Search

Medical imaging is an essential medical diagnosis system subsequently integrated with artificial intelligence for assistance in clinical diagnosis. The actual medical images acquired during the image capturing procedures generate poor quality images as a result of numerous physical restrictions of the imaging equipment and time constraints. Recently, medical image super-resolution (SR) has emerged as an indispensable research subject in the community of image processing to address such limitations. SR is a classical computer vision operation that attempts to restore a visually sharp high-resolution images from the degraded low-resolution images. In this study, an effective medical super-resolution approach based on weighted least squares optimisation via multiscale convolutional neural networks (CNNs) has been proposed for lesion localisation. The weighted least squares optimisation strategy that particularly is well-suited for progressively coarsening the original images and simultaneously extract multiscale information has been executed. Subsequently, a SR model by training CNNs based on wavelet analysis has been designed by carrying out wavelet decomposition of optimized images for multiscale representations. Then multiple CNNs have been trained separately to approximate the wavelet multiscale representations. The trained multiple convolutional neural networks characterize medical images in many directions and multiscale frequency bands, and thus facilitate image restoration subject to increased number of variations depicted in different dimensions and orientations. Finally, the trained CNNs regress wavelet multiscale representations from a LR medical images, followed by wavelet synthesis that forms a reconstructed HR medical image. The experimental performance indicates that the proposed model SR restoration approach achieve superior SR efficiency over existing comparative methods

Extreme learning machine (ELM) is a training algorithm for single hidden layer feedforward neural network (SLFN), which converges much faster than traditional methods and yields promising performance. In this paper, we hope to present a comprehensive review on ELM. Firstly, we will focus on the theoretical analysis including universal approximation theory and generalization. Then, the various improvements are listed, which help ELM works better in terms of stability, efficiency, and accuracy. Because of its outstanding performance, ELM has been successfully applied in many real-time learning tasks for classification, clustering, and regression. Besides, we report the applications of ELM in medical imaging: MRI, CT, and mammogram. The controversies of ELM were also discussed in this paper. We aim to report these advances and find some future perspectives.

Most intelligent algorithms used for recognizing epilepsy electroencephalogram (EEG) have two major deficiencies. The one is the lack of interpretability and the other is unsatisfactory recognition results. In response to these challenges, we propose a dedicated model called multi-view Takagi–Sugeno–Kang (TSK) fuzzy system (MV-TSK-FS) for the epilepsy EEG detection. Our contributions lie in three aspects. First, TSK-FS is selected as the basic model. As one of the most famous fuzzy systems, TSK-FS has the advantage of nice interpretability and thus meets the requirement of clinic trials and applications. Second, MV-TSK-FS uses a multi-view framework to collaboratively handle the collective feature data extracted from diverse extraction perspectives, which strives to avoid the potential performance degradation commonly incurred with single feature extraction. Third, we propose a view-weighted mechanism based on the quadratic regularization to distinguish the importance of each view. The more important the view, the larger the corresponding weight is. The final decision is consequently figured out with the weighted outputs of all views. Experimental results demonstrate that, compared with other epilepsy EEG detection ones, our proposed method has better classification performance as well as more satisfied interpretability on results.

In the clinical study of Alzheimer’s Disease (AD) with neuroimaging data, it is challenging to identify the progressive Mild Cognitive Impairment (pMCI) subjects from the stableMCI (sMCI) subjects (i.e., the pMCI/sMCI classification) in an individual level because of small inter-group differences between two groups (i.e., pMCIs and sMCIs) as well as high intra-group variations within each group. Moreover, there are a very limited number of subjects available, which cannot guarantee to find informative and discriminative patterns for achieving high diagnostic accuracy. In this paper, we propose a novel sparse regression method to fuse the auxiliary data into the predictor data for the pMCI/sMCI classification, where the predictor data is structural Magnetic Resonance Imaging (MRI) information of both pMCI and sMCI subjects and the auxiliary data includes the ages of the subjects, the Positron Emission Tomography (PET) information of the predictor data, and the structural MRI information of AD and Normal Controls (NC). Specifically, we incorporate the auxiliary data and the predictor data into a unified framework to jointly achieve the following objectives: i) jointly selecting informative features from both the auxiliary data and the predictor data; ii) robust to outliers from both the auxiliary data and the predictor data; and iii) reducing the aging effect due to the possible cause of brain atrophy induced by both the normal aging and the disease progression. As a result, our proposed method jointly selects the useful features from the auxiliary data and the predictor data by taking into account the influence of outliers and the age of the two kinds of data, i.e., the pMCI and sMCI subjects as well as the AD and NC subjects. We further employ the linear Support Vector Machine (SVM) with the selected features of the predictor data to conduct the pMCI/sMCI classification. Experimental results on the public data of Alzheimer’s Disease Neuroimaging Initiative (ADNI) show the proposed method achieved the best classification performance, compared to the best comparison method, in terms of four evaluation metrics.

Alzheimer’s disease (AD) will become a global burden in the coming decades according to the latest statistical survey. How to effectively detect AD or MCI (mild cognitive impairment) using reliable biomarkers and robust machine learning methods has become a challenging problem. In this study, we propose a novel AD multiclass classification framework with embedding feature selection and fusion based on multimodal neuroimaging. The framework has three novel aspects: (1) An l 2 , 1 -norm regularization term combined with the multiclass hinge loss is used to naturally select features across all the classes in each modality. (2) To fuse the complementary information contained in each modality, an l p -norm ( 1 p ∞ ) regularization term is introduced to combine different kernels to perform multiple kernel learning to avoid a sparse kernel coefficient distribution, thereby effectively exploiting complementary modalities. (3) A theorem that transforms the multiclass hinge loss minimization problem using the l 2 , 1 -norm and l p -norm regularizations to a previous solvable optimization problem and its proof are given. Additionally, it is theoretically proved that the optimization process converges to the global optimum. Extensive comparison experiments and analysis support the promising performance of the proposed method.

