SecureNet: Proactive intellectual property protection and model security defense for DNNs based on backdoor learning

With the widespread application of deep neural networks (DNNs), the risk of privacy breaches against DNN models is constantly on the rise, resulting in an increasing need for intellectual property (IP) protection for such models. Although neural network watermarking techniques are widely used to safeguard the IP of DNNs, they can only achieve passive protection and cannot actively prevent unauthorized users from illicit use or embezzlement of the trained DNN models. Therefore, the development of proactive protection techniques to prevent IP infringement is imperative. To this end, we propose SecureNet, a key-based access license framework for DNN models. The proposed approach involves injecting license keys into the model through backdoor learning, enabling correct model functionality only when the appropriate license key is included in the input. To ensure the reusability of DNN models, we also propose a license key replacement algorithm. In addition, based on SecureNet, we designed defense mechanisms against adversarial attacks and backdoor attacks, respectively. Furthermore, we introduce a fine-grained authorization method that enables flexible granting of model permissions to different users. We have designed four license-key schemes with different privileges, tailored to various scenarios. We evaluated SecureNet on five benchmark datasets including MNIST, Cifar10, Cifar100, FaceScrub, and CelebA, and assessed its performance on six classic DNN models: LeNet-5, VGG16, ResNet18, ResNet101, NFNet-F5, and MobileNetV3. The results demonstrate that our approach outperforms the state-of-the-art model parameter encryption methods by at least 95% in terms of computational efficiency. Additionally, it provides effective defense against adversarial attacks and backdoor attacks without compromising the model's overall performance.

Attention-based investigation and solution to the trade-off issue of adversarial training

Adversarial training has become the mainstream method to boost adversarial robustness of deep models. However, it often suffers from the trade-off dilemma, where the use of adversarial examples hurts the standard generalization of models on natural data. To study this phenomenon, we investigate it from the perspective of spatial attention. In brief, standard training typically encourages a model to conduct a comprehensive check to input space. But adversarial training often causes a model to overly concentrate on sparse spatial regions. This reduced tendency is beneficial to avoid adversarial accumulation but easily makes the model ignore abundant discriminative information, thereby resulting in weak generalization. To address this issue, this paper introduces an Attention-Enhanced Learning Framework (AELF) for robustness training. The main idea is to enable the model to inherit the attention pattern of standard pre-trained model through an embedding-level regularization. To be specific, given a teacher model built on natural examples, the embedding distribution of teacher model is used as a static constraint to regulate the embedding outputs of the objective model. This design is mainly supported with that the embedding feature of standard model is usually recognized as a rich semantic integration of input. For implementation, we present a simplified AELFs that can achieve the regularization with single cross entropy loss via the parameter initialization and parameter update strategy. This avoids the extra consistency comparison operation between embedding vectors. Experimental observations verify the rationality of our argument, and experimental results demonstrate that it can achieve remarkable improvements in generalization under the high-level robustness.

Enhancing adversarial attacks with resize-invariant and logical ensemble

In black-box scenarios, most transfer-based attacks usually improve the transferability of adversarial examples by optimizing the gradient calculation of the input image. Unfortunately, since the gradient information is only calculated and optimized for each pixel point in the image individually, the generated adversarial examples tend to overfit the local model and have poor transferability to the target model. To tackle the issue, we propose a resize-invariant method (RIM) and a logical ensemble transformation method (LETM) to enhance the transferability of adversarial examples. Specifically, RIM is inspired by the resize-invariant property of Deep Neural Networks (DNNs). The range of resizable pixel is first divided into multiple intervals, and then the input image is randomly resized and padded within each interval. Finally, LETM performs logical ensemble of multiple images after RIM transformation to calculate the final gradient update direction. The proposed method adequately considers the information of each pixel in the image and the surrounding pixels. The probability of duplication of image transformations is minimized and the overfitting effect of adversarial examples is effectively mitigated. Numerous experiments on the ImageNet dataset show that our approach outperforms other advanced methods and is capable of generating more transferable adversarial examples.

Multi-scale feature fusion for prediction of IDH1 mutations in glioma histopathological images

BACKGROUND AND OBJECTIVE

Mutations in isocitrate dehydrogenase 1 (IDH1) play a crucial role in the prognosis, diagnosis, and treatment of gliomas. However, current methods for determining its mutation status, such as immunohistochemistry and gene sequencing, are difficult to implement widely in routine clinical diagnosis. Recent studies have shown that using deep learning methods based on pathological images of glioma can predict the mutation status of the IDH1 gene. However, our research focuses on utilizing multi-scale information in pathological images to improve the accuracy of predicting IDH1 gene mutations, thereby providing an accurate and cost-effective prediction method for routine clinical diagnosis.

METHODS

In this paper, we propose a multi-scale fusion gene identification network (MultiGeneNet). The network first uses two feature extractors to obtain feature maps at different scale images, and then by employing a bilinear pooling layer based on Hadamard product to realize the fusion of multi-scale features. Through fully exploiting the complementarity among features at different scales, we are able to obtain a more comprehensive and rich representation of multi-scale features.

RESULTS

Based on the Hematoxylin and Eosin stained pathological section dataset of 296 patients, our method achieved an accuracy of 83.575 % and an AUC of 0.886, thus significantly outperforming other single-scale methods.

CONCLUSIONS

Our method can be deployed in medical aid systems at very low cost, serving as a diagnostic or prognostic tool for glioma patients in medically underserved areas.

Semi-supervised medical image classification via distance correlation minimization and graph attention regularization

We propose a novel semi-supervised learning method to leverage unlabeled data alongside minimal annotated data and improve medical imaging classification performance in realistic scenarios with limited labeling budgets to afford data annotations. Our method introduces distance correlation to minimize correlations between feature representations from different views of the same image encoded with non-coupled deep neural networks architectures. In addition, it incorporates a data-driven graph-attention based regularization strategy to model affinities among images within the unlabeled data by exploiting their inherent relational information in the feature space. We conduct extensive experiments on four medical imaging benchmark data sets involving X-ray, dermoscopic, magnetic resonance, and computer tomography imaging on single and multi-label medical imaging classification scenarios. Our experiments demonstrate the effectiveness of our method in achieving very competitive performance and outperforming several state-of-the-art semi-supervised learning methods. Furthermore, they confirm the suitability of distance correlation as a versatile dependence measure and the benefits of the proposed graph-attention based regularization for semi-supervised learning in medical imaging analysis.

Anatomically aware dual-hop learning for pulmonary embolism detection in CT pulmonary angiograms

Pulmonary Embolisms (PE) represent a leading cause of cardiovascular death. While medical imaging, through computed tomographic pulmonary angiography (CTPA), represents the gold standard for PE diagnosis, it is still susceptible to misdiagnosis or significant diagnosis delays, which may be fatal for critical cases. Despite the recently demonstrated power of deep learning to bring a significant boost in performance in a wide range of medical imaging tasks, there are still very few published researches on automatic pulmonary embolism detection. Herein we introduce a deep learning based approach, which efficiently combines computer vision and deep neural networks for pulmonary embolism detection in CTPA. Our method brings novel contributions along three orthogonal axes: (1) automatic detection of anatomical structures; (2) anatomical aware pretraining, and (3) a dual-hop deep neural net for PE detection. We obtain state-of-the-art results on the publicly available multicenter large-scale RSNA dataset.

Hybrid disease prediction approach leveraging digital twin and metaverse technologies for health consumer

Emerging from the convergence of digital twin technology and the metaverse, consumer health (MCH) is witnessing a transformative shift. The amalgamation of bioinformatics with healthcare Big Data has ushered in a new era of disease prediction models that harness comprehensive medical data, enabling the anticipation of illnesses even before the onset of symptoms. In this model, deep neural networks stand out because they improve accuracy remarkably by increasing network depth and making weight changes using gradient descent. Nonetheless, traditional methods face their own set of challenges, including the issues of gradient instability and slow training. In this case, the Broad Learning System (BLS) stands out as a good alternative. It gets around the problems with gradient descent and lets you quickly rebuild a model through incremental learning. One problem with BLS is that it has trouble extracting complex features from complex medical data. This makes it less useful in a wide range of healthcare situations. In response to these challenges, we introduce DAE-BLS, a novel hybrid model that marries Denoising AutoEncoder (DAE) noise reduction with the efficiency of BLS. This hybrid approach excels in robust feature extraction, particularly within the intricate and multifaceted world of medical data. Validation using diverse datasets yields impressive results, with accuracies reaching as high as 98.50%. DAE-BLS's ability to rapidly adapt through incremental learning holds great promise for accurate and agile disease prediction, especially within the complex and dynamic healthcare scenarios of today.

