Enhancing plant morphological trait identification in herbarium collections through deep learning–based segmentation

This paper investigates the applicability of deep learning models for predicting the severity of forest wildfires, utilizing an innovative benchmark dataset called EO4WildFires. EO4WildFires integrates multispectral imagery from Sentinel-2, SAR data from Sentinel-1, and meteorological data from NASA Power annotated with EFFIS data for forest fire detection and size estimation. These data cover 45 countries with a total of 31,730 wildfire events from 2018 to 2022. All of these various sources of data are archived into data cubes, with the intention of assessing wildfire severity by considering both current and historical forest conditions, utilizing a broad range of data including temperature, precipitation, and soil moisture. The experimental setup has been arranged to test the effectiveness of different deep learning architectures in predicting the size and shape of wildfire-burned areas. This study incorporates both image segmentation networks and visual transformers, employing a consistent experimental design across various models to ensure the comparability of the results. Adjustments were made to the training data, such as the exclusion of empty labels and very small events, to refine the focus on more significant wildfire events and potentially improve prediction accuracy. The models’ performance was evaluated using metrics like F1 score, IoU score, and Average Percentage Difference (aPD). These metrics offer a multi-faceted view of model performance, assessing aspects such as precision, sensitivity, and the accuracy of the burned area estimation. Through extensive testing the final model utilizing LinkNet and ResNet-34 as backbones, we obtained the following metric results on the test set: 0.86 F1 score, 0.75 IoU, and 70% aPD. These results were obtained when all of the available samples were used. When the empty labels were absent during the training and testing, the model increased its performance significantly: 0.87 F1 score, 0.77 IoU, and 44.8% aPD. This indicates that the number of samples, as well as their respectively size (area), tend to have an impact on the model’s robustness. This restriction is well known in the remote sensing domain, as accessible, accurately labeled data may be limited. Visual transformers like TeleViT showed potential but underperformed compared to segmentation networks in terms of F1 and IoU scores.

Herbarium specimen scans constitute a valuable source of raw data. Herbarium collections are gaining interest in the scientific community as their exploration can lead to understanding serious threats to biodiversity. Data derived from scanned specimen images can be analyzed to answer important questions such as how plants respond to climate change, how different species respond to biotic and abiotic influences, or what role a species plays within an ecosystem. However, exploiting such large collections is challenging and requires automatic processing. A promising solution lies in the use of computer-based processing techniques, such as Deep Learning (DL). But herbarium specimens can be difficult to process and analyze as they contain several kinds of visual noise, including information labels, scale bars, color palettes, envelopes containing seeds or other organs, collection-specific barcodes, stamps, and other notes that are placed on the mounting sheet. Moreover, the paper on which the specimens are mounted can degrade over time for multiple reasons, and often the paper's color darkens and, in some cases, approaches the color of the plants.

This study aims to develop an attention-based deep learning model for distinguishing benign from malignant breast lesions on CESM. Preoperative CESM images of 1239 patients, which were definitely diagnosed on pathology in a multicentre cohort, were divided into training and validation sets, internal and external test sets. The regions of interest of the breast lesions were outlined manually by a senior radiologist. We adopted three conventional convolutional neural networks (CNNs), namely, DenseNet 121, Xception, and ResNet 50, as the backbone architectures and incorporated the convolutional block attention module (CBAM) into them for classification. The performance of the models was analysed in terms of the receiver operating characteristic (ROC) curve, accuracy, the positive predictive value (PPV), the negative predictive value (NPV), the F1 score, the precision recall curve (PRC), and heat maps. The final models were compared with the diagnostic performance of conventional CNNs, radiomics models, and two radiologists with specialised breast imaging experience. The best-performing deep learning model, that is, the CBAM-based Xception, achieved an area under the ROC curve (AUC) of 0.970, a sensitivity of 0.848, a specificity of 1.000, and an accuracy of 0.891 on the external test set, which was higher than those of other CNNs, radiomics models, and radiologists. The PRC and the heat maps also indicated the favourable predictive performance of the attention-based CNN model. The diagnostic performance of two radiologists improved with deep learning assistance. Using an attention-based deep learning model based on CESM images can help to distinguishing benign from malignant breast lesions, and the diagnostic performance of radiologists improved with deep learning assistance.