COVID-19 is an ongoing pandemic disease. To make more accurate diagnosis on COVID-19 than existing approaches, this paper proposed a novel method combining DenseNet and optimization of transfer learning setting (OTLS) strategy. Preprocessing was used to enhance, crop, and resize the collected chest CT images. Data augmentation method was used to increase the size of training set. A composite learning factor (CLF) was employed which assigned different learning factor to three types of layers: frozen layers, middle layers, and new layers. Meanwhile, the OTLS strategy was proposed. Finally, precomputation method was utilized to reduce RAM storage and accelerate the algorithm. We observed that optimization setting "201-IV" can achieve the best performance by proposed OTLS strategy. The sensitivity, specificity, precision, and accuracy of our proposed method were 96.35 ± 1.07, 96.25 ± 1.16, 96.29 ± 1.11, and 96.30 ± 0.56, respectively. The proposed DenseNet-OTLS method achieved better performances than state-of-the-art approaches in diagnosing COVID-19.

Computer aided diagnosis systems are playing an important part in clinical treatment. They can help the doctors and physicians to verify the diagnosis decisions. In this study, a new classification algorithm for the brain magnetic resonance image is proposed. Initially, we utilized a MobileNetV2 to extract features from the input brain images, which was pre-trained on ImageNet dataset. Instead of training the deep network, we simply calculate the output of its certain layer to form the feature vector. The optimal feature layer is obtained by the experiment. Then, three different feedforward networks: extreme learning machine, Schmidt neural network and random vector functional-link net, are trained for classification. Chaotic bat algorithm was proposed to optimize the weights and biases in the three randomized neural networks to boost their classification accuracy. The result from 5×hold-out validation reveals that our method can achieve good generalization performance which is comparable to state-of-the-art pathological brain detection methods. The trained model can serve as a visual question answering system and produce accurate results.

Compared with common deep learning methods (e.g., convolutional neural networks), transfer learning is characterized by simplicity, efficiency and its low training cost, breaking the curse of small datasets. Medical image analysis plays an indispensable role in both scientific research and clinical diagnosis. Common medical image acquisition methods include Computer Tomography (CT), Magnetic Resonance Imaging (MRI), Ultrasound (US), X-Ray, etc. Although these medical imaging methods can be applied for non-invasive qualitative and quantitative analysis of patients—compared with image datasets in other computer vision fields such like faces—medical images, especially its labeling, is still scarce and insufficient. Therefore, more and more researchers adopted transfer learning for medical image processing. In this study, after reviewing one hundred representative papers from IEEE, Elsevier, Google Scholar, Web of Science and various sources published from 2000 to 2020, a comprehensive review is presented, including (i) structure of CNN, (ii) background knowledge of transfer learning, (iii) different types of strategies performing transfer learning, (iv) application of transfer learning in various sub-fields of medical image analysis, and (v) discussion on the future prospect of transfer learning in the field of medical image analysis. Through this review paper, beginners could receive an overall and systematic knowledge of transfer learning application in medical image analysis. And policymaker of related realm will benefit from the summary of the trend of transfer learning in medical imaging field and may be encouraged to make policy positive to the future development of transfer learning in the field of medical image analysis.

As two different modalities of medical images, Magnetic Resonance (MR) and Computer Tomography (CT), provide mutually-complementary information to doctors in clinical applications. However, to obtain both images sometimes is cost-consuming and unavailable, particularly for special populations. For example, patients with metal implants are not suitable for MR scanning. Also, it is probably infeasible to acquire multi-contrast MR images during once clinical scanning. In this context, to synthesize needed MR images for patients whose CT images are available becomes valuable. To this end, we present a novel generative network, called CAE-ACGAN, which incorporates the advantages of Variational Auto-Encoder (VAE) and Generative Adversarial Network (GAN) with an auxiliary discriminative classifier network. We apply this network to synthesize multi-contrast MR images from single CT and conduct experiments on brain datasets. Our main contributions can be summarized as follows: 1)We alleviate the problems of images blurriness and mode collapse by integrating the advantages of VAE and GAN; 2) We solve the complicated cross-domain, multi-contrast MR synthesis task using the proposed network; 3) The technique of random-extraction-patches is used to lower the limit of insufficient training data, enabling to obtain promising results even with limited available data; 4) By comparing with other typical networks, we are able to yield nearer-real, higher-quality synthetic MR images, demonstrating the effectiveness and stability of our proposed network.

Melanoma skin cancer is one of the most common diseases in the world. It is essential to diagnose melanoma at an early stage. Visual inspection during the medical examination of skin lesions is not a simple task, as there is a similarity between lesions. Also, medical experience and disposition can result in inaccurate diagnoses. Technologies such as the Internet of Things (IoT) have helped to create effective health systems. Doctors can use them anywhere, with the guarantee that more people can be diagnosed without prejudice to subjective factors. Transfer Learning and Deep Learning are increasingly significant in the clinical diagnosis of different diseases. This work proposes the use of Transfer Learning and Deep Learning in an IoT system to assist doctors in the diagnosis of common skin lesions, typical nevi, and melanoma. This work uses Convolutional Neural Networks (CNNs) as resource extractors. The CNN models used were: Visual Geometry Group (VGG), Inception, Residual Networks (ResNet), Inception-ResNet, Extreme Inception (Xception), MobileNet, Dense Convolutional Network (DenseNet), and Neural Architecture Search Network (NASNet). For the classification of injuries, the Bayes, Support Vector Machines (SVM), Random Forest (RF), Perceptron Multilayer (MLP), and the K-Nearest Neighbors (KNN) classifiers are used. This study used two datasets: the first provided by the International Skin Imaging Collaboration (ISIC) at the International Biomedical Imaging Symposium (ISBI); the second is PH2. For ISBI-ISIC, this study examined lesions between nevi and melanomas. In PH2, this work analyzed the diagnosis based on lesions of common nevus, atypical nevi, and melanomas. The DenseNet201 extraction model, combined with the KNN classifier achieved an accuracy of 96.805% for the ISBI-ISIC dataset and 93.167% for the PH2. Thus, an approach focused on the IoT system is reliable and efficient for doctors who assist in the diagnosis of skin lesions.

Traditional technology uses serial processing method in CT image feature extraction. It is prone to loss of image data, which causes problems such as ring distortion of the reconstructed image and long reconstruction time. Therefore, a three-dimensional (3D) reconstruction algorithm for CT image features based on multi-threaded deep learning calculation is designed. Feature image textures are segmented using parameters such as gray, mean, and variance. Fusion calculates co-occurrence logarithm of the gray level with segmented sub-block and its pixels. Co-occurrence probability of the sub-block gray levels is obtained, followed by getting optimal feature volume data, which is stored in 3D texture and 1D texture for interpolation calculation.The optimal feature volume data is stored in 3D and 1D texture for interpolation calculation. Rotation matrix is stored in the global storage space of CUDA, which is used for multi-threaded calculations. After completing multi-threaded batch CT image hardware configuration and algorithm flow settings, the thread's index and bounding box is calculated. 3D reconstruction of CT image features is achieved by model accumulation. The experimental simulation proves that local detail information loss is small after reconstruction of the proposed method. Its reconstruction time is short and has good applicability.

