Public Datasets Research Articles

Cost-Aware Calibration of Classifiers

Most classification techniques in machine learning are able to produce probability predictions in addition to class predictions. However, these predicted probabilities are often not well calibrated in that they deviate from the actual outcome rates (i.e., the proportion of data instances that actually belong to a certain class). A lack of calibration can jeopardize downstream decision tasks that rely on accurate probability predictions. Although several post hoc calibration methods have been proposed, they generally do not consider the potentially asymmetric costs associated with overprediction versus underprediction. In this research, we formally define the problem of cost-aware calibration and propose a metric to quantify the cost of miscalibration for a given classifier. Next, we propose three approaches to achieve cost-aware calibration, two of which are cost-aware adaptations of existing calibration algorithms; the third one (named MetaCal) is a Bayes optimal learning algorithm inspired by prior work on cost-aware classification. We carry out systematic empirical evaluations on multiple public data sets to demonstrate the effectiveness of the proposed approaches in reducing the cost of miscalibration. Finally, we generalize the definition and metric as well as solution algorithms of cost-aware calibration to account for nonlinear cost structures that may arise in real-world decision tasks. Data Ethics & Reproducibility Note: There are no data ethics considerations. The code capsule is available on Code Ocean at https://doi.org/10.24433/CO.8552538.v1 and in the e-Companion to this article (available at https://doi.org/10.1287/ijds.2024.0038 ).

From Detection to Action: A Multimodal AI Framework for Traffic Incident Response

With the rising incidence of traffic accidents and growing environmental concerns, the demand for advanced systems to ensure traffic and environmental safety has become increasingly urgent. This paper introduces an automated highway safety management framework that integrates computer vision and natural language processing for real-time monitoring, analysis, and reporting of traffic incidents. The system not only identifies accidents but also aids in coordinating emergency responses, such as dispatching ambulances, fire services, and police, while simultaneously managing traffic flow. The approach begins with the creation of a diverse highway accident dataset, combining public datasets with drone and CCTV footage. YOLOv11s is retrained on this dataset to enable real-time detection of critical traffic elements and anomalies, such as collisions and fires. A vision–language model (VLM), Moondream2, is employed to generate detailed scene descriptions, which are further refined by a large language model (LLM), GPT 4-Turbo, to produce concise incident reports and actionable suggestions. These reports are automatically sent to relevant authorities, ensuring prompt and effective response. The system’s effectiveness is validated through the analysis of diverse accident videos and zero-shot simulation testing within the Webots environment. The results highlight the potential of combining drone and CCTV imagery with AI-driven methodologies to improve traffic management and enhance public safety. Future work will include refining detection models, expanding dataset diversity, and deploying the framework in real-world scenarios using live drone and CCTV feeds. This study lays the groundwork for scalable and reliable solutions to address critical traffic safety challenges.

Modelling Concept Drift in Dynamic Data Streams for Recommender Systems

Recommendation systems play a crucial role in modern e-commerce and streaming services. However, the limited availability of public datasets hampers the rapid development of more efficient and accurate recommendation algorithms within the research community. This work introduces a stream-based data generator designed to generate user preferences for a set of items while accommodating progressive changes in user preferences. The underlying principle involves using user/item embeddings to derive preferences by exploring the proximity of these embeddings. Whether randomly generated or learned from a real finite data stream, these embeddings serve as the basis for generating new preferences. We investigate how this fundamental model can adapt to shifts in user behavior over time; in our framework, changes correspond to alterations in the structure of the tripartite graph, reflecting modifications in the underlying embeddings. Through an analysis of real-life data streams, we demonstrate that the proposed model is effective in capturing actual preferences and the changes that they can exhibit over time. Thus, we characterize these changes and develop a generalized method capable of simulating realistic data, thereby generating streams with similar yet controllable drift dynamics.