Optimization of medium components for protein production by Escherichia coli with a high-throughput pipeline that uses a deep neural network

To optimize rapidly the medium for green fluorescent protein expression by Escherichia coli with an introduced plasmid, pRSET/emGFP, a single-cycle optimization pipeline was applied. The pipeline included a deep neural network (DNN) and mathematical optimization algorithms with simultaneous optimization of 18 medium components. To evaluate the DNN data sampling method, two methods, orthogonal array (OA) and Latin hypercube sampling (LHS), were used to design 64 initial media for each sampling method. The OA- and LHS-based data sampling resulted in green fluorescent protein fluorescence intensities of 0.088 × 103-1.85 × 104 and 3.30 × 103-1.50 × 104, respectively. Fifty DNN models were built using the OA and LHS datasets. Hold-out validation was performed using 15 % test of OA and LHS data. Mean square errors of the DNN models were 0.015-0.64, indicating the estimation accuracies were sufficient. However, the sensitivities of components in the DNN models varied and were grouped into six major classes by the index of k-means clustering. A representative model was selected for each class. Mathematical optimization algorithms using Bayesian optimization and genetic algorithm were applied to the representative models, and representative optimized medium (OM) compositions were selected by k-means clustering from the proposed OMs. A total of 54 OMs were obtained from the OA and LHS datasets. In the validating cultivation, the best OMs of OA and LHS were 2.12-fold and 2.13-fold higher, respectively, than those of the learning data.

Adversarially robust neural networks with feature uncertainty learning and label embedding

Deep neural networks (DNNs) are vulnerable to the attacks of adversarial examples, which bring serious security risks to the learning systems. In this paper, we propose a new defense method to improve the adversarial robustness of DNNs based on stochastic neural networks (SNNs), termed as Margin-SNN. The proposed Margin-SNN mainly includes two modules, i.e., feature uncertainty learning module and label embedding module. The first module introduces uncertainty to the latent feature space by giving each sample a distributional representation rather than a fixed point representation, and leverages the advantages of variational information bottleneck method in achieving good intra-class compactness in latent space. The second module develops a label embedding mechanism to take advantage of the semantic information underlying the labels, which maps the labels into the same latent space with the features, in order to capture the similarity between sample and its class centroid, where a penalty term is equipped to elegantly enlarge the margin between different classes for better inter-class separability. Since no adversarial information is introduced, the proposed model can be learned in standard training to improve adversarial robustness, which is much more efficient than adversarial training. Extensive experiments on data sets MNIST, FASHION MNIST, CIFAR10, CIFAR100 and SVHN demonstrate superior defensive ability of the proposed method. Our code is available at https://github.com/humeng24/Margin-SNN.

Representational formats of human memory traces

Neural representations are internal brain states that constitute the brain's model of the external world or some of its features. In the presence of sensory input, a representation may reflect various properties of this input. When perceptual information is no longer available, the brain can still activate representations of previously experienced episodes due to the formation of memory traces. In this review, we aim at characterizing the nature of neural memory representations and how they can be assessed with cognitive neuroscience methods, mainly focusing on neuroimaging. We discuss how multivariate analysis techniques such as representational similarity analysis (RSA) and deep neural networks (DNNs) can be leveraged to gain insights into the structure of neural representations and their different representational formats. We provide several examples of recent studies which demonstrate that we are able to not only measure memory representations using RSA but are also able to investigate their multiple formats using DNNs. We demonstrate that in addition to slow generalization during consolidation, memory representations are subject to semantization already during short-term memory, by revealing a shift from visual to semantic format. In addition to perceptual and conceptual formats, we describe the impact of affective evaluations as an additional dimension of episodic memories. Overall, these studies illustrate how the analysis of neural representations may help us gain a deeper understanding of the nature of human memory.

Neural networks with ReLU powers need less depth

Despite the widespread success of deep learning in various applications, neural network theory has been lagging behind. The choice of the activation function plays a critical role in the expressivity of a neural network but for reasons that are not yet fully understood. While the rectified linear unit (ReLU) is currently one of the most popular activation functions, ReLU squared has only recently been empirically shown to be pivotal in producing consistently superior results for state-of-the-art deep learning tasks (So et al., 2021). To analyze the expressivity of neural networks with ReLU powers, we employ the novel framework of Gribonval et al. (2022) based on the classical concept of approximation spaces. We consider the class of functions for which the approximation error decays at a sufficiently fast rate as network complexity, measured by the number of weights, increases. We show that when approximating sufficiently smooth functions that cannot be represented by sufficiently low-degree polynomials, networks with ReLU powers need less depth than those with ReLU. Moreover, if they have the same depth, networks with ReLU powers can have potentially faster approximation rates. Lastly, our computational experiments on approximating the Rastrigin and Ackley functions with deep neural networks showed that ReLU squared and ReLU cubed networks consistently outperform ReLU networks.

SME-MFP: A novel spatiotemporal neural network with multiangle initialization embedding toward multifunctional peptides prediction

As a promising alternative to conventional antibiotic drugs in the biomedical field, functional peptide has been widely used in disease treatment owing to its low toxicity, high absorption rate, and biological activity. Recently, several machine learning methods have been developed for functional peptide prediction. However, the main research heavily relies on statistical features and few consider multifunctional peptide identification. So, we propose SME-MFP, a novel predictor in the imbalanced multi-label functional peptide datasets. First, we employ physicochemical and evolutionary information to represent the peptide sequence's initialization features from multiple perspectives. Second, the features are fused and then put into spatial feature extractors, where the residual connection and multiscale convolutional neural network extract more discriminative features of different lengths' peptide sequences. Besides, we also design AFT-based temporal feature extractors to fully capture the global interactions of the sequences. Finally, devising a new loss to replace the traditional cross entropy loss to settle the class imbalance problems. The results show that our framework not only enhances the model's ability to capture sequence features effectively, but also accuracy improves by 3.89% over existing methods on public peptide datasets.

Understanding the role of pathways in a deep neural network

Deep neural networks have demonstrated superior performance in artificial intelligence applications, but the opaqueness of their inner working mechanism is one major drawback in their application. The prevailing unit-based interpretation is a statistical observation of stimulus-response data, which fails to show a detailed internal process of inherent mechanisms of neural networks. In this work, we analyze a convolutional neural network (CNN) trained in the classification task and present an algorithm to extract the diffusion pathways of individual pixels to identify the locations of pixels in an input image associated with object classes. The pathways allow us to test the causal components which are important for classification and the pathway-based representations are clearly distinguishable between categories. We find that the few largest pathways of an individual pixel from an image tend to cross the feature maps in each layer that is important for classification. And the large pathways of images of the same category are more consistent in their trends than those of different categories. We also apply the pathways to understanding adversarial attacks, object completion, and movement perception. Further, the total number of pathways on feature maps in all layers can clearly discriminate the original, deformed, and target samples.

Super-resolution deep neural network (SRDNN) based multi-image steganography for highly secured lossless image transmission

Information exchange and communication through the Internet are one of the most crucial aspects of today's information technology world. The security of information transmitted online has grown to be a critical concern, particularly in the transfer of medical data. To overcome this, the data must be delivered securely without being altered or lost. This can be possibly done by combining the principles of cryptography and steganography. In the recent past, steganography is used with simpler methods like the least significant bit manipulation technique, in order to encode a lower-resolution image into a higher-resolution image. Here, we attempt to use deep neural networks to combine many two-dimensional colour images of the same resolution into a single cover image with the same resolution. In this technique, many secret images are concealed inside a single cover image using deep neural networks. The embedded cover image is then encrypted using a 3D chaotic map for diffusion and elliptic curve cryptography (ECC) for confusion to increase security.Supporting the fact that neural networks experience losses, the proposed system recovers up to 93% of the hidden image concealed in the original image. As the secret image features are identified and combined along with the cover image, the time complexity involved in the security process is minimized by 78% compared to securing the original data.

Understanding skin color bias in deep learning-based skin lesion segmentation

BACKGROUND

The field of dermatological image analysis using deep neural networks includes the semantic segmentation of skin lesions, pivotal for lesion analysis, pathology inference, and diagnoses. While biases in neural network-based dermatoscopic image classification against darker skin tones due to dataset imbalance and contrast disparities are acknowledged, a comprehensive exploration of skin color bias in lesion segmentation models is lacking. It is imperative to address and understand the biases in these models.

