer is segmented using an ensemble of four different convolutional neural networks. En-face representations of the layer thickness are produced to characterize the photoreceptors. The pixel-wise standard deviation of the score maps produced by the individual models is also taken to indicate areas of photoreceptor abnormality or ambiguous results. Experimental results showed that our ensemble is able to produce results in pair with a human expert, outperforming each of its constitutive models. No statistically significant differences were observed between mean thickness estimates obtained from automated and manually generated annotations. Therefore, our model is able to reliable quantify photoreceptors, which can be used to improve prognosis and managment of macular diseases.

Recent advancements in deep learning have revolutionized the way microscopy images of cells are processed. Deep learning network architectures have a large number of parameters, thus, in order to reach high accuracy, they require a massive amount of annotated data. A common way of improving accuracy builds on the artificial increase of the training set by using different augmentation techniques. A less common way relies on test-time augmentation (TTA) which yields transformed versions of the image for prediction and the results are merged. In this paper we describe how we have incorporated the test-time argumentation prediction method into two major segmentation approaches utilized in the single-cell analysis of microscopy images. These approaches are semantic segmentation based on the U-Net, and instance segmentation based on the Mask R-CNN models. Our findings show that even if only simple test-time augmentations (such as rotation or flipping and proper merging methods) are applied, TTA can significantly improve prediction accuracy. We have utilized images of tissue and cell cultures from the Data Science Bowl (DSB) 2018 nuclei segmentation competition and other sources. Additionally, boosting the highest-scoring method of the DSB with TTA, we could further improve prediction accuracy, and our method has reached an ever-best score at the DSB.

Multivariate time series data in practical applications, such as health care, geoscience, and biology, are characterized by a variety of missing values. In time series prediction and other related tasks, it has been noted that missing values and their missing patterns are often correlated with the target labels, a.k.a., informative missingness. There is very limited work on exploiting the missing patterns for effective imputation and improving prediction performance. In this paper, we develop novel deep learning models, namely GRU-D, as one of the early attempts. GRU-D is based on Gated Recurrent Unit (GRU), a state-of-the-art recurrent neural network. It takes two representations of missing patterns, i.e., masking and time interval, and effectively incorporates them into a deep model architecture so that it not only captures the long-term temporal dependencies in time series, but also utilizes the missing patterns to achieve better prediction results. Experiments of time series classification tasks on real-world clinical datasets (MIMIC-III, PhysioNet) and synthetic datasets demonstrate that our models achieve state-of-the-art performance and provide useful insights for better understanding and utilization of missing values in time series analysis.

Translating the vast data generated by genomic platforms into accurate predictions of clinical outcomes is a fundamental challenge in genomic medicine. Many prediction methods face limitations in learning from the high-dimensional profiles generated by these platforms, and rely on experts to hand-select a small number of features for training prediction models. In this paper, we demonstrate how deep learning and Bayesian optimization methods that have been remarkably successful in general high-dimensional prediction tasks can be adapted to the problem of predicting cancer outcomes. We perform an extensive comparison of Bayesian optimized deep survival models and other state of the art machine learning methods for survival analysis, and describe a framework for interpreting deep survival models using a risk backpropagation technique. Finally, we illustrate that deep survival models can successfully transfer information across diseases to improve prognostic accuracy. We provide an open-source software implementation of this framework called SurvivalNet that enables automatic training, evaluation and interpretation of deep survival models.

Modeling in-vivo protein-DNA binding is not only fundamental for further understanding of the regulatory mechanisms, but also a challenging task in computational biology. Deep-learning based methods have succeed in modeling in-vivo protein-DNA binding, but they often (1) follow the fully supervised learning framework and overlook the weakly supervised information of genomic sequences that a bound DNA sequence may has multiple TFBS(s), and, (2) use one-hot encoding to encode DNA sequences and ignore the dependencies among nucleotides. In this paper, we propose a weakly supervised framework, which combines multiple-instance learning with a hybrid deep neural network and uses k-mer encoding to transform DNA sequences, for modeling in-vivo protein-DNA binding. Firstly, this framework segments sequences into multiple overlapping instances using a sliding window, and then encodes all instances into image-like inputs of high-order dependencies using k-mer encoding. Secondly, it separately computes a score for all instances in the same bag using a hybrid deep neural network that integrates convolutional and recurrent neural networks. Finally, it integrates the predicted values of all instances as the final prediction of this bag using the Noisy-and method. The experimental results on in-vivo datasets demonstrate the superior performance of the proposed framework. In addition, we also explore the performance of the proposed framework when using k-mer encoding, and demonstrate the performance of the Noisy-and method by comparing it with other fusion methods, and find that adding recurrent layers can improve the performance of the proposed framework.