UnifiedCut: A Simple and Efficient Neural Model for Thai, Burmese and Khmer Word Segmentation

Word segmentation is a critical task in natural language processing for southeast Asian Abugida languages, including Thai, Burmese, and Khmer. Existing approaches demonstrate that models using fixed-length windowed context inputs can achieve high segmentation accuracy; however, they often rely on low-level character features or language-specific preprocessing. Character-based methods can limit feature learning, while language-specific features add complexity due to specialized preprocessing requirements. This paper introduces UnifiedCut, which is a neural model that leverages multiple n-grams within a windowed multi-head attention mechanism. This design captures segmentation features from local contexts and multi-perspective n-gram inputs, enhancing generalization and recall, particularly for out-of-vocabulary words. Compared to CNN- and RNN-based approaches, UnifiedCut’s multi-head attention enables finer-grained feature extraction and greater parallelism, resulting in a faster, more scalable solution. Comprehensive experiments on public datasets for Thai, Burmese, and Khmer show that UnifiedCutachieves state-of-the-art performance in word segmentation.

MTGWNN: A Multi‐Template Graph Wavelet Neural Network Identification Model for Autism Spectrum Disorder

ABSTRACTFunctional magnetic resonance imaging (fMRI) has been widely applied in studying various brain disorders. However, current studies typically model regions of interest (ROIs) in brains with a single template. This approach generally examines only the connectivity between ROIs to identify autism spectrum disorder (ASD), ignoring the structural features of the brain. This study proposes a multi‐template graph wavelet neural network (GWNN) identification model for ASD called MTGWNN. First, the brain is segmented with multiple templates and the BOLD time series are extracted from fMRI data to construct brain networks. Next, a graph attention network (GAT) is applied to automatically learn interactions between nodes, capturing local information in the node features. These features are then further processed by a convolutional neural network (CNN) to learn global connectivity representations and achieve feature dimensionality reduction. Finally, the features and phenotypic data from each subject are integrated by GWNN to identify ASD at the optimal scale. Experimental results indicate that MTGWNN outperforms the comparative models. Testing on the public dataset ABIDE‐I achieved an accuracy (ACC) of 87.25% and an area under the curve (AUC) of 92.49%. MTGWNN effectively integrates brain network features from multiple templates, providing a more comprehensive characterization of brain abnormalities in patients with ASD. It incorporates population information from phenotypic data, which helps to compensate for the limited sample size of individual patients and improves the robustness and generalization of ASD identification.

Trends and variation in issuance of high-cost narcolepsy drugs by NHS England organisations and regions from 2019 to 2022.

Clinicians and people with narcolepsy report varied access to higher-cost narcolepsy treatments in England associated with variations in national and local commissioning. There are no publicly available data quantifying use of these drugs to support policy decisions. We therefore aimed to describe national, regional and local prescribing trends for higher-cost narcolepsy drugs using new national databases. We used the English prescribing dataset and secondary care medicines data to quantify volumes of high-cost narcolepsy drugs issued between 01 January 2019 and 31 December 2022. Volumes were converted to World Health Organisation defined daily doses, to estimate the monthly number of defined daily doses of sodium oxybate, pitolisant and solriamfetol issued by each integrated care board and region. We compared national, integrated care board, and regional level issuance of each drug over time. Analysis of almost 6000 primary care prescriptions and 2000 cumulative months of secondary care pharmacy stock data, issued across 41/42 integrated care boards in England, revealed a 49.1% increase in issuance of high-cost narcolepsy drugs between 2019 and 2022. In 2022, sodium oxybate accounted for 52.66% of issuance, pitolisant 43.09% and solriamfetol 4.25%, with 22.31% of defined daily doses issued in primary care. Three integrated care boards (NHS Southeast London, NHS Cumbria and North-East, NHS Cheshire and Merseyside) predominate, issuing 56.33% of all defined daily doses. Variations between integrated care boards and regions differ substantially by drug and route of issuance. Our findings describe substantial variation in the use of specialist narcolepsy drugs in England, and highlight the untapped potential of using large, public domain datasets to publicly review higher-cost drug prescribing.