In recent years, the machine learning field has been inundated with a variety of deep learning methods. Different deep learning model types, including recurrent neural networks (RNNs), convolutional neural networks (CNNs), adversarial neural networks (ANNs), and autoencoders, are successfully tackling challenging computer vision problems including image detection and segmentation in an unconstrained environment. Although image segmentation has received a lot of interest, there have been several new deep learning methods discovered with regard to object detection and recognition. An academic review of deep learning image segmentation methods is presented in this article. In this study, the major goal is to offer a sensible comprehension of the basic approaches that have already made a substantial contribution to the domain of image segmentation throughout the years. The article describes the existing state of image segmentation, and goes on to make the argument that deep learning has revolutionized this field. Afterwards, segmentation algorithms have been scientifically classified and optimized, each with their own special contribution. With a variety of informative narratives, the reader may be able to understand the internal workings of these processes more quickly.

Deep learning methods have gained significant attention in sleep science. This study aimed to assess the performance of a deep learning-based sleep stage classification model constructed using fewer physiological parameters derived from cardiorespiratory and body movement data. Overnight polysomnography (PSG) data from 123 participants (age: 19–82 years) with suspected sleep disorders were analyzed. Multivariate time series data, including heart rate, respiratory rate, cardiorespiratory coupling, and body movement frequency, were input into a bidirectional long short-term memory (biLSTM) network model to train and predict five-class sleep stages. The trained model's performance was evaluated using balanced accuracy, Cohen's κ coefficient, and F1 scores on an epoch-per-epoch basis and compared with the ground truth using the leave-one-out cross-validation scheme. The model achieved an accuracy of 71.2 ± 5.8%, Cohen's κ of 0.425 ± 0.115, and an F1 score of 0.650 ± 0.083 across all sleep stages, and all metrics were negatively correlated with the apnea–hypopnea index, as well as age, but positively correlated with sleep efficiency. Moreover, the model performance varied for each sleep stage, with the highest F1 score observed for N2 and the lowest for N3. Regression and Bland–Altman analyses between sleep parameters of interest derived from deep learning and PSG showed substantial correlations (r = 0.33–0.60) with low bias. The findings demonstrate the efficacy of the biLSTM deep learning model in accurately classifying sleep stages and in estimating sleep parameters for sleep structure analysis using a reduced set of physiological parameters. The current model without using EEG information may expand the application of unobtrusive in-home monitoring to clinically assess the prevalence of sleep disorders outside of a sleep laboratory.

Objective:This work aimed to detect automatically periapical lesion on panoramic radiographs (PRs) using deep learning.Methods:454 objects in 357 PRs were anonymized and manually labeled. They are then pre-processed to improve image quality and enhancement purposes. The data were randomly assigned into the training, validation, and test folders with ratios of 0.8, 0.1, and 0.1, respectively. The state-of-art 10 different deep learning-based detection frameworks including various backbones were applied to periapical lesion detection problem. Model performances were evaluated by mean average precision, accuracy, precision, recall, F1 score, precision-recall curves, area under curve and several other Common Objects in Context detection evaluation metrics.Results:Deep learning-based detection frameworks were generally successful in detecting periapical lesions on PRs. Detection performance, mean average precision, varied between 0.832 and 0.953 while accuracy was between 0.673 and 0.812 for all models. F1 score was between 0.8 and 0.895. RetinaNet performed the best detection performance, similarly Adaptive Training Sample Selection provided F1 score of 0.895 as highest value. Testing with external data supported our findings.Conclusion:This work showed that deep learning models can reliably detect periapical lesions on PRs. Artificial intelligence-based on deep learning tools are revolutionizing dental healthcare and can help both clinicians and dental healthcare system.

Traditional Chinese medicine diagnostic prediction model for holistic syndrome differentiation based on deep learning

Multiclass autoencoder-based active learning for sensor-based human activity recognition

The primary aim is to develop a model that can achieve high accuracy in solving multi label classification problems by training, testing, and analyzing deep learning models that utilize both image and text data. In this paper, we propose a novel hybrid model that jointly considers textual and visual data for the task of movie genre prediction, which is a representative example of multi label classification problems. In the proposed model, textual features extracted from movie summaries are obtained using the DistilBERT model, while visual features derived from movie posters are extracted using the ConvNeXt deep learning model. These features from the two modalities are then combined using the XGBoost machine learning algorithm to perform genre prediction. This approach aims to achieve higher accuracy and better generalizability in movie genre classification by integrating information from different modalities through a late fusion method. The ConvNeXt architecture was adapted to the problem using transfer learning and fine-tuning techniques. To achieve the highest performance from the DistilBERT model, optimization was performed for the token length and threshold hyperparameters, and the model with a token length of 256 and a threshold of 0.5 was used in the hybrid model. Furthermore, to maximize the overall performance of the novel hybrid model, optimization was conducted using the Grid Search algorithm. All three proposed models were trained and tested on a dataset obtained from the IMDB website. The performances of the models were evaluated using hamming loss, precision and F1 score metrics. Experimental results revealed that, overall, the text-based model outperformed the image-based model, while the proposed hybrid model achieved higher performance than both individual models. It was demonstrated that textual and visual features complement each other and positively enhance the overall performance. This paper presents an original and effective study for multi label movie genre classification, combining the fields of computer vision, natural language processing, and machine learning.