METHODS

Our study comprehensively evaluates skin tone bias within prevalent neural networks for skin lesion segmentation. Since no information about skin color exists in widely used datasets, to quantify the bias we use three distinct skin color estimation methods: Fitzpatrick skin type estimation, Individual Typology Angle estimation as well as manual grouping of images by skin color. We assess bias across common models by training a variety of U-Net-based models on three widely-used datasets with 1758 different dermoscopic and clinical images. We also evaluate commonly suggested methods to mitigate bias.

RESULTS

Our findings expose a significant and large correlation between segmentation performance and skin color, revealing consistent challenges in segmenting lesions for darker skin tones across diverse datasets. Using various methods of skin color quantification, we have found significant bias in skin lesion segmentation against darker-skinned individuals when evaluated both in and out-of-sample. We also find that commonly used methods for bias mitigation do not result in any significant reduction in bias.

CONCLUSIONS

Our findings suggest a pervasive bias in most published lesion segmentation methods, given our use of commonly employed neural network architectures and publicly available datasets. In light of our findings, we propose recommendations for unbiased dataset collection, labeling, and model development. This presents the first comprehensive evaluation of fairness in skin lesion segmentation.

Markerless gait analysis through a single camera and computer vision

The assessment of gait performance using quantitative measures can yield crucial insights into an individual's health status. Recently, computer vision-based human pose estimation has emerged as a promising solution for markerless gait analysis, as it allows for the direct extraction of gait parameters from videos. This study aimed to compare the lower extremity kinematics and spatiotemporal gait parameters obtained from a single-camera-based markerless method with those acquired from a marker-based motion tracking system across a healthy population. Additionally, we investigated the impact of camera viewing angles and distances on the accuracy of the markerless method. Our findings demonstrated a robust correlation and agreement (Rxy > 0.75, Rc > 0.7) between the markerless and marker-based methods for most spatiotemporal gait parameters. We also observed strong correlations (Rxy > 0.8) between the two methods for hip flexion/extension, knee flexion/extension, hip abduction/adduction, and hip internal/external rotation. Statistical tests revealed significant effects of viewing angles and distances on the accuracy of the identified gait parameters. While the markerless method offers an alternative for general gait analysis, particularly when marker use is impractical, its accuracy for clinical applications remains insufficient and requires substantial improvement. Future investigations should explore the potential of the markerless system to measure gait parameters in pathological gaits.

Automatic Detection of 30 Fundus Diseases Using Ultra-Widefield Fluorescein Angiography with Deep Experts Aggregation

INTRODUCTION

Inaccurate, untimely diagnoses of fundus diseases leads to vision-threatening complications and even blindness. We built a deep learning platform (DLP) for automatic detection of 30 fundus diseases using ultra-widefield fluorescein angiography (UWFFA) with deep experts aggregation.

METHODS

This retrospective and cross-sectional database study included a total of 61,609 UWFFA images dating from 2016 to 2021, involving more than 3364 subjects in multiple centers across China. All subjects were divided into 30 different groups. The state-of-the-art convolutional neural network architecture, ConvNeXt, was chosen as the backbone to train and test the receiver operating characteristic curve (ROC) of the proposed system on test data and external test date. We compared the classification performance of the proposed system with that of ophthalmologists, including two retinal specialists.

RESULTS

We built a DLP to analyze UWFFA, which can detect up to 30 fundus diseases, with a frequency-weighted average area under the receiver operating characteristic curve (AUC) of 0.940 in the primary test dataset and 0.954 in the external multi-hospital test dataset. The tool shows comparable accuracy with retina specialists in diagnosis and evaluation.

CONCLUSIONS

This is the first study on a large-scale UWFFA dataset for multi-retina disease classification. We believe that our UWFFA DLP advances the diagnosis by artificial intelligence (AI) in various retinal diseases and would contribute to labor-saving and precision medicine especially in remote areas.

Co-registered optical coherence tomography and X-ray angiography for the prediction of fractional flow reserve

Cardiovascular disease (CVD) stands as the leading global cause of mortality, and coronary artery disease (CAD) has the highest prevalence, contributing to 42% of these fatalities. Recognizing the constraints inherent in the anatomical assessment of CAD, Fractional Flow Reserve (FFR) has emerged as a pivotal functional diagnostic metric. Herein, we assess the potential of employing an ensemble approach with deep neural networks (DNN) to predict invasively measured Fractional Flow Reserve (FFR) using raw anatomical data extracted from both optical coherence tomography (OCT) and X-ray coronary angiography (XA). In this study, we used a challenging dataset, with 46% of the lesions falling within the FFR range of 0.75 to 0.85. Despite this complexity, our model achieved an accuracy of 84.3%, demonstrating a sensitivity of 87.5% and a specificity of 81.4%. Our results demonstrate that incorporating both OCT and XA signals, co-registered, as inputs for the DNN model leads to an important increase in overall accuracy.

Structural-topic aware deep neural networks for information cascade prediction

It is critical to accurately predict the future popularity of information cascades for many related applications, such as online opinion warning or academic influence evaluation. Despite many efforts devoted to developing effective prediction approaches, especially the recent presence of deep learning-based model, the structural information of the cascade network is ignored. Thus, to make use of the structural information in cascade prediction task, we propose a structural-topic aware deep neural networks (STDNN), which firstly learns the structure topic distribution of each node in the cascade, feeds it to a sequential neural network, and finally predicts the future popularity of the cascades. It can inherit the high interpretability of Hawkes process and possesses the high predictive power of deep learning methods, bridging the gap between prediction and understanding of information cascades by capturing indicative graph structures. We evaluate our model through quantitative experiments, where our model exhibits promising performance, efficiency higher than the baselines.

Deep neural networks for wearable sensor-based activity recognition in Parkinson's disease: investigating generalizability and model complexity

BACKGROUND

The research gap addressed in this study is the applicability of deep neural network (NN) models on wearable sensor data to recognize different activities performed by patients with Parkinson's Disease (PwPD) and the generalizability of these models to PwPD using labeled healthy data.

METHODS

The experiments were carried out utilizing three datasets containing wearable motion sensor readings on common activities of daily living. The collected readings were from two accelerometer sensors. PAMAP2 and MHEALTH are publicly available datasets collected from 10 and 9 healthy, young subjects, respectively. A private dataset of a similar nature collected from 14 PwPD patients was utilized as well. Deep NN models were implemented with varying levels of complexity to investigate the impact of data augmentation, manual axis reorientation, model complexity, and domain adaptation on activity recognition performance.

RESULTS

A moderately complex model trained on the augmented PAMAP2 dataset and adapted to the Parkinson domain using domain adaptation achieved the best activity recognition performance with an accuracy of 73.02%, which was significantly higher than the accuracy of 63% reported in previous studies. The model's F1 score of 49.79% significantly improved compared to the best cross-testing of 33.66% F1 score with only data augmentation and 2.88% F1 score without data augmentation or domain adaptation.

CONCLUSION

These findings suggest that deep NN models originating on healthy data have the potential to recognize activities performed by PwPD accurately and that data augmentation and domain adaptation can improve the generalizability of models in the healthy-to-PwPD transfer scenario. The simple/moderately complex architectures tested in this study could generalize better to the PwPD domain when trained on a healthy dataset compared to the most complex architectures used. The findings of this study could contribute to the development of accurate wearable-based activity monitoring solutions for PwPD, improving clinical decision-making and patient outcomes based on patient activity levels.

Boundary uncertainty aware network for automated polyp segmentation

Recently, leveraging deep neural networks for automated colorectal polyp segmentation has emerged as a hot topic due to the favored advantages in evading the limitations of visual inspection, e.g., overwork and subjectivity. However, most existing methods do not pay enough attention to the uncertain areas of colonoscopy images and often provide unsatisfactory segmentation performance. In this paper, we propose a novel boundary uncertainty aware network (BUNet) for precise and robust colorectal polyp segmentation. Specifically, considering that polyps vary greatly in size and shape, we first adopt a pyramid vision transformer encoder to learn multi-scale feature representations. Then, a simple yet effective boundary exploration module (BEM) is proposed to explore boundary cues from the low-level features. To make the network focus on the ambiguous area where the prediction score is biased to neither the foreground nor the background, we further introduce a boundary uncertainty aware module (BUM) that explores error-prone regions from the high-level features with the assistance of boundary cues provided by the BEM. Through the top-down hybrid deep supervision, our BUNet implements coarse-to-fine polyp segmentation and finally localizes polyp regions precisely. Extensive experiments on five public datasets show that BUNet is superior to thirteen competing methods in terms of both effectiveness and generalization ability.