Computer-aided diagnosis system is becoming a more and more important tool in clinical treatment, which can provide a verification of the doctors’ decisions. In this paper, we proposed a novel abnormal brain detection method for magnetic resonance image. Firstly, a pre-trained AlexNet was modified with batch normalization layers and trained on our brain images. Then, the last several layers were replaced with an extreme learning machine. A searching method was proposed to find the best number of layers to be replaced. Finally, the extreme learning machine was optimized by chaotic bat algorithm to obtain better classification performance. Experiment results based on 5 × hold-out validation revealed that our method achieved state-of-the-art performance.

As an important component of universal sign language and the basis of other sign language learning, finger sign language is of great significance. This paper proposed a novel fingerspelling identification method for Chinese Sign Language via AlexNet-based transfer learning and Adam optimizer, which tested four different configurations of transfer learning. Besides, in the experiment, Adam algorithm was compared with stochastic gradient descent with momentum (SGDM) and root mean square propagation (RMSProp) algorithms, and comparison of using data augmentation (DA) against not using DA was executed to pursue higher performance. Finally, the best accuracy of 91.48% and average accuracy of 89.48 ± 1.16% were yielded by configuration M1 (replacing the last FCL8) with Adam algorithm and using 181x DA, which indicates that our method can identify Chinese finger sign language effectively and stably. Meanwhile, the proposed method is superior to other five state-of-the-art approaches.

Gliomas are characterized by high morbidity and high mortality in primary tumors. The identification of glioma type is helpful for radiologists to facilitate correct medical judgments and better prognosis for patients. In order to avoid harm to patients caused by a biopsy, radiologists attempt to classify Magnetic Resonance Images(MRI) using deep learning methods. In the present paper, we propose a deep learning convolutional neural network ResNet based on the pyramid dilated convolution for Gliomas classification. The pyramid dilated convolution is integrated into the bottom of Resnet to increase the receptive field of the original network and improve the classification accuracy. After adding the pyramid dilated convolution model, the receptive field of the original network underlying convolution was improved. A clinical dataset is used to test the pyramid dilated convolution ResNet neural network model proposed in this paper. The experimental results demonstrate that the proposed method can effectively improve glioma classification performance.

( Aim ) Multiple sclerosis is a neurological condition that may cause neurologic disability. Convolutional neural network can achieve good results, but tuning hyperparameters of CNN needs expert knowledge and are difficult and time-consuming. To identify multiple sclerosis more accurately, this article proposed a new transfer-learning-based approach. ( Method ) DenseNet-121, DenseNet-169, and DenseNet-201 neural networks were compared. In addition, we proposed the use of a composite learning factor (CLF) that assigns different learning factor to three types of layers: early frozen layers, middle layers, and late replaced layers. How to allocate layers into those three layers remains a problem. Hence, four transfer learning settings (viz., Settings A, B, C, and D) were tested and compared. A precomputation method was utilized to reduce the storage burden and accelerate the program. ( Results ) We observed that DenseNet-201-D (the layers from CP to T3 are frozen, the layers of D4 are updated with learning factor of 1, and the final new layers of FCL are randomly initialized with learning factor of 10) can achieve the best performance. The sensitivity, specificity, and accuracy of DenseNet-201-D was 98.27± 0.58, 98.35± 0.69, and 98.31± 0.53, respectively. ( Conclusion ) Our method gives better performances than state-of-the-art approaches. Furthermore, this composite learning rate gives superior results to traditional simple learning factor (SLF) strategy.

Based on deep learning technology, this paper proposes a two-stage colorectal image feature mining and fast recognition model to achieve fully automatic medical image pathology discrimination. Drawing on the ideas of multi-factor Meta-regression analysis widely used in the medical field and the model aggregation framework based on Bayesian prior probability theory, a prognostic model of colorectal tumors suitable for various situations and scenarios is constructed. And using a combination of public data sets and real data sets, design two sets of experiments to verify these models from different angles. The algorithm was used to select one, four, and five related features from three sequences to construct three sets of prediction models. The application of the six algorithms failed to obtain a better predictive model (AUC value range 0.439 ~ 0.640). The algorithm (AUC value 0.750± 0.137) and the algorithm (AUC value 0.764± 0.128) can be used to obtain models with better predictive performance, and the four models are less effective (AUC value

A fully automated design is proposed in this work employing optimal deep learning features for classifying gastrointestinal infections. Here, three prominent infections– ulcer, bleeding, polyp and a healthy class are considered as class labels. In the initial stage, the contrast is improved by fusing bi-directional histogram equalization with top-hat filtering output. The resultant fusion images are then passed to ResNet101 pre-trained model and trained once again using deep transfer learning. However, there are challenges involved in extracting deep learning features including impertinent information and redundancy. To mitigate this problem, we took advantage of two metaheuristic algorithms– Enhanced Crow Search and Differential Evolution. These algorithms are implemented in parallel to obtain optimal feature vectors. Following this, a maximum correlation-based fusion approach is applied to fuse optimal vectors from the previous step to obtain an enhanced vector. This final vector is given as input to Extreme Learning Machine (ELM) classifier for final classification. The proposed method is evaluated on a combined database. It accomplished an accuracy of 99.46%, which shows significant improvement over preceding techniques and other neural network architectures.

Glomerulosclerosis is a pathomorphological feature of glomerular lesions. Early detection, accurate judgement and effective prevention of the glomeruli is crucial not only for people with kidney disease, but also for the general population. We proposed a method in combination of traditional image analysis with modern machine learning diagnosis system model based on GoogLeNet for recognizing and distinguishing different categories of glomerulus in order to efficiently capture the important structures as well as to minimize manual effort and supervision. We proposed a novel deep learning model based on GoogLeNet with added batch-normalization layers to extract useful features and subsequently entered the features into SoftMax for classification. We also incorporated Bayesian Optimization algorithm and k-fold cross validation in this system for achieving a more reliable result. Our method has eventually achieved an overall accuracy of 95.04±4.99%, and F1 score of 94.44±3.11% for no glomerulus category, 96.73±5.23% for normal glomerulus category and 93.66±7.82% for globally sclerosed glomerulus category, which means this method can accurately determine the degree of glomerulosclerosis with little supervision. The experimental result also shows that this method has better performance when compared with other state-of art methods.