Effect of patient-contextual skin images in human- and artificial intelligence-based diagnosis of melanoma: Results from the 2020 SIIM-ISIC melanoma classification challenge.

While the high accuracy of reported AI tools for melanoma detection is promising, the lack of holistic consideration of the patient is often criticized. Along with medical history, a dermatologist would also consider intra-patient nevi patterns, such that nevi that are different from others on a given patient are treated with suspicion. To evaluate whether patient-contextual lesion-images improves diagnostic accuracy for melanoma in a dermoscopic image-based AI competition and a human reader study. An international online AI competition was held in 2020. The task was to classify dermoscopy images as melanoma or benign lesions. A multi-source dataset of dermoscopy images grouped by patient were provided, and additional use of public datasets was permitted. Competitors were judged on area under the receiver operating characteristic (AUROC) on a private leaderboard. Concurrently, a human reader study was hosted using a subset of the test data. Participants gave their initial diagnosis of an index case (melanoma vs. benign) and were then presented with seven additional lesion-images of that patient before giving a second prediction of the index case. Outcome measures were sensitivity and specificity. The top 50 of 3308 AI competition entries achieved AUROC scores ranging from 0.943 to 0.949. Few algorithms considered intra-patient lesion patterns and instead most evaluated images independently. The median sensitivity and specificity of human readers before receiving contextual images were 60.0% and 86.7%, and after were 60.0% and 85.7%. Human and AI algorithm performance varied by image source. This study provided an open-source state-of-the-art algorithm for melanoma detection that has been evaluated at multiple centres. Patient-contextual images did not positively impact performance of AI algorithms or human readers. Providing seven contextual images and no total body image may have been insufficient to test the applicability of the intra-patient lesion patterns.

Local Pyramid Vision Transformer: Millimeter-Wave Radar Gesture Recognition Based on Transformer with Integrated Local and Global Awareness

A millimeter-wave radar is widely accepted by the public due to its low susceptibility to interference, such as changes in light, and the protection of personal privacy. With the development of the deep learning theory, the deep learning method has been dominant in the millimeter-wave radar field, which usually uses convolutional neural networks for feature extraction. In recent years, transformer networks have also been highly valued by researchers due to their parallel processing capabilities and long-distance dependency modeling capabilities. However, traditional convolutional neural networks (CNNs) and vision transformers each have their limitations: CNNs usually overlook the global features of images and vision transformers may neglect local image continuity, and both of them may impede gesture recognition performance. In addition, whether CNN or transformer, their implementation is hindered by the scarcity of public radar gesture datasets. To address these limitations, this paper proposes a new recognition method using a local pyramid visual transformer (LPVT) based on millimeter-wave radar. LPVT can capture global and local features in dynamic gesture spectrograms, ultimately improving the recognition ability of gestures. In this paper, we mainly carried out the following two tasks: building the corresponding datasets and executing gesture recognition. First, we constructed a gesture dataset for training. In this stage, we use a 77 GHz radar to collect the echo signals of gestures and preprocess them to build a dataset. Second, we propose the LPVT network specifically designed for gesture recognition tasks. By integrating local sensing into the globally focused transformer, we improve its capacity to capture both global and local features in dynamic gesture spectrograms. The experimental results using the dataset we constructed show that the proposed LPVT network achieved a gesture recognition accuracy of 92.2%, which exceeds the performance of other networks.

ChromTR: chromosome detection in raw metaphase cell images via deformable transformers.