Weakly supervised image segmentation beyond tight bounding box annotations

Weakly supervised image segmentation approaches in the literature usually achieve high segmentation performance using tight bounding box supervision and decrease the performance greatly when supervised by loose bounding boxes. However, compared with loose bounding box, it is much more difficult to acquire tight bounding box due to its strict requirements on the precise locations of the four sides of the box. To resolve this issue, this study investigates whether it is possible to maintain good segmentation performance when loose bounding boxes are used as supervision. For this purpose, this work extends our previous parallel transformation based multiple instance learning (MIL) for tight bounding box supervision by integrating an MIL strategy based on polar transformation to assist image segmentation. The proposed polar transformation based MIL formulation works for both tight and loose bounding boxes, in which a positive bag is defined as pixels in a polar line of a bounding box with one endpoint located inside the object enclosed by the box and the other endpoint located at one of the four sides of the box. Moreover, a weighted smooth maximum approximation is introduced to incorporate the observation that pixels closer to the origin of the polar transformation are more likely to belong to the object in the box. The proposed approach was evaluated on two public datasets using dice coefficient when bounding boxes at different precision levels were considered in the experiments. The results demonstrate that the proposed approach achieves state-of-the-art performance for bounding boxes at all precision levels and is robust to mild and moderate errors in the loose bounding box annotations. The codes are available at https://github.com/wangjuan313/wsis-beyond-tightBB.

Indirect estimation of pediatric reference interval via density graph deep embedded clustering

Establishing reference intervals (RIs) for pediatric patients is crucial in clinical decision-making, and there is a critical gap of pediatric RIs in China. However, the direct sampling technique for establishing RIs is resource-intensive and ethically challenging. Indirect estimation methods, such as unsupervised clustering algorithms, have emerged as potential alternatives for predicting reference intervals. This study introduces deep graph clustering methods into indirect estimation of pediatric reference intervals. Specifically, we propose a Density Graph Deep Embedded Clustering (DGDEC) algorithm, which incorporates a density feature extractor to enhance sample representation and provides additional perspectives for distinguishing different levels of health status among populations. Additionally, we construct an adjacency matrix by computing the similarity between samples after feature enhancement. The DGDEC algorithm leverages the adjacency matrix to capture the interrelationships between patients and divides patients into different groups, thereby estimating reference intervals for the potential healthy population. The experimental results demonstrate that when compared to other indirect estimation techniques, our method ensures the predicted pediatric reference intervals in different age and gender groups are closer to the true values while maintaining good generalization performance. Additionally, through ablation experiments, our study confirms that the similarity between patients and the multi-scale density features of samples can effectively describe the potential health status of patients.

Star algorithm for neural network ensembling

Neural network ensembling is a common and robust way to increase model efficiency. In this paper, we propose a new neural network ensemble algorithm based on Audibert's empirical star algorithm. We provide optimal theoretical minimax bound on the excess squared risk. Additionally, we empirically study this algorithm on regression and classification tasks and compare it to most popular ensembling methods.

An uncertainty approach for Electric Submersible Pump modeling through Deep Neural Network

This work proposes a new methodology to identify and validate deep learning models for artificial oil lift systems that use submersible electric pumps. The proposed methodology allows for obtaining the models and evaluating the prediction's uncertainty jointly and systematically. The methodology employs a nonlinear model to generate training and validation data and the Markov Chain Monte Carlo algorithm to assess the neural network's epistemic uncertainty. The nonlinear model was used to overcome the limitations of the need for big datasets for training deep learning models. However, the developed models are validated against experimental data after training and validation with synthetic data. The validation is also performed through the models' uncertainty assessment and experimental data. From the implementation point of view, the method was coded in Python with Tensorflow and Keras libraries used to build the neural Networks and find the hyperparameters. The results show that the proposed methodology obtained models representing both the nonlinear model's dynamic behavior and the experimental data. It provides a most probable value close to the experimental data, and the uncertainty of the generated deep learning models has the same order of magnitude as that of the nonlinear model. This uncertainty assessment shows that the built models were adequately validated. The proposed deep learning models can be applied in several applications requiring a reliable and computationally lighter model. Hence, the obtained AI dynamic models can be employed for digital twin construction, control, and optimization.

Learning self-supervised molecular representations for drug-drug interaction prediction

Drug-drug interactions (DDI) are a critical concern in healthcare due to their potential to cause adverse effects and compromise patient safety. Supervised machine learning models for DDI prediction need to be optimized to learn abstract, transferable features, and generalize to larger chemical spaces, primarily due to the scarcity of high-quality labeled DDI data. Inspired by recent advances in computer vision, we present SMR-DDI, a self-supervised framework that leverages contrastive learning to embed drugs into a scaffold-based feature space. Molecular scaffolds represent the core structural motifs that drive pharmacological activities, making them valuable for learning informative representations. Specifically, we pre-trained SMR-DDI on a large-scale unlabeled molecular dataset. We generated augmented views for each molecule via SMILES enumeration and optimized the embedding process through contrastive loss minimization between views. This enables the model to capture relevant and robust molecular features while reducing noise. We then transfer the learned representations for the downstream prediction of DDI. Experiments show that the new feature space has comparable expressivity to state-of-the-art molecular representations and achieved competitive DDI prediction results while training on less data. Additional investigations also revealed that pre-training on more extensive and diverse unlabeled molecular datasets improved the model's capability to embed molecules more effectively. Our results highlight contrastive learning as a promising approach for DDI prediction that can identify potentially hazardous drug combinations using only structural information.

Recognition of eye diseases based on deep neural networks for transfer learning and improved D-S evidence theory

BACKGROUND

Human vision has inspired significant advancements in computer vision, yet the human eye is prone to various silent eye diseases. With the advent of deep learning, computer vision for detecting human eye diseases has gained prominence, but most studies have focused only on a limited number of eye diseases.

RESULTS

Our model demonstrated a reduction in inherent bias and enhanced robustness. The fused network achieved an Accuracy of 0.9237, Kappa of 0.878, F1 Score of 0.914 (95% CI [0.875-0.954]), Precision of 0.945 (95% CI [0.928-0.963]), Recall of 0.89 (95% CI [0.821-0.958]), and an AUC value of ROC at 0.987. These metrics are notably higher than those of comparable studies.

CONCLUSIONS

Our deep neural network-based model exhibited improvements in eye disease recognition metrics over models from peer research, highlighting its potential application in this field.

METHODS

In deep learning-based eye recognition, to improve the learning efficiency of the model, we train and fine-tune the network by transfer learning. In order to eliminate the decision bias of the models and improve the credibility of the decisions, we propose a model decision fusion method based on the D-S theory. However, D-S theory is an incomplete and conflicting theory, we improve and eliminate the existed paradoxes, propose the improved D-S evidence theory(ID-SET), and apply it to the decision fusion of eye disease recognition models.

SVFR: A novel slice-to-volume feature representation framework using deep neural networks and a clustering model for the diagnosis of Alzheimer's disease

Deep neural networks (DNNs) have been effective in classifying structural magnetic resonance imaging (sMRI) images for Alzheimer's disease (AD) diagnosis. In this study, we propose a novel two-phase slice-to-volume feature representation (SVFR) framework for AD diagnosis. Specifically, we design a slice-level feature extractor to automatically select informative slice images and extract their slice-level features, by combining DNN and clustering models. Furthermore, we propose a joint volume-level feature generator and classifier to hierarchically aggregate the slice-level features into volume-level features and to classify images, by devising a spatial pyramid set pooling module and a fusion module. Experimental results demonstrate the superior performance of the proposed SVFR, surpassing the majority of the state-of-the-art methods and achieving comparable results to the best-performing approach. Experimental results also showcase the efficacy of the slice-level feature extractor in the selection of informative slice images, as well as the effectiveness of the volume-level feature generator and classifier in the integration of slice-level features for image classification. The source code for this study is publicly available at https://github.com/gll89/SVFR.