Diffusion tensor imaging (DTI) is a non-invasive magnetic resonance imaging technique and a special type of magnetic resonance imaging, which has been widely used to study the diffusion process in the brain. The signal-to-noise ratio of DTI data is relatively low, the shape and direction of the noisy tensor data are destroyed. This limits the development of DTI in clinical applications. In order to remove the Rician noise and preserve the diffusion tensor geometry of DTI, we propose a DTI denoising algorithm based on Riemann nonlocal similarity. Firstly, DTI tensor is mapped to the Riemannian manifold to preserve the structural properties of the tensor. The Riemann similarity measure is used to search for non-local similar blocks to form similar patch groups. Then the Gaussian mixture model is used to learn the prior distribution of patch groups. Finally, the noisy patch group is denoised by Bayesian inference and the denoised patch group is reconstructed to obtain the final denoised image. The denoising experiments of real and simulated DTI data are carried out to verify the feasibility and effectiveness of the proposed algorithm. The experimental results show that our algorithm not only effectively removes the Rician noise in the DTI image, but also preserves the nonlinear structure of the DTI image. Comparing to the existing denoising algorithms, our algorithm has better improvement of the principal diffusion direction, lower absolute error of fractional anisotropy and higher peak signal-to-noise ratio.

Magnetic resonance imaging (MRI), which assists doctors in determining clinical staging and expected surgical range, has high medical value. A large number of MRI images require a large amount of storage space and the transmission bandwidth of the PACS system in offline storage and remote diagnosis. Therefore, high-quality compression of MRI images is very research-oriented. Current compression methods for MRI images with high compression ratio cause loss of information on lesions, leading to misdiagnosis; compression methods for MRI images with low compression ratio does not achieve the desired effect. Therefore, a fast fractal-based compression algorithm for MRI images is proposed in this paper. First, three-dimensional (3D) MRI images are converted into a two-dimensional (2D) image sequence, which facilitates the image sequence based on the fractal compression method. Then, range and domain blocks are classified according to the inherent spatiotemporal similarity of 3D objects. By using self-similarity, the number of blocks in the matching pool is reduced to improve the matching speed of the proposed method. Finally, a residual compensation mechanism is introduced to achieve compression of MRI images with high decompression quality. The experimental results show that compression speed is improved by 2-3 times, and the PSNR is improved by nearly 10. It indicates the proposed algorithm is effective and solves the contradiction between high compression ratio and high quality of MRI medical images.

Due to the space inconsistency between benchmark image and segmentation result in many existing semantic segmentation algorithms for abdominal CT images, an improved model based on the basic framework of DeepLab-v3 is proposed, and Pix2pix network is introduced as the generation adversarial model. Our proposed model realizes the segmentation framework combining deep feature with multi-scale semantic feature. In order to improve the generalization ability and training accuracy of the model, this paper proposes a combination of the traditional multi-classification cross-entropy loss function with the content loss function of generator output and the adversarial loss function of discriminator output. A large number of qualitative and quantitative experimental results show that the performance of our proposed semantic segmentation algorithm is better than the existing algorithms, and can improve the segmentation efficiency while ensuring the space consistency of the semantics segmentation for abdominal CT images.

We propose a novel graph rank-based average pooling neural network (GRAPNN) to detect secondary pulmonary tuberculosis patients via chest CT imaging. First, we propose a novel rank-based pooling neural network (RAPNN) to learn the individual image-level features from chest CT images. Second, we integrate the graph convolutional network (GCN), which learns relation-aware representation among the batch of chest CT images, to RAPNN. Third, we build a novel Graph RAPNN (GRAPNN) model based on the previous integration via k-means clustering and k-nearest neighbors’ algorithm. Besides, an improved data augmentation is utilized to handle overfitting problem. Grad-ACM is used to make this GRAPNN model explainable. This proposed GRAPNN method is compared with seven state-of-the-art algorithms. The results showed GRAPNN model yields the best performances with a sensitivity of 94.65%, a specificity of 95.12%, a precision of 95.17%, an accuracy of 94.88%, and an F1 score of 94.87%. Our GRAPNN is superior to other seven state-of-the-art approaches. The explainable mechanism in our method can identify the lesions of important lung parts (tuberculosis cavities and surrounding small lesions) for transparent decision.

Aim. COVID-19 has caused large death tolls all over the world. Accurate diagnosis is of significant importance for early treatment. Methods. In this study, we proposed a novel PSSPNN model for classification between COVID-19, secondary pulmonary tuberculosis, community-captured pneumonia, and healthy subjects. PSSPNN entails five improvements: we first proposed the n-conv stochastic pooling module. Second, a novel stochastic pooling neural network was proposed. Third, PatchShuffle was introduced as a regularization term. Fourth, an improved multiple-way data augmentation was used. Fifth, Grad-CAM was utilized to interpret our AI model. Results. The 10 runs with random seed on the test set showed our algorithm achieved a microaveraged F1 score of 95.79%. Moreover, our method is better than nine state-of-the-art approaches. Conclusion. This proposed PSSPNN will help assist radiologists to make diagnosis more quickly and accurately on COVID-19 cases.