Chromosome karyotyping is a critical way to diagnose various hematological malignancies and genetic diseases, of which chromosome detection in raw metaphase cell images is the most critical and challenging step. In this work, focusing on the joint optimization of chromosome localization and classification, we propose ChromTR to accurately detect and classify 24 classes of chromosomes in raw metaphase cell images. ChromTR incorporates semantic feature learning and class distribution learning into a unified DETR-based detection framework. Specifically, we first propose a Semantic Feature Learning Network (SFLN) for semantic feature extraction and chromosome foreground region segmentation with object-wise supervision. Next, we construct a Semantic-Aware Transformer (SAT) with two parallel encoders and a Semantic-Aware decoder to integrate global visual and semantic features. To provide a prediction with a precise chromosome number and category distribution, a Category Distribution Reasoning Module (CDRM) is built for foreground-background objects and chromosome class distribution reasoning. We evaluate ChromTR on 1404 newly collected R-band metaphase images and the public G-band dataset AutoKary2022. Our proposed ChromTR outperforms all previous chromosome detection methods with an average precision of 92.56% in R-band chromosome detection, surpassing the baseline method by 3.02%. In a clinical test, ChromTR is also confident in tackling normal and numerically abnormal karyotypes. When extended to the chromosome enumeration task, ChromTR also demonstrates state-of-the-art performances on R-band and G-band two metaphase image datasets. Given these superior performances to other methods, our proposed method has been applied to assist clinical karyotype diagnosis.

Identification of genetic associations between acute myocardial infarction and non-small cell lung cancer

IntroductionA growing body of evidence suggests a potential connection between myocardial infarction (MI) and lung cancer (LC). However, the underlying pathogenesis and molecular mechanisms remain unclear. This research aims to identify common genes and pathways between MI and LC through bioinformatics analysis.MethodsTwo public datasets (GSE166780 and GSE8569) were analyzed to identify differentially expressed genes (DEGs). Common DEGs were enriched using Gene Ontology (GO) and the Kyoto Encyclopedia of Genes and Genomes (KEGG). Hub genes were identified and their diagnostic performance was evaluated. Gene co-expression networks, as well as regulatory networks involving miRNA-hub genes and TF-hub genes, were also constructed. Finally, candidate drugs were predicted.ResultsAmong the datasets, 34 common trend DEGs were identified. Enrichment analysis linked these DEGs to key biological processes, cellular components, and molecular functions. Eight hub genes (CEBPA, TGFBR2, EZH2, JUNB, JUN, FOS, PLAU, COL1A1) were identified, demonstrating promising diagnostic accuracy. Key transcription factors associated with these hub genes include SP1, ESR1, CREB1, ETS1, NFKB1, and RELA, while key miRNAs include hsa-mir-101-3p, hsa-mir-124-3p, hsa-mir-29c-3p, hsa-mir-93-5p, and hsa-mir-155-5p. Additionally, potential therapeutic drugs were identified, with zoledronic acid anhydrous showing potential value in reducing the co-occurrence of the two diseases.DiscussionThis study identified eight common signature genes shared between NSCLC and AMI. Validation datasets confirmed the diagnostic value of key hub genes COL1A1 and PLAU. These findings suggest that shared hub genes may serve as novel therapeutic targets for patients with both diseases. Ten candidate drugs were predicted, with zoledronic acid showing potential for targeting dual hub genes, offering a promising therapeutic approach for the comorbidity of lung cancer and myocardial infarction.

A YOLO Network Based on Depthwise Convolution Attention, Feature Fusion, and KL Divergence (DFK-YOLO): A Deep Learning Method for Infrared Small Target Detection Based on YOLOv7

Infrared imaging technology has a wide range of applications across various fields, with one of its most critical uses being the detection of small infrared targets. However, model-driven approaches often lack robustness in identifying these small targets, while current deep learning-based methods face challenges in effectively extracting and integrating features. Additionally, appropriate labeling strategies for small infrared targets remain underdeveloped. To address these limitations, this paper proposes a novel detection method based on YOLOv7. Specifically, an attention module leveraging Depthwise Convolution is incorporated into the backbone of YOLOv7. Furthermore, a new Feature Fusion Neck is designed to replace the original neck component of YOLOv7. Lastly, a novel label assignment strategy is introduced. The proposed method achieves a mAP@0.5 of 99.5% and a mAP@0.75 of 71.6% on a public dataset, surpassing the baseline YOLOv7 by 1% and 4.6%, respectively. Compared to state-of-the-art deep learning object detection methods, the proposed approach demonstrates superior performance.