Discovering individual fingerprints in resting-state functional connectivity using deep neural networks

Non-negligible idiosyncrasy due to interindividual differences is an ongoing issue in resting-state functional MRI (rfMRI) analysis. We show that a deep neural network (DNN) can be employed for individual identification by learning important features from the time-varying functional connectivity (FC) of rfMRI in the Human Connectome Project. We employed the trained DNN to identify individuals from an independent dataset acquired at our institution. The results revealed that the DNN could successfully identify 300 individuals with an error rate of 2.9% using 15 s time-window and 870 individuals with an error rate of 6.7%. A trained DNN with nonlinear hidden layers led to the proposal of the "fingerprint of FC" (fpFC) as representative edges of individual FC. The fpFCs for individuals exhibited commonly important and individual-specific edges across time-window lengths (from 5 min to 15 s). Furthermore, the utility of our model for another group of subjects was validated, supporting the feasibility of our technique in the context of transfer learning. In conclusion, our study offers an insight into the discovery of the intrinsic mode of the human brain using whole-brain resting-state FC and DNNs.

Advances in medical image analysis with vision Transformers: A comprehensive review

The remarkable performance of the Transformer architecture in natural language processing has recently also triggered broad interest in Computer Vision. Among other merits, Transformers are witnessed as capable of learning long-range dependencies and spatial correlations, which is a clear advantage over convolutional neural networks (CNNs), which have been the de facto standard in Computer Vision problems so far. Thus, Transformers have become an integral part of modern medical image analysis. In this review, we provide an encyclopedic review of the applications of Transformers in medical imaging. Specifically, we present a systematic and thorough review of relevant recent Transformer literature for different medical image analysis tasks, including classification, segmentation, detection, registration, synthesis, and clinical report generation. For each of these applications, we investigate the novelty, strengths and weaknesses of the different proposed strategies and develop taxonomies highlighting key properties and contributions. Further, if applicable, we outline current benchmarks on different datasets. Finally, we summarize key challenges and discuss different future research directions. In addition, we have provided cited papers with their corresponding implementations in https://github.com/mindflow-institue/Awesome-Transformer.

Use of extreme gradient boosting, light gradient boosting machine, and deep neural networks to evaluate the activity stage of extraocular muscles in thyroid-associated ophthalmopathy

PURPOSE

To develop a machine learning model to evaluate the activity stage of extraocular muscles in thyroid-associated ophthalmopathy (TAO).

METHODS

This study retrospectively analysed data from patients with TAO who underwent contrast-enhanced magnetic resonance imaging (MRI) from 2015 to 2022. Three independent machine learning models, namely, extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), and deep neural networks (DNNs), were constructed using common clinical features. The performance of these models was compared using evaluation metrics such as the area under the receiver operating curve (AUC), accuracy, precision, recall, and F1 score. The importance of features was explained using Shapley additive explanations (SHAP).

RESULTS

A total of 2561 eyes of 1479 TAO patients were included in this study. The original dataset was randomly divided into a training set (80%, n = 2048) and a test set (20%, n = 513). In the performance evaluation of the test set, the LightGBM model had the best diagnostic performance (AUC 0.9260). According to the SHAP results, features such as conjunctival congestion, swollen caruncles, oedema of the upper eyelid, course of TAO, and intraocular pressure had the most significant impact on the LightGBM model.

CONCLUSION

This study used contrast-enhanced MRI as an objective evaluation criterion and constructed a LightGBM model based on readily accessible clinical data. The model had good classification performance, making it a promising artificial intelligence (AI)-assisted tool to help community hospitals evaluate the inflammatory activity of extraocular muscles in TAO patients in a timely manner.

A review of signal processing and machine learning techniques for interictal epileptiform discharge detection

Brain interictal epileptiform discharges (IEDs), as one of the hallmarks of epileptic brain, are transient events captured by electroencephalogram (EEG). IEDs are generated by seizure networks, and they occur between seizures (interictal periods). The development of a robust method for IED detection could be highly informative for clinical treatment procedures and epileptic patient management. Since 1972, different machine learning techniques, from template matching to deep learning, have been developed to automatically detect IEDs from scalp EEG (scEEG) and intracranial EEG (iEEG). While the scEEG signals suffer from low information details and high attenuation of IEDs due to the high skull electrical impedance, the iEEG signals recorded using implanted electrodes enjoy higher details and are more suitable for identifying the IEDs. In this review paper, we group IED detection techniques into six categories: (1) template matching, (2) feature representation (mimetic, time-frequency, and nonlinear features), (3) matrix decomposition, (4) tensor factorization, (5) neural networks, and (6) estimation of the iEEG from the concurrent scEEG followed by detection and classification. The methods are compared quantitatively (e.g., in terms of accuracy, sensitivity, and specificity), and their general advantages and limitations are described. Finally, current limitations and possible future research paths related to this field are mentioned.

GA-Net: A geographical attention neural network for the segmentation of body torso tissue composition

PURPOSE

Body composition analysis (BCA) of the body torso plays a vital role in the study of physical health and pathology and provides biomarkers that facilitate the diagnosis and treatment of many diseases, such as type 2 diabetes mellitus, cardiovascular disease, obstructive sleep apnea, and osteoarthritis. In this work, we propose a body composition tissue segmentation method that can automatically delineate those key tissues, including subcutaneous adipose tissue, skeleton, skeletal muscle tissue, and visceral adipose tissue, on positron emission tomography/computed tomography scans of the body torso.

METHODS

To provide appropriate and precise semantic and spatial information that is strongly related to body composition tissues for the deep neural network, first we introduce a new concept of the body area and integrate it into our proposed segmentation network called Geographical Attention Network (GA-Net). The body areas are defined following anatomical principles such that the whole body torso region is partitioned into three non-overlapping body areas. Each body composition tissue of interest is fully contained in exactly one specific minimal body area. Secondly, the proposed GA-Net has a novel dual-decoder schema that is composed of a tissue decoder and an area decoder. The tissue decoder segments the body composition tissues, while the area decoder segments the body areas as an auxiliary task. The features of body areas and body composition tissues are fused through a soft attention mechanism to gain geographical attention relevant to the body tissues. Thirdly, we propose a body composition tissue annotation approach that takes the body area labels as the region of interest, which significantly improves the reproducibility, precision, and efficiency of delineating body composition tissues.

RESULTS

Our evaluations on 50 low-dose unenhanced CT images indicate that GA-Net outperforms other architectures statistically significantly based on the Dice metric. GA-Net also shows improvements for the 95% Hausdorff Distance metric in most comparisons. Notably, GA-Net exhibits more sensitivity to subtle boundary information and produces more reliable and robust predictions for such structures, which are the most challenging parts to manually mend in practice, with potentially significant time-savings in the post hoc correction of these subtle boundary placement errors. Due to the prior knowledge provided from body areas, GA-Net achieves competitive performance with less training data. Our extension of the dual-decoder schema to TransUNet and 3D U-Net demonstrates that the new schema significantly improves the performance of these classical neural networks as well. Heatmaps obtained from attention gate layers further illustrate the geographical guidance function of body areas for identifying body tissues.

CONCLUSIONS

(i) Prior anatomic knowledge supplied in the form of appropriately designed anatomic container objects significantly improves the segmentation of bodily tissues. (ii) Of particular note are the improvements achieved in the delineation of subtle boundary features which otherwise would take much effort for manual correction. (iii) The method can be easily extended to existing networks to improve their accuracy for this application.

Data-driven supervised machine learning to predict the compressive response of porous PVA/Gelatin hydrogels and in-vitro assessments: Employing design of experiments

The purpose of this study is to design and evaluate a series of porous hydrogels by considering three independent variables using the Box-Behnken method. Accordingly, concentrations of the constituent macromolecules of the hydrogels, Polyvinyl Alcohol and Gelatin, and concentration of the crosslinking agent are varied to fabricate sixteen different porous samples utilizing the lyophilization process. Subsequently, the porous hydrogels are subjected to a battery of tests, including Fourier Transform Infrared spectroscopy, morphology assessment, pore-size study, porosimetry, uniaxial compression, and swelling measurements. Additionally, in-vitro cell assessments are performed by culturing mouse fibroblast cells (L-929) on the hydrogels, where viability, proliferation, adhesion, and morphology of the L-929 cells are monitored over 24, 48, and 72 h to evaluate the biocompatibility of these biomaterials. To better understand the mechanical behavior of the hydrogels under compressive loadings, Deep Neural Networks (DNNs) are implemented to predict and capture their compressive stress-strain responses as a function of the constituent materials' concentrations and duration of the performed mechanical tests. Overall, this study emphasizes the importance of considering multiple variables in the design of porous hydrogels, provides a comprehensive evaluation of their mechanical and biological properties, and, particularly, implements DNNs in the prediction of the hydrogels' stress-strain responses.