Highlights • A high-performance pneumonia detection system was proposed in this paper. • Transfer learning is used to obtain feature extractor. • A novel graph-based feature reconstruction method was proposed. • The proposed feature reconstruction is efficient yet transplantable to other scenarios., Pneumonia is a global disease that causes high children mortality. The situation has even been worsening by the outbreak of the new coronavirus named COVID-19, which has killed more than 983,907 so far. People infected by the virus would show symptoms like fever and coughing as well as pneumonia as the infection progresses. Timely detection is a public consensus achieved that would benefit possible treatments and therefore contain the spread of COVID-19. X-ray, an expedient imaging technique, has been widely used for the detection of pneumonia caused by COVID-19 and some other virus. To facilitate the process of diagnosis of pneumonia, we developed a deep learning framework for a binary classification task that classifies chest X-ray images into normal and pneumonia based on our proposed CGNet. In our CGNet, there are three components including feature extraction, graph-based feature reconstruction and classification. We first use the transfer learning technique to train the state-of-the-art convolutional neural networks (CNNs) for binary classification while the trained CNNs are used to produce features for the following two components. Then, by deploying graph-based feature reconstruction, we, therefore, combine features through the graph to reconstruct features. Finally, a shallow neural network named GNet, a one layer graph neural network, which takes the combined features as the input, classifies chest X-ray images into normal and pneumonia. Our model achieved the best accuracy at 0.9872, sensitivity at 1 and specificity at 0.9795 on a public pneumonia dataset that includes 5,856 chest X-ray images. To evaluate the performance of our proposed method on detection of pneumonia caused by COVID-19, we also tested the proposed method on a public COVID-19 CT dataset, where we achieved the highest performance at the accuracy of 0.99, specificity at 1 and sensitivity at 0.98, respectively.

Aim. This study proposes a new artificial intelligence model based on cardiovascular computed tomography for more efficient and precise recognition of Tetralogy of Fallot (TOF). Methods. Our model is a structurally optimized stochastic pooling convolutional neural network (SOSPCNN), which combines stochastic pooling, structural optimization, and convolutional neural network. In addition, multiple-way data augmentation is used to overcome overfitting. Grad-CAM is employed to provide explainability to the proposed SOSPCNN model. Meanwhile, both desktop and web apps are developed based on this SOSPCNN model. Results. The results on ten runs of 10-fold crossvalidation show that our SOSPCNN model yields a sensitivity of 92.25 ± 2.19 , a specificity of 92.75 ± 2.49 , a precision of 92.79 ± 2.29 , an accuracy of 92.50 ± 1.18 , an F1 score of 92.48 ± 1.17 , an MCC of 85.06 ± 2.38 , an FMI of 92.50 ± 1.17 , and an AUC of 0.9587. Conclusion. The SOSPCNN method performed better than three state-of-the-art TOF recognition approaches.

To more efficiently diagnose secondary pulmonary tuberculosis, we build an improved convolutional neural network (ICNN) based on recent deep learning technologies. First, a 12-way data augmentation (DA-12) was proposed to increase size of training set. Second, stochastic pooling was introduced to replace the standard average pooling and max pooling. Third, batch normalization and dropout techniques were included and associated with conv layers and fully-connected layers, respectively. Fourth, a dynamic learning rate was employed to replace traditional fixed learning rate. Fifth, hyperparameter optimization was used to optimize the number of layers within proposed network. Our eight-layer ICNN demonstrated excellent results on the test set, yielding a sensitivity of 94.19%, a specificity of 93.72%, and an accuracy of 93.95%. Our ICNN provides better performances than other four state-of-the-art algorithms. It can help radiologists to make more accurate diagnosis on secondary pulmonary tuberculosis.

Highlights • We proposed a novel (L, 2) transfer feature learning (L2TFL) approach. • L2TFL can elucidate the optimal layers to be removed prior to selection. • We developed a novel selection algorithm of pretrained network for fusion approach. • SAPNF can determine the best two pretrained models for fusion. • We introduced a deep CCT fusion discriminant correlation analysis fusion method., Aim : COVID-19 is a disease caused by a new strain of coronavirus. Up to 18th October 2020, worldwide there have been 39.6 million confirmed cases resulting in more than 1.1 million deaths. To improve diagnosis, we aimed to design and develop a novel advanced AI system for COVID-19 classification based on chest CT (CCT) images. Methods : Our dataset from local hospitals consisted of 284 COVID-19 images, 281 community-acquired pneumonia images, 293 secondary pulmonary tuberculosis images; and 306 healthy control images. We first used pretrained models (PTMs) to learn features, and proposed a novel (L, 2) transfer feature learning algorithm to extract features, with a hyperparameter of number of layers to be removed (NLR, symbolized as L). Second, we proposed a selection algorithm of pretrained network for fusion to determine the best two models characterized by PTM and NLR. Third, deep CCT fusion by discriminant correlation analysis was proposed to help fuse the two features from the two models. Micro-averaged (MA) F1 score was used as the measuring indicator. The final determined model was named CCSHNet. Results : On the test set, CCSHNet achieved sensitivities of four classes of 95.61%, 96.25%, 98.30%, and 97.86%, respectively. The precision values of four classes were 97.32%, 96.42%, 96.99%, and 97.38%, respectively. The F1 scores of four classes were 96.46%, 96.33%, 97.64%, and 97.62%, respectively. The MA F1 score was 97.04%. In addition, CCSHNet outperformed 12 state-of-the-art COVID-19 detection methods. Conclusions : CCSHNet is effective in detecting COVID-19 and other lung infectious diseases using first-line clinical imaging and can therefore assist radiologists in making accurate diagnoses based on CCTs.

Highlights • We analysed over 320 COVID-19 images and 320 healthy control images. • We proposed an improved CNN to extract individual image-level features. • We proposed to use GCN to extract relation-aware representations. • We proposed a DFF technology to combine features from GCN and CNN. • The proposed FCGNet gives better performance than 15 state-of-the-art approaches., (Aim) COVID-19 is an infectious disease spreading to the world this year. In this study, we plan to develop an artificial intelligence based tool to diagnose on chest CT images. (Method) On one hand, we extract features from a self-created convolutional neural network (CNN) to learn individual image-level representations. The proposed CNN employed several new techniques such as rank-based average pooling and multiple-way data augmentation. On the other hand, relation-aware representations were learnt from graph convolutional network (GCN). Deep feature fusion (DFF) was developed in this work to fuse individual image-level features and relation-aware features from both GCN and CNN, respectively. The best model was named as FGCNet. (Results) The experiment first chose the best model from eight proposed network models, and then compared it with 15 state-of-the-art approaches. (Conclusion) The proposed FGCNet model is effective and gives better performance than all 15 state-of-the-art methods. Thus, our proposed FGCNet model can assist radiologists to rapidly detect COVID-19 from chest CT images.Keywords:Deep feature fusion; Convolutional neural network; Graph convolutional network; Multiple-way data augmentation; Batch normalization; Dropout; Rank-based average pooling

Epilepsy is a prevalent neurological disorder that threatens human health in the world. The most commonly used method to detect epilepsy is using the electroencephalogram (EEG). However, epilepsy detection from the EEG is time-consuming and error-prone work because of the varying levels of experience we find in physicians. To tackle this challenge, in this paper, we propose a multi-scale non-local (MNL) network to achieve automatic EEG signal detection. Our MNL-Network is based on 1D convolution neural network involving two specific layers to improve the classification performance. One layer is named the signal pooling layer which incorporates three different sizes of 1D max-pooling layers to learn the multi-scale features from the EEG signal. The other one is called a multi-scale non-local layer, which calculates the correlation of different multi-scale extracted features and outputs the correlative encoded features to further enhance the classification performance. To evaluate the effectiveness of our model, we conduct experiments on the Bonn dataset. The experimental results demonstrate that our MNL-Network could achieve competitive results in the EEG classification task.