Coverage-enhanced fault diagnosis for Deep Learning programs: A learning-based approach with hybrid metrics

Short-term solar radiation forecasting using hybrid deep residual learning and gated LSTM recurrent network with differential covariance matrix adaptation evolution strategy

In recent years, deep reinforcement learning has combined the advantages of reinforcement learning and deep learning, and has made great progress in decision-making tasks. However, the training of deep reinforcement learning requires frequent interactions between the agent and the environment and repeated experiments. Adversaries have chances to poison the sample data collected by the agent by attacking the experimental environment in the training process, thereby bringing security risks and serious consequences to the training process of reinforcement learning. This work is committed to addressing the security risks in the field of deep reinforcement learning. However, this work improves the algorithm from the perspective of sample data filtering, and improves the security performance of deep reinforcement learning algorithm. There are two contributions in this work: one is to defend against adversarial attacks against deep reinforcement learning through cluster analysis and sample value evaluation; the other is to propose a deep reinforcement learning algorithm based on sample value evaluation on the basis of deterministic strategy gradient algorithm. The algorithm uses the clustering method to classify the sample pool, and measures the contribution value and security risk of the sample to the model training through the sample value evaluation. The classic game experiments show that the proposed algorithm is safe and effective. It reduces the threat of the agent falling into the adversarial sample attack and improves the training performance of deep reinforcement learning.

Use of Machine Learning to Identify Follow-Up Recommendations in Radiology Reports

Deep learning (DL) has experienced an exponential development in recent years, with major impact in many medical fields, especially in the field of medical image and, respectively, as a specific task, in the segmentation of the medical image. We aim to create a computer assisted diagnostic method, optimized by the use of deep learning (DL) and validated by a randomized controlled clinical trial, is a highly automated tool for diagnosing and staging precancerous and cervical cancer and thyroid cancers. We aim to design a high-performance deep learning model, combined from convolutional neural network (U-Net)-based architectures, for segmentation of the medical image that is independent of the type of organs/tissues, dimensions or type of image (2D/3D) and to validate the DL model in a randomized, controlled clinical trial. We used as a methodology primarily the analysis of U-Net-based architectures to identify the key elements that we considered important in the design and optimization of the combined DL model, from the U-Net-based architectures, imagined by us. Secondly, we will validate the performance of the DL model through a randomized controlled clinical trial. The DL model designed by us will be a highly automated tool for diagnosing and staging precancers and cervical cancer and thyroid cancers. The combined model we designed takes into account the key features of each of the architectures Overcomplete Convolutional Network Kite-Net (Kite-Net), Attention gate mechanism is an improvement added on convolutional network architecture for fast and precise segmentation of images (Attention U-Net), Harmony Densely Connected Network-Medical image Segmentation (HarDNet-MSEG). In this regard, we will create a comprehensive computer assisted diagnostic methodology validated by a randomized controlled clinical trial. The model will be a highly automated tool for diagnosing and staging precancers and cervical cancer and thyroid cancers. This would help drastically minimize the time and effort that specialists put into analyzing medical images, help to achieve a better therapeutic plan, and can provide a “second opinion” of computer assisted diagnosis.

Forecasting the behavior of the stock market is a classic but difficult topic, one that has attracted the interest of both economists and computer scientists. Over the course of the last couple of decades, researchers have investigated linear models as well as models that are based on machine learning (ML), deep learning (DL), reinforcement learning (RL), and deep reinforcement learning (DRL) in order to create an accurate predictive model. Machine learning algorithms can now extract high-level financial market data patterns. Investors are using deep learning models to anticipate and evaluate stock and foreign exchange markets due to the advantage of artificial intelligence. Recent years have seen a proliferation of the deep reinforcement learning algorithm’s application in algorithmic trading. DRL agents, which combine price prediction and trading signal production, have been used to construct several completely automated trading systems or strategies. Our objective is to enable interested researchers to stay current and easily imitate earlier findings. In this paper, we have worked to explain the utility of Machine Learning, Deep Learning, Reinforcement Learning, and Deep Reinforcement Learning in Quantitative Finance (QF) and the Stock Market. We also outline potential future study paths in this area based on the overview that was presented before.