End to End Leaf Detection Using Convolutional Neural Network

Global food security is seriously threatened by crop diseases, yet in many places, it is still difficult to identify them quickly because of a lack of infrastructure. In tackling this problem, recent developments in leaf-based image categorization methods have produced encouraging outcomes. This study investigates how to differentiate between healthy and unhealthy leaves using curated datasets using the Random Forest algorithm. The development of the dataset, feature extraction, classifier training, and classification are the various stages of the implementation process. To efficiently distinguish between diseased and healthy samples, the Random Forest classifier is used to train the dataset, which includes pictures of both diseased and healthy leaves. A significant obstacle to agricultural production, plant diseases have an impact on crop yield and quality. Precision farming and sustainable agriculture depend on the timely and precise identification of leaf diseases. In this work, we present a complete system that automatically detects and classifies leaf diseases using Convolutional Neural Networks (CNN), starting with picture capture. The technology offers a quicker method of identifying leaf illnesses and does away with the requirement for manual feature engineering. Our tests, carried out on several public datasets, show the model's efficacy and achieve high accuracy for a range of illness categories. These results underline the practical uses that might improve crop management and early intervention techniques. Using machine learning on massive datasets that are accessible to the public, this approach offers a scalable solution for detecting plant diseases on a large scale.

Multi-source domain generalization tool wear prediction based on wide convolution weighted antagonism

Abstract While traditional deep learning models achieve high accuracy in predicting tool wear under consistent working conditions, actual production processes frequently involve varying conditions due to different processing methods. The wear data of different working conditions have a large difference in distribution, so that the wear signal of milling cutter trained in one working condition can only predict the wear value of the working condition, which will cause a large waste of material and manpower for actual production. Therefore, a domain generalization method utilizing wide deep convolutional neural networks and weighted antagonistic multi-source domain generalization is introduced for tool wear prediction, and multi-source domain wide depth convolutional weighted adversarial domain generalization model (WDCNNWAL) model is constructed. After filtering and de-noising the original data, it is input into WDCNN for automatic feature learning, and the resulting multi-source feature signals are input into the prediction model and discriminator respectively to obtain corresponding loss values. Furthermore, Wasserstein distance is used to measure the probability distribution distance of multi-source feature signals as a weighted value combined with the discriminator loss function, and the final loss value is obtained by adding together. Finally, the tool wear prediction model is set up using a backpropagation neural network. Various milling wear experimental data along with the NASA tool wear public dataset were utilized to assess the predictive performance of the trained model. The model achieves a high level of generalization in tool wear prediction, with average root mean squared error (RMSE) of 0.0599 and 0.2075, as well as average R 2 of 0.9037 and 0.9196 on the NASA and self-collecting datasets respectively. To validate its exceptional precision in generalization performance, comparative experiments were conducted with other data-driven methods resulting in over a 50% reduction in average RMSE and more than a 40% increase in average R 2.

Photonic diffractive generators through sampling noises from scattering media

Photonic computing, with potentials of high parallelism, low latency and high energy efficiency, have gained progressive interest at the forefront of neural network (NN) accelerators. However, most existing photonic computing accelerators concentrate on discriminative NNs. Large-scale generative photonic computing machines remain largely unexplored, partly due to poor data accessibility, accuracy and hardware feasibility. Here, we harness random light scattering in disordered media as a native noise source and leverage large-scale diffractive optical computing to generate images from above noise, thereby achieving hardware consistency by solely pursuing the spatial parallelism of light. To realize experimental data accessibility, we design two encoding strategies between images and optical noise latent space that effectively solves the training problem. Furthermore, we utilize advanced photonic NN architectures including cascaded and parallel configurations of diffraction layers to enhance the image generation performance. Our results show that the photonic generator is capable of producing clear and meaningful synthesized images across several standard public datasets. As a photonic generative machine, this work makes an important contribution to photonic computing and paves the way for more sophisticated applications such as real world data augmentation and multi modal generation.