Voice spoofing detection using a neural networks assembly considering spectrograms and mel frequency cepstral coefficients

Nowadays, biometric authentication has gained relevance due to the technological advances that have allowed its inclusion in many daily-use devices. However, this same advantage has also brought dangers, as spoofing attacks are now more common. This work addresses the vulnerabilities of automatic speaker verification authentication systems, which are prone to attacks arising from new techniques for the generation of spoofed audio. In this article, we present a countermeasure for these attacks using an approach that includes easy to implement feature extractors such as spectrograms and mel frequency cepstral coefficients, as well as a modular architecture based on deep neural networks. Finally, we evaluate our proposal using the well-know ASVspoof 2017 V2 database, the experiments show that using the final architecture the best performance is obtained, achieving an equal error rate of 6.66% on the evaluation set.

Gray-to-color image conversion in the classification of breast lesions on ultrasound using pre-trained deep neural networks

Breast ultrasound (BUS) image classification in benign and malignant classes is often based on pre-trained convolutional neural networks (CNNs) to cope with small-sized training data. Nevertheless, BUS images are single-channel gray-level images, whereas pre-trained CNNs learned from color images with red, green, and blue (RGB) components. Thus, a gray-to-color conversion method is applied to fit the BUS image to the CNN's input layer size. This paper evaluates 13 gray-to-color conversion methods proposed in the literature that follow three strategies: replicating the gray-level image to all RGB channels, decomposing the image to enhance inherent information like the lesion's texture and morphology, and learning a matching layer. Besides, we introduce an image decomposition method based on the lesion's structural information to describe its inner and outer complexity. These gray-to-color conversion methods are evaluated under the same experimental framework using a pre-trained CNN architecture named ResNet-18 and a BUS dataset with more than 3000 images. In addition, the Matthews correlation coefficient (MCC), sensitivity (SEN), and specificity (SPE) measure the classification performance. The experimental results show that decomposition methods outperform replication and learning-based methods when using information from the lesion's binary mask (obtained from a segmentation method), reaching an MCC value greater than 0.70 and specificity up to 0.92, although the sensitivity is about 0.80. On the other hand, regarding the proposed method, the trade-off between sensitivity and specificity is better balanced, obtaining about 0.88 for both indices and an MCC of 0.73. This study contributes to the objective assessment of different gray-to-color conversion approaches in classifying breast lesions, revealing that mask-based decomposition methods improve classification performance. Besides, the proposed method based on structural information improves the sensitivity, obtaining more reliable classification results on malignant cases and potentially benefiting clinical practice.

A framework for generalizable neural networks for robust estimation of eyelids and pupils

Deep neural networks (DNNs) have enabled recent advances in the accuracy and robustness of video-oculography. However, to make robust predictions, most DNN models require extensive and diverse training data, which is costly to collect and label. In this work, we seek to improve the codevelop pylids, a pupil- and eyelid-estimation DNN model based on DeepLabCut. We show that performance of pylids-based pupil estimation can be related to the distance of test data from the distribution of training data. Based on this principle, we explore methods for efficient data selection for training our DNN. We show that guided sampling of new data points from the training data approaches state-of-the-art pupil and eyelid estimation with fewer training data points. We also demonstrate the benefit of using an efficient sampling method to select data augmentations for training DNNs. These sampling methods aim to minimize the time and effort required to label and train DNNs while promoting model generalization on new diverse datasets.

Predicting anticancer synergistic drug combinations based on multi-task learning

BACKGROUND

The discovery of anticancer drug combinations is a crucial work of anticancer treatment. In recent years, pre-screening drug combinations with synergistic effects in a large-scale search space adopting computational methods, especially deep learning methods, is increasingly popular with researchers. Although achievements have been made to predict anticancer synergistic drug combinations based on deep learning, the application of multi-task learning in this field is relatively rare. The successful practice of multi-task learning in various fields shows that it can effectively learn multiple tasks jointly and improve the performance of all the tasks.

METHODS

In this paper, we propose MTLSynergy which is based on multi-task learning and deep neural networks to predict synergistic anticancer drug combinations. It simultaneously learns two crucial prediction tasks in anticancer treatment, which are synergy prediction of drug combinations and sensitivity prediction of monotherapy. And MTLSynergy integrates the classification and regression of prediction tasks into the same model. Moreover, autoencoders are employed to reduce the dimensions of input features.

RESULTS

Compared with the previous methods listed in this paper, MTLSynergy achieves the lowest mean square error of 216.47 and the highest Pearson correlation coefficient of 0.76 on the drug synergy prediction task. On the corresponding classification task, the area under the receiver operator characteristics curve and the area under the precision-recall curve are 0.90 and 0.62, respectively, which are equivalent to the comparison methods. Through the ablation study, we verify that multi-task learning and autoencoder both have a positive effect on prediction performance. In addition, the prediction results of MTLSynergy in many cases are also consistent with previous studies.

CONCLUSION

Our study suggests that multi-task learning is significantly beneficial for both drug synergy prediction and monotherapy sensitivity prediction when combining these two tasks into one model. The ability of MTLSynergy to discover new anticancer synergistic drug combinations noteworthily outperforms other state-of-the-art methods. MTLSynergy promises to be a powerful tool to pre-screen anticancer synergistic drug combinations.

Data mining techniques in breast cancer diagnosis at the cellular-molecular level

INTRODUCTION

Studies in the field of better diagnosis of breast cancer using machine learning and data mining techniques have always been promising. A new diagnostic method can detect the characteristics of breast cancer in the early stages and help in better treatment. The aim of this study is to provide a method for early detection of breast cancer by reducing human errors based on data mining techniques in medicine using accurate and rapid screening.

METHODOLOGY

The proposed method includes data pre-processing and image quality improvement in the first step. The second step consists of separating cancer cells from healthy breast tissue and removing outliers using image segmentation. Finally, a classification model is configured by combining deep neural networks in the third phase. The proposed ensemble classification model uses several effective features extracted from images and is based on majority vote. This model can be used as a screening system to diagnose the grade of invasive ductal carcinoma of the breast.

RESULTS

Evaluations have been done using two histopathological microscopic datasets including patients with invasive ductal carcinoma of the breast. With extracting high-level features with average accuracies of 92.65% and 93.34% in these two datasets, the proposed method has succeeded in quickly diagnosing and classifying breast cancer with high performance.

CONCLUSION

By combining deep neural networks and extracting features affecting breast cancer, the ability to diagnose with the highest accuracy is provided, and this is a step toward helping specialists and increasing the chances of patients' survival.

Development of generalizable automatic sleep staging using heart rate and movement based on large databases

Purpose

With the advancement of deep neural networks in biosignals processing, the performance of automatic sleep staging algorithms has improved significantly. However, sleep staging using only non-electroencephalogram features has not been as successful, especially following the current American Association of Sleep Medicine (AASM) standards. This study presents a fine-tuning based approach to widely generalizable automatic sleep staging using heart rate and movement features trained and validated on large databases of polysomnography.

Methods

A deep neural network is used to predict sleep stages using heart rate and movement features. The model is optimized on a dataset of 8731 nights of polysomnography recordings labeled using the Rechtschaffen & Kales scoring system, and fine-tuned to a smaller dataset of 1641 AASM-labeled recordings. The model prior to and after fine-tuning is validated on two AASM-labeled external datasets totaling 1183 recordings. In order to measure the performance of the model, the output of the optimized model is compared to reference expert-labeled sleep stages using accuracy and Cohen's κ as key metrics.

Results

The fine-tuned model showed accuracy of 76.6% with Cohen's κ of 0.606 in one of the external validation datasets, outperforming a previously reported result, and showed accuracy of 81.0% with Cohen's κ of 0.673 in another external validation dataset.

Conclusion

These results indicate that the proposed model is generalizable and effective in predicting sleep stages using features which can be extracted from non-contact sleep monitors. This holds valuable implications for future development of home sleep evaluation systems.