Highlights • We proposed a general ResGNet Framework that is suitable for image classification tasks. • We propose three novel models for COVID-19 detection. • It is the first attempt at applying graph convolutional neural network for COVID-19 detection. • Compared to SOTA, our model achieved the best performance in terms of accuracy., The widely spreading COVID-19 has caused thousands of hundreds of mortalities over the world in the past few months. Early diagnosis of the virus is of great significance for both of infected patients and doctors providing treatments. Chest Computerized tomography (CT) screening is one of the most straightforward techniques to detect pneumonia which was caused by the virus and thus to make the diagnosis. To facilitate the process of diagnosing COVID-19, we therefore developed a graph convolutional neural network ResGNet-C under ResGNet framework to automatically classify lung CT images into normal and confirmed pneumonia caused by COVID-19. In ResGNet-C, two by-products named NNet-C, ResNet101-C that showed high performance on detection of COVID-19 are simultaneously generated as well. Our best model ResGNet-C achieved an averaged accuracy at 0.9662 with an averaged sensitivity at 0.9733 and an averaged specificity at 0.9591 using five cross-validations on the dataset, which is comprised of 296 CT images. To our best knowledge, this is the first attempt at integrating graph knowledge into the COVID-19 classification task. Graphs are constructed according to the Euclidean distance between features extracted by our proposed ResNet101-C and then are encoded with the features to give the prediction results of CT images. Besides the high-performance system, which surpassed all state-of-the-art methods, our proposed graph construction method is simple, transferrable yet quite helpful for improving the performance of classifiers, as can be justified by the experimental results.

Traditional deep learning methods are suboptimal in classifying ambiguity features, which often arise in noisy and hard to predict categories, especially, to distinguish semantic scoring. Semantic scoring, depending on semantic logic to implement evaluation, inevitably contains fuzzy description and misses some concepts, for example, the ambiguous relationship between normal and probably normal always presents unclear boundaries (normal—more likely normal—probably normal). Thus, human error is common when annotating images. Differing from existing methods that focus on modifying kernel structure of neural networks, this article proposes a dominant fuzzy fully connected layer (FFCL) for breast imaging reporting and data system (BI-RADS) scoring and validates the universality of this proposed structure. This proposed model aims to develop complementary properties of scoring for semantic paradigms, while constructing fuzzy rules based on analyzing human thought patterns, and to particularly reduce the influence of semantic conglutination. Specifically, this semantic-sensitive defuzzifier layer projects features occupied by relative categories into semantic space, and a fuzzy decoder modifies probabilities of the last output layer referring to the global trend. Moreover, the ambiguous semantic space between two relative categories shrinks during the learning phases, as the positive and negative growth trends of one category appearing among its relatives were considered. We first used the Euclidean distance to zoom in the distance between the real scores and the predicted scores, and then employed two sample t test method to evidence the advantage of the FFCL architecture. Extensive experimental results performed on the curated breast imaging subset of digital database of screening mammography dataset show that our FFCL structure can achieve superior performances for both triple and multiclass classification in BI-RADS scoring, outperforming the state-of-the-art methods.

Ductal carcinoma in situ (DCIS) is a pre-cancerous lesion in the ducts of the breast, and early diagnosis is crucial for optimal therapeutic intervention. Thermography imaging is a non-invasive imaging tool that can be utilized for detection of DCIS and although it has high accuracy (~88%), it is sensitivity can still be improved. Hence, we aimed to develop an automated artificial intelligence-based system for improved detection of DCIS in thermographs. This study proposed a novel artificial intelligence based system based on convolutional neural network (CNN) termed CNN-BDER on a multisource dataset containing 240 DCIS images and 240 healthy breast images. Based on CNN, batch normalization, dropout, exponential linear unit and rank-based weighted pooling were integrated, along with L-way data augmentation. Ten runs of tenfold cross validation were chosen to report the unbiased performances. Our proposed method achieved a sensitivity of 94.08±1.22%, a specificity of 93.58±1.49 and an accuracy of 93.83±0.96. The proposed method gives superior performance than eight state-of-theart approaches and manual diagnosis. The trained model could serve as a visual question answering system and improve diagnostic accuracy., British Heart Foundation Accelerator Award, UK, Royal Society International Exchanges Cost Share Award, UK RP202G0230, Hope Foundation for Cancer Research, UK RM60G0680, Medical Research Council Confidence in Concept Award, UK MC_PC_17171, MINECO/FEDER, Spain/Europe RTI2018-098913-B100 A-TIC-080-UGR18

The outbreak of COVID-19 has been striking the world for months and caused hundreds of thousands of mortality. Early and accurate detection turns out to be one of the most effective ways to slow the spreading of the virus. To help radiologists interpret images, we developed an automatic CT image-based detection system, which achieved high accuracy on the detection of COVID-19. The proposed model in the detection system is codenamed GoogLeNet-COD, which utilizes one of the state-of-the-art deep convolutional neural networks GooLeNet as the backbone. As GoogLeNet was initially trained on ImageNet, we first replaced the last top two layers with four new layers, which include the dropout layer, two fully-connected layers and the output layer. The dropout technique is utilized to prevent overfitting in the system by inserting a dropout layer in the network. The newly added fully-connected layer serves as a transitional layer that prevents significant information loss while the last fully-connected layer is used to generate possibilities for the final output layer. The hold-out validation method is used to evaluate the performance of the proposed system. The experiment on a private COVID-19 dataset showed a high accuracy of our system.