LR-SLAM: Visual Inertial SLAM System with Redundant Line Feature Elimination

The present study focuses on the simultaneous localization and mapping (SLAM) system based on point and line features. Aiming to address the prevalent issue of repeated detection during line feature extraction in low-texture environments, a novel method for merging redundant line features is proposed. This method effectively mitigates the problem of increased initial pose estimation error that arises when the same line is erroneously detected as multiple lines in adjacent frames. Furthermore, recognizing the potential for the introduction of line features to prolong the marginalization process of the information matrix, optimization strategies are employed to accelerate this process. Additionally, to tackle the issue of insufficient point feature accuracy, subpixel technology is introduced to enhance the precision of point features, thereby further reducing errors. Experimental results on the European Robotics Challenge (EUROC) public dataset demonstrate that the proposed LR-SLAM system exhibits significant advantages over mainstream SLAM systems such as ORB-SLAM3, VINS-Mono, and PL-VIO in terms of accuracy, efficiency, and robustness.

Improve myocardial strain estimation based on deformable groupwise registration with a locally low-rank dissimilarity metric

BackgroundCurrent mainstream cardiovascular magnetic resonance-feature tracking (CMR-FT) methods, including optical flow and pairwise registration, often suffer from the drift effect caused by accumulative tracking errors. Here, we developed a CMR-FT method based on deformable groupwise registration with a locally low-rank (LLR) dissimilarity metric to improve myocardial tracking and strain estimation accuracy.MethodsThe proposed method, Groupwise-LLR, performs feature tracking by iteratively updating the entire displacement field across all cardiac phases to minimize the sum of the patchwise signal ranks of the deformed movie. The method was compared with alternative CMR-FT methods including the Farneback optical flow, a sequentially pairwise registration method, and a global low rankness-based groupwise registration method via a simulated dataset (n = 20), a public cine data set (n = 100), and an in-house tagging-MRI patient dataset (n = 16). The proposed method was also compared with two general groupwise registration methods, nD + t B-Splines and pTVreg, in simulations and in vivo tracking.ResultsOn the simulated dataset, Groupwise-LLR achieved the lowest point tracking errors (p = 0.13 against pTVreg for the temporally averaged point tracking errors in the long-axis view, and p < 0.05 for all other cases), voxelwise strain errors (all p < 0.05), and global strain errors (p = 0.05 against pTVreg for the longitudinal global strain errors, and p < 0.05 for all other cases). On the public dataset, Groupwise-LLR achieved the lowest contour tracking errors (all p < 0.05), reduced the drift effect in late-diastole, and preserved similar inter-observer reproducibility as the alternative methods. On the patient dataset, Groupwise-LLR correlated better with tagging-MRI for radial strains than the other CMR-FT methods in multiple myocardial segments and levels.ConclusionsThe proposed Groupwise-LLR reduces the drift effect and provides more accurate myocardial tracking and strain estimation than the alternative methods. The method may thus facilitate a more accurate estimation of myocardial strains for clinical assessments of cardiac function.