Enhanced joint hybrid deep neural network explainable artificial intelligence model for 1-hr ahead solar ultraviolet index prediction

BACKGROUND AND OBJECTIVE

Exposure to solar ultraviolet (UV) radiation can cause malignant keratinocyte cancer and eye disease. Developing a user-friendly, portable, real-time solar UV alert system especially or wearable electronic mobile devices can help reduce the exposure to UV as a key measure for personal and occupational management of the UV risks. This research aims to design artificial intelligence-inspired early warning tool tailored for short-term forecasting of UV index (UVI) integrating satellite-derived and ground-based predictors for Australian hotspots receiving high UV exposures. The study further improves the trustworthiness of the newly designed tool using an explainable artificial intelligence approach.

METHODS

An enhanced joint hybrid explainable deep neural network model (called EJH-X-DNN) is constructed involving two phases of feature selection and hyperparameter tuning using Bayesian optimization. A comprehensive assessment of EJH-X- DNN is conducted with six other competing benchmarked models. The proposed model is explained locally and globally using robust model-agnostic explainable artificial intelligence frameworks such as Local Interpretable Model-Agnostic Explanations (LIME), Shapley additive explanations (SHAP), and permutation feature importance (PFI).

RESULTS

The newly proposed model outperformed all benchmarked models for forecasting hourly horizons UVI, with correlation coefficients of 0.900, 0.960, 0.897, and 0.913, respectively, for Darwin, Alice Springs, Townsville, and Emerald hotspots. According to the combined local and global explainable model outcomes, the site-based results indicate that antecedent lagged memory of UVI and solar zenith angle are influential features. Predictions made by EJH-X-DNN model are strongly influenced by factors such as ozone effect, cloud conditions, and precipitation.

CONCLUSION

With its superiority and skillful interpretation, the UVI prediction system reaffirms its benefits for providing real-time UV alerts to mitigate risks of skin and eye health complications, reducing healthcare costs and contributing to outdoor exposure policy.

Deep learning based source imaging provides strong sublobar localization of epileptogenic zone from MEG interictal spikes

Electromagnetic source imaging (ESI) offers unique capability of imaging brain dynamics for studying brain functions and aiding the clinical management of brain disorders. Challenges exist in ESI due to the ill-posedness of the inverse problem and thus the need of modeling the underlying brain dynamics for regularizations. Advances in generative models provide opportunities for more accurate and realistic source modeling that could offer an alternative approach to ESI for modeling the underlying brain dynamics beyond equivalent physical source models. However, it is not straightforward to explicitly formulate the knowledge arising from these generative models within the conventional ESI framework. Here we investigate a novel source imaging framework based on mesoscale neuronal modeling and deep learning (DL) that can learn the sensor-source mapping relationship directly from MEG data for ESI. Two DL-based ESI models were trained based on data generated by neural mass models and either generic or personalized head models. The robustness of the two DL models was evaluated by systematic computer simulations and clinical validation in a cohort of 29 drug-resistant focal epilepsy patients who underwent intracranial EEG (iEEG) evaluation or surgical resection. Results estimated from pre-operative MEG interictal spikes were quantified using the overlap with resection regions and the distance to the seizure-onset zone (SOZ) defined by iEEG recordings. The DL-based ESI provided robust results when no personalized head geometry is considered, reaching a spatial dispersion of 21.90 ± 19.03 mm, sublobar concordance of 83 %, and sublobar sensitivity and specificity of 66 and 97 % respectively. When using personalized head geometry derived from individual patients' MRI in the training data, personalized DL-based ESI model can further improve the performance and reached a spatial dispersion of 8.19 ± 8.14 mm, sublobar concordance of 93 %, and sublobar sensitivity and specificity of 77 and 99 % respectively. When compared to the SOZ, the localization error of the personalized approach is 15.78 ± 5.54 mm, outperforming the conventional benchmarks. This work demonstrates that combining generative models and deep learning enables an accurate and robust imaging of epileptogenic zone from MEG recordings with strong sublobar precision, suggesting its added value to enhancing MEG source localization and imaging, and to epilepsy source localization and other clinical applications.

Forecasting actual evapotranspiration without climate data based on stacked integration of DNN and meta-heuristic models across China from 1958 to 2021

As a non-linear phenomenon that varies along with agro-climatic conditions alongside many other factors, Evapotranspiration (ET) process represents a complexity when be assessed especially if there is a data scarcity in the weather data. However, even under such a data scarcity, the accurate estimates of ET values remain necessary for precise irrigation. So, the present study aims to: i) evaluate the performance of six hybrid machine learning (ML) models in estimating the monthly actual ET values under different agro-climatic conditions in China for seven provinces (Shandong, Jiangsu, Zhejiang, Fujian, Jiangxi, Hubei, and Henan), and ii) select the best-developed model based on statistical metrics and reduce errors between predicted and actual ET (AET) values. AET datasets were divided into 78% for model training (from 1958 to 2007) and the remaining was used for testing (from 2008 to 2021). Deep Neural Networks (DNN) was used as a standalone model at first then the stacking method was applied to integrate DNN with data-driven models such as Additive regression (AR), Random Forest (RF), Random Subspace (RSS), M5 Burned Tree (M5P) and Reduced Error Purning Tree (REPTree). Partial Auto-Correlation Function (PACF) was used for selection of the best lags inputs to the developed models. Results have revealed that DNN-based hybrid models held better performance than non-hybrid DNN models, such that the DNN-RF algorithm outperformed others during both training and testing stages, followed by DNN-RSS. This model has acquired the best values of every statistical measure [MAE (10.8, 12.9), RMSE (15.6, 17.4), RAE (31.9%, 41.4%), and RRSE (39.3%, 47.2%)] for training and testing, respectively. In contrast, the DNN model held the worst performance [MAE (14.9, 13.7), RMSE (20.1, 18.2), RAE (43.9%, 43.7%), and RRSE (50.6%, 49.3%)], for training and testing, respectively. Results from the study presented have revealed the capability of DNN-based hybrid models for long-term predictions of the AET values. Moreover, the DNN-RF model has been suggested as the most suitable model to improve future investigation for AET predictions, which could benefit the enhancement of the irrigation process and increase crop yield.

Predicting survival of Iranian COVID-19 patients infected by various variants including omicron from CT Scan images and clinical data using deep neural networks

Purpose: The rapid spread of the COVID-19 omicron variant virus has resulted in an overload of hospitals around the globe. As a result, many patients are deprived of hospital facilities, increasing mortality rates. Therefore, mortality rates can be reduced by efficiently assigning facilities to higher-risk patients. Therefore, it is crucial to estimate patients' survival probability based on their conditions at the time of admission so that the minimum required facilities can be provided, allowing more opportunities to be available for those who need them. Although radiologic findings in chest computerized tomography scans show various patterns, considering the individual risk factors and other underlying diseases, it is difficult to predict patient prognosis through routine clinical or statistical analysis. Method: In this study, a deep neural network model is proposed for predicting survival based on simple clinical features, blood tests, axial computerized tomography scan images of lungs, and the patients' planned treatment. The model's architecture combines a Convolutional Neural Network and a Long Short Term Memory network. The model was trained using 390 survivors and 108 deceased patients from the Rasoul Akram Hospital and evaluated 109 surviving and 36 deceased patients infected by the omicron variant. Results: The proposed model reached an accuracy of 87.5% on the test data, indicating survival prediction possibility. The accuracy was significantly higher than the accuracy achieved by classical machine learning methods without considering computerized tomography scan images (p-value <= 4E-5). The images were also replaced with hand-crafted features related to the ratio of infected lung lobes used in classical machine-learning models. The highest-performing model reached an accuracy of 84.5%, which was considerably higher than the models trained on mere clinical information (p-value <= 0.006). However, the performance was still significantly less than the deep model (p-value <= 0.016). Conclusion: The proposed deep model achieved a higher accuracy than classical machine learning methods trained on features other than computerized tomography scan images. This proves the images contain extra information. Meanwhile, Artificial Intelligence methods with multimodal inputs can be more reliable and accurate than computerized tomography severity scores.