Alcoholism changes the structure of brain. Several somatic marker hypothesis network-related regions are known to be damaged in chronic alcoholism. Neuroimaging approach can help us better understanding the impairment discovered in alcohol-dependent subjects. In this research, we recruited subjects from participating hospitals. In total, 188 abstinent long-term chronic alcoholic participants (95 men and 93 women) and 191 non-alcoholic control participants (95 men and 96 women) were enrolled in our experiment via computerized diagnostic interview schedule version IV and medical history interview employed to determine whether the applicants can be enrolled or excluded. The Siemens Verio Tim 3.0 T MR scanner (Siemens Medical Solutions, Erlangen, Germany) was employed to scan the subjects. Then, we proposed a 10-layer convolutional neural network for the diagnosis based on imaging, including three advanced techniques: parametric rectified linear unit (PReLU); batch normalization; and dropout. The structure of network is fine-tuned. The results show that our method secured a sensitivity of 97.73 ± 1.04%, a specificity of 97.69 ± 0.87%, and an accuracy of 97.71 ± 0.68%. We observed the PReLU gives better performance than ordinary ReLU, clipped ReLU, and leaky ReLU. The batch normalization and dropout gained enhanced performance as batch normalization overcame the internal covariate shift and dropout got over the overfitting. The results of our proposed 10-layer CNN model show its performance better than seven state-of-the-art approaches.

Background The number of patients with Alzheimer's disease is increasing rapidly every year. Scholars often use computer vision and machine learning methods to develop an automatic diagnosis system. Objective In this study, we developed a novel machine learning system that can make diagnoses automatically from brain magnetic resonance images. Methods First, the brain imaging was processed, including skull stripping and spatial normalization. Second, one axial slice was selected from the volumetric image, and stationary wavelet entropy (SWE) was done to extract the texture features. Third, a single-hidden-layer neural network was used as the classifier. Finally, a predator-prey particle swarm optimization was proposed to train the weights and biases of the classifier. Results Our method used 4-level decomposition and yielded 13 SWE features. The classification yielded an overall accuracy of 92.73±1.03%, a sensitivity of 92.69±1.29%, and a specificity of 92.78±1.51%. The area under the curve is 0.95±0.02. Additionally, this method only cost 0.88 s to identify a subject in online stage, after its volumetric image is preprocessed. Conclusion In terms of classification performance, our method performs better than 10 state-of-the-art approaches and the performance of human observers. Therefore, this proposed method is effective in the detection of Alzheimer's disease.

Hubs in complex networks can greatly influence the integration of network functions, and recognition of hubs helps to better understand the interaction between pairs of network nodes. This paper proposes a new hub recognition method with multiple-feature combination for the brain functional networks constructed by resting-state functional Magnetic Resonance Imaging (fMRI). Three single-feature methods, including degree centrality, betweenness centrality and closeness centrality, are used to calculate hubs of the brain functional network separately. For reordering the nodes, a composite equation is constructed based on the three recognition parameters. Network vulnerability and average shortest path length are used to evaluate the importance of the hubs recognized by above four methods. Experimental result demonstrates that, the hubs recognized by multiple-feature combination have more significant differences from ordinary nodes than those by single-feature methods, and they have an important impact on the global efficiency of brain functional networks.

Cerebral Microbleeds (CMBs) are considered as an essential indicator in the diagnosis of critical cerebrovascular diseases such as ischemic stroke and dementia. Manual detection of CMBs is prone to errors due to complex morphological nature of CMBs. In this paper, an efficient method is presented for CMB detection in Susceptibility-Weighted Imaging (SWI) scans. The proposed framework consists of three phases: i) brain extraction, ii) extraction of initial candidates based on threshold and size based filtering, and iii) feature extraction and classification of CMBs from other healthy tissues in order to remove false positives using Support Vector Machine, Quadratic Discriminant Analysis (QDA) and ensemble classifiers. The proposed technique is validated on a dataset of 20 subjects with CMBs that consists of 14 subjects for training and 6 subjects for testing. QDA classifier achieved the best sensitivity of 93.7% with 56 false positives per patient and 5.3 false positives per CMB.

Development of automated diagnosis systems has taken a major place in current research practice to assist medical experts in decision-making. This paper presents a new automatic system for detection of pathological brain through magnetic resonance imaging (MRI). The proposed system involves contrast enhancement of input MR images using contrast limited adaptive histogram equalization (CLAHE). Then, the curve like features are computed from the preprocessed MR brain images using fast discrete curvelet transform via unequally-spaced FFT (FDCT-USFFT). Subsequently, a combined technique known as PCA+LDA is employed to derive more discriminative and reduced feature sets. Finally, a novel learning approach dubbed as extreme learning machine with modified sine cosine algorithm (MSCA-ELM) is proposed by combining ELM and MSCA for classification of MR images into two categories: pathological and healthy. A mutation operator is introduced to basic SCA (MSCA). In MSCA-ELM, MSCA is used to optimize the input weights and hidden biases of single-hidden layer feed-forward neural network (SLFN) and an analytical procedure is used to compute the output weights. The proposed scheme is rigorously evaluated on three standard datasets and the results are compared against other competent schemes. The experimental results demonstrate that the proposed scheme outperforms its counterparts in terms of classification accuracy and number of features required. It has also been noticed that MSCA-ELM yields superior performance than conventional learning methods. Hence, the proposed system can effectively recognize pathological brain in real-time and can possibly be installed on medical robots.

In many cases, users of smart wheelchairs have difficulties with daily maneuvering tasks and would benefit from an automated navigation system. With multi-colony division and cooperation mechanism, the polymorphic ant colony algorithm is helpful to solve optimal path planning problems by greatly improving search and convergence speed. In this paper, a path planning method for smart wheelchairs is proposed based on the adaptive polymorphic ant colony algorithm. To avoid ant colony from getting into local optimum in the process of reaching a solution, the adaptive state transition strategy and the adaptive information updating strategy were employed in the polymorphic ant colony algorithm to guarantee the relative importance of pheromone intensity and desirability. Subsequently, the search ant maintains the randomness for the search of the global optimal solution, and then the deadlock problem is solved by means of the direction determination method that improves the global search ability of the algorithm. The target path planning and obstacle path planning are respectively carried out by using the adaptive polymorphic ant colony algorithm. Experimental results indicate that the proposed method provides better performance than the improved ant colony algorithm and the polymorphic ant colony algorithm. Furthermore, the efficiency of finding an optimum solution is higher than the average polymorphic ant colony algorithm. The proposed method, which achieves superior performance in path planning for smart wheelchairs, is even racing ahead of other state-of-the-art solutions. In addition, this study reveals the feasibility of using it as an effective and feasible planning path tool for future healthcare systems.