PURE: a Prompt-based framework with dynamic Update mechanism for educational Relation Extraction

Traditional education systems obscure the diverse interconnections inherent within subject knowledge, thus failing to meet the current demand for personalized and adaptive learning experiences. Recent advances have explored various relation extraction techniques to construct educational knowledge graphs that integrate dispersed subject knowledge into a unified framework. However, educational conceptual entities are far more abstract and intricate compared to their real-world equivalents, and these techniques primarily focus on static knowledge, overlooking the dynamic nature of knowledge in practical learning and application scenarios. To address these issues, we propose a Prompt-based framework with dynamic Update mechanism for educational Relation Extraction (PURE). This framework embraces a prompt-tuning strategy and employs a more appropriate MacBERT-large model to encode the instances wrapped by prompt templates. Furthermore, we construct an instance-relation database that serves as an external knowledge base of our framework. A dynamic instance-relation update mechanism is proposed to refine the database, thus enhancing the accuracy of PURE in predicting new triples. We conduct experiments on a Data Structure course relation extraction dataset and three public datasets. The experimental results demonstrate that PURE achieves significant improvements and outperforms several state-of-the-art baselines in efficiency of extraction and utilization of educational information. Comparable performance is achieved even in more complex biomedical relation extraction, validating its robustness and applicability to other domains.

A motor imagery classification model based on hybrid brain-computer interface and multitask learning of electroencephalographic and electromyographic deep features

ObjectiveExtracting deep features from participants’ bioelectric signals and constructing models are key research directions in motor imagery (MI) classification tasks. In this study, we constructed a multimodal multitask hybrid brain-computer interface net (2M-hBCINet) based on deep features of electroencephalogram (EEG) and electromyography (EMG) to effectively accomplish motor imagery classification tasks.MethodsThe model first used a variational autoencoder (VAE) network for unsupervised learning of EEG and EMG signals to extract their deep features, and subsequently applied the channel attention mechanism (CAM) to select these deep features and highlight the advantageous features and minimize the disadvantageous ones. Moreover, in this study, multitask learning (MTL) was applied to train the 2M-hBCINet model, incorporating the primary task that is the MI classification task, and auxiliary tasks including EEG reconstruction task, EMG reconstruction task, and a feature metric learning task, each with distinct loss functions to enhance the performance of each task. Finally, we designed module ablation experiments, multitask learning comparison experiments, multi-frequency band comparison experiments, and muscle fatigue experiments. Using leave-one-out cross-validation(LOOCV), the accuracy and effectiveness of each module of the 2M-hBCINet model were validated using the self-made MI-EEMG dataset and the public datasets WAY-EEG-GAL and ESEMIT.ResultsThe results indicated that compared to comparative models, the 2M-hBCINet model demonstrated good performance and achieved the best results across different frequency bands and under muscle fatigue conditions.ConclusionThe 2M-hBCINet model constructed based on EMG and EEG data innovatively in this study demonstrated excellent performance and strong generalization in the MI classification task. As an excellent end-to-end model, 2M-hBCINet can be generalized to be used in EEG-related fields such as anomaly detection and emotion analysis.

Self-supervised learning improves robustness of deep learning lung tumor segmentation models to CT imaging differences.

Self-supervised learning (SSL) is an approach to extract useful feature representations from unlabeled data, and enable fine-tuning on downstream tasks with limited labeled examples. Self-pretraining is a SSL approach that uses curated downstream task dataset for both pretraining and fine-tuning. Availability of large, diverse, and uncurated public medical image sets presents the opportunity to potentially create foundation models by applying SSL in the "wild" that are robust to imaging variations. However, the benefit of wild- versus self-pretraining has not been studied for medical imageanalysis. Compare robustness of wild versus self-pretrained models created using convolutional neural network (CNN) and transformer (vision transformer [ViT] and hierarchical shifted window [Swin]) models for non-small cell lung cancer (NSCLC) segmentation from 3D computed tomography (CT) scans. CNN, ViT, and Swin models were wild-pretrained using unlabeled 10,412 3D CTs sourced from the cancer imaging archive and internal datasets. Self-pretraining was applied to same networks using a curated public downstream task dataset (n = 377) of patients with NSCLC. Pretext tasks introduced in self-distilled masked image