Improved Giza pyramids construction algorithm for Modify the deep neural network-based method for energy demand forecasting

Accurate prediction of energy demand is crucial for improving services, reducing costs, and optimizing operations in energy systems. Deep neural networks (DNNs) have emerged as a popular method for energy demand forecasting. However, the performance of DNNs can be affected by data quality and hyperparameter selection. To address these concerns, this study proposes a novel energy demand forecasting technique that combines DNNs with an enhanced Giza pyramid construction methodology. The aim of this study is to provide a more reliable and effective approach for forecasting energy demand. The DNNs are employed to capture the complex relationships between input and output variables, while the Giza pyramids algorithm is utilized to optimal selection of hyperparameters of the network. Real-world energy demand data is used to evaluate the proposed approach, comparing it with state-of-the-art baseline models. The research methodology involves assessing the suggested approach using real-world energy demand information and conducting a comparative analysis with cutting-edge baseline models, including modified BP neural network (MBPNN), Neural Network based Genetic Algorithm (NNGA), and reinforcement learning and deep neural network (RLDNN). The IGPCA/CNN method outperforms other methods in energy prediction accuracy across short-term, medium-term, and long-term time scales. It achieves an MSE score of 0.564, lower than MBPNN, NNGA, and RLDNN. In medium-term prediction, it achieves an MSE score of 0.587, better than MBPNN, NNGA, and RLDNN. In long-term prediction, it achieves an MSE score of 0.629, lower than MBPNN and RLDNN. Further analysis and validation experiments are needed to ensure robustness and generalizability. Comparing the method with other state-of-the-art approaches can provide a comprehensive understanding of its superiority. The performance of the models is evaluated based on reliability and effectiveness in energy demand forecasting. The major conclusion of this study is that the proposed approach outperforms the initial models in accurately forecasting energy demand. The combination of DNNs and the improved Giza pyramid construction methodology results in enhanced performance, demonstrating superior reliability and effectiveness compared to other models. The study highlights the significance of accurate energy demand prediction for optimizing energy systems and reducing costs.

ReGen-DTI: A novel generative drug target interaction model for predicting potential drug candidates against SARS-COV2

Covid-19 has caused massive numbers of infections and fatalities globally. In response, there has been a large-scale experimental and computational research effort to study and develop drugs. Towards this, Deep learning techniques are used for the generation of potential novel drug candidates that are proven to be effective against exploring large molecular search spaces. Recent advances in reinforcement learning in conjunction with generative techniques has proven to be a promising field in the area of drug discovery. In this regard, we propose a generative drug discovery approach using reinforcement techniques for sampling novel molecules that bind to the main protease of SARS-COV2. The generative method reported significant validity scores for the generated novel molecules and captured the underlying features of the training molecules. Further, the model is fine-tuned on existing re-purposed molecules which are active towards specific target proteins based on similarity metrics. Upon fine tuning the model generated 92.71% valid, 93.55% unique, and 100% novel molecules. Unlike previous methods which are dependent on docking procedures, we proposed a deep learning based novel drug target interaction (DTI) model to find the binding affinity between candidate molecules and target protease sequence. Finally, the binding affinity of the generated molecules is predicted against the 3CLPro main protease by using the proposed DTI model. Most of the generated molecules have shown binding affinity scores <100 nM (lower the better), which are significantly better compared to the existing commercial drugs including Remdesevir.

Deep Ensembles Are Robust to Occasional Catastrophic Failures of Individual DNNs for Organs Segmentations in CT Images

Deep neural networks (DNNs) have recently showed remarkable performance in various computer vision tasks, including classification and segmentation of medical images. Deep ensembles (an aggregated prediction of multiple DNNs) were shown to improve a DNN's performance in various classification tasks. Here we explore how deep ensembles perform in the image segmentation task, in particular, organ segmentations in CT (Computed Tomography) images. Ensembles of V-Nets were trained to segment multiple organs using several in-house and publicly available clinical studies. The ensembles segmentations were tested on images from a different set of studies, and the effects of ensemble size as well as other ensemble parameters were explored for various organs. Compared to single models, Deep Ensembles significantly improved the average segmentation accuracy, especially for those organs where the accuracy was lower. More importantly, Deep Ensembles strongly reduced occasional "catastrophic" segmentation failures characteristic of single models and variability of the segmentation accuracy from image to image. To quantify this we defined the "high risk images": images for which at least one model produced an outlier metric (performed in the lower 5% percentile). These images comprised about 12% of the test images across all organs. Ensembles performed without outliers for 68%-100% of the "high risk images" depending on the performance metric used.

Increasing-Margin Adversarial (IMA) training to improve adversarial robustness of neural networks

BACKGROUND AND OBJECTIVE

Deep neural networks (DNNs) are vulnerable to adversarial noises. Adversarial training is a general and effective strategy to improve DNN robustness (i.e., accuracy on noisy data) against adversarial noises. However, DNN models trained by the current existing adversarial training methods may have much lower standard accuracy (i.e., accuracy on clean data), compared to the same models trained by the standard method on clean data, and this phenomenon is known as the trade-off between accuracy and robustness and is commonly considered unavoidable. This issue prevents adversarial training from being used in many application domains, such as medical image analysis, as practitioners do not want to sacrifice standard accuracy too much in exchange for adversarial robustness. Our objective is to lift (i.e., alleviate or even avoid) this trade-off between standard accuracy and adversarial robustness for medical image classification and segmentation.

METHODS

We propose a novel adversarial training method, named Increasing-Margin Adversarial (IMA) Training, which is supported by an equilibrium state analysis about the optimality of adversarial training samples. Our method aims to preserve accuracy while improving robustness by generating optimal adversarial training samples. We evaluate our method and the other eight representative methods on six publicly available image datasets corrupted by noises generated by AutoAttack and white-noise attack.

RESULTS

Our method achieves the highest adversarial robustness for image classification and segmentation with the smallest reduction in accuracy on clean data. For one of the applications, our method improves both accuracy and robustness.

CONCLUSIONS

Our study has demonstrated that our method can lift the trade-off between standard accuracy and adversarial robustness for the image classification and segmentation applications. To our knowledge, it is the first work to show that the trade-off is avoidable for medical image segmentation.

ImageCAS: A large-scale dataset and benchmark for coronary artery segmentation based on computed tomography angiography images

Cardiovascular disease (CVD) accounts for about half of non-communicable diseases. Vessel stenosis in the coronary artery is considered to be the major risk of CVD. Computed tomography angiography (CTA) is one of the widely used noninvasive imaging modalities in coronary artery diagnosis due to its superior image resolution. Clinically, segmentation of coronary arteries is essential for the diagnosis and quantification of coronary artery disease. Recently, a variety of works have been proposed to address this problem. However, on one hand, most works rely on in-house datasets, and only a few works published their datasets to the public which only contain tens of images. On the other hand, their source code have not been published, and most follow-up works have not made comparison with existing works, which makes it difficult to judge the effectiveness of the methods and hinders the further exploration of this challenging yet critical problem in the community. In this paper, we propose a large-scale dataset for coronary artery segmentation on CTA images. In addition, we have implemented a benchmark in which we have tried our best to implement several typical existing methods. Furthermore, we propose a strong baseline method which combines multi-scale patch fusion and two-stage processing to extract the details of vessels. Comprehensive experiments show that the proposed method achieves better performance than existing works on the proposed large-scale dataset. The benchmark and the dataset are published at https://github.com/XiaoweiXu/ImageCAS-A-Large-Scale-Dataset-and-Benchmark-for-Coronary-Artery-Segmentation-based-on-CT.

Decoding movement kinematics from EEG using an interpretable convolutional neural network

Continuous decoding of hand kinematics has been recently explored for the intuitive control of electroencephalography (EEG)-based Brain-Computer Interfaces (BCIs). Deep neural networks (DNNs) are emerging as powerful decoders, for their ability to automatically learn features from lightly pre-processed signals. However, DNNs for kinematics decoding lack in the interpretability of the learned features and are only used to realize within-subject decoders without testing other training approaches potentially beneficial for reducing calibration time, such as transfer learning. Here, we aim to overcome these limitations by using an interpretable convolutional neural network (ICNN) to decode 2-D hand kinematics (position and velocity) from EEG in a pursuit tracking task performed by 13 participants. The ICNN is trained using both within-subject and cross-subject strategies, and also testing the feasibility of transferring the knowledge learned on other subjects on a new one. Moreover, the network eases the interpretation of learned spectral and spatial EEG features. Our ICNN outperformed most of the other state-of-the-art decoders, showing the best trade-off between performance, size, and training time. Furthermore, transfer learning improved kinematics prediction in the low data regime. The network attributed the highest relevance for decoding to the delta-band across all subjects, and to higher frequencies (alpha, beta, low-gamma) for a cluster of them; contralateral central and parieto-occipital sites were the most relevant, reflecting the involvement of sensorimotor, visual and visuo-motor processing. The approach improved the quality of kinematics prediction from the EEG, at the same time allowing interpretation of the most relevant spectral and spatial features.