AimEmotion recognition based on facial expression is an important field in affective computing. Current emotion recognition systems may suffer from two shortcomings: translation in facial image may deteriorate the recognition performance, and the classifier is not robust. MethodTo solve above two problems, our team proposed a novel intelligent emotion recognition system. Our method used stationary wavelet entropy to extract features, and employed a single hidden layer feedforward neural network as the classifier. To prevent the training of the classifier fall into local optimum points, we introduced the Jaya algorithm. ResultsThe simulation results over a 20-subject 700-image dataset showed our algorithm reached an overall accuracy of 96.800.14%. ConclusionThis proposed approach performs better than five state-of-the-art approaches in terms of overall accuracy. Besides, the db4 wavelet performs the best among other whole db wavelet family. The 4-level wavelet decomposition is superior to other levels. In the future, we shall test other advanced features and training algorithms.

The alcohol use disorder (AUD) is an important brain disease, which could cause the damage and alteration of brain structure. The current diagnosis of AUD is mainly done manually by radiologists. This study proposes a novel computer-vision-based method for automatic detection of AUD based on wavelet Renyi entropy and three-segment encoded Jaya algorithm from MRI scans. The wavelet Renyi entropy is proposed to provide multiresolution and multiscale analysis of features, describe the complexity of the brain structure, and extract the distinctive features. Grid search method was used to select the optimal wavelet decomposition level and Renyi order. The classifier was constructed based on feedforward neural network and a three-segment encoded (TSE) Jaya algorithm providing parameter-free training of the weights, biases, and number of hidden neurons. We have conducted the experimental evaluation on 235 subjects (114 are AUDs and 121 healthy). k-fold cross validation has been used to avoid overfitting and report out-of-sample errors. The results showed that the proposed method outperforms four state-of-the-art approaches in terms of accuracy. The proposed TSE-Jaya provides a better performance, compared to the conventional approaches including plain Jaya, multiobjective genetic algorithm, particle swarm optimization, bee colony optimization, modified ant colony system, and real-coded biogeography-based optimization.

Cerebral microbleed (CMB) is a type of biomarker, which is related to cerebrovascular diseases. In this paper, a novel computer aided diagnosis method for CMB detection was presented. Firstly, sliding neighborhood algorithm was used to generate CMB and non-CMB samples from brain susceptibility weighted images. Then, a 15-layer proposed FeatureNet was trained for extracting features from the input samples. Afterwards, structure after the first fully connected layer in FeatureNet was replaced by three randomized neural networks for classification: Schmidt neural network, random vector functional-link net, and extreme learning machine, and the weights and biases in early layers of FeatureNet were frozen during the training of those three classifiers. Finally, the output of the three classifiers was ensemble by majority voting mechanism to get better classification performance. In our experiment, five-fold cross validation was employed for evaluation. Results revealed that our FeatureNet-SNN, FeatureNet-RVFL and FeatureNet-ELM yielded accuracy of 98.22%, 98.23%, and 97.54%, respectively, and the ensembled FeatureNet-EN improved the accuracy to 98.60%, which outperformed several existing state-of-the-art approaches. The proposed FeatureNet-EN model could provide accurate CMB detection, and thus reduce death tolls. Impact Statement — We propose a 15-layer FeatureNet to detect cerebral microbleed (CMB). We propose three FeatureNet variants: FeatureNet-SNN, FeatureNet-RVFL and FeatureNet-ELM. We use ensemble learning to combine three FeatureNet variants, and generate a FeatureNet-EN. The proposed FeatureNet-SNN, FeatureNet-RVFL and FeatureNet-ELM yielded accuracy of 98.22%, 98.23%, and 97.54%, respectively, and the ensembled FeatureNet-EN improved the accuracy to 98.60%, better than state-of-the-art methods. This method could provide accurate CMB detection, and thus reduce death tolls.

This work is aimed to develop the key algorithms used in medical robots, which can detect alcohol use disorder from structural magnetic resonance imaging of brains. We enrolled 30 alcoholic participants and 30 nonalcoholic participants. In the algorithm stage, we suggested to use Hu moment invariant to extract global features, and use single-hidden layer neural-network as the classifier. Afterwards, we proposed a novel predator–prey adaptive-inertia chaotic particle swarm optimization algorithm to train the classifier. The ten-fold stratified cross validation showed that our method achieves a sensitivity of 90.67 ± 3.16%, a specificity of 91.33 ± 3.06%, and an accuracy of 91.00 ± 1.41%. Our results are better than genetic algorithm, firefly algorithm, and particle swarm optimization. This proposes algorithm is effective in alcoholism detection. It can be installed on medical robots.

Image segmentation is an important application of polarimetric synthetic aperture radar. This study aimed to create an 11-layer deep convolutional neural network for this task. The Pauli decomposition formed the RGB image and was used as the input. We created an 11-layer convolutional neural network (CNN). L-band data over the San Francisco bay area and C-band data over Flevoland area were employed as the dataset. For the San Francisco bay PSAR image, our method achieved an overall accuracy of 97.32%, which was at least 2% superior to four state-of-the-art approaches. We provided the confusion matrix over test area, and the kernel visualization. We compared the max pooling and average pooling. We validated by experiment that four convolution layers perform the best. Besides, our method gave better results than AlexNet. The GPU yields a 173× acceleration on the training samples, and a 181× acceleration on the test samples, compared to standard CPU. For the Flevoland PSAR image, our 11-layer CNN also gives better overall accuracy than five state-of-the-art approaches. The convolutional neural network is better than traditional classifiers and is effective in remote sensing image segmentation.

In order to detect the cerebral microbleed (CMB) voxels within brain, we used susceptibility weighted imaging to scan the subjects. Then, we used undersampling to solve the accuracy paradox caused from the imbalanced data between CMB voxels and non-CMB voxels. we developed a seven-layer deep neural network (DNN), which includes one input layer, four sparse autoencoder layers, one softmax layer, and one output layer. Our simulation showed this method achieved a sensitivity of 95.13%, a specificity of 93.33%, and an accuracy of 94.23%. The result is better than three state-of-the-art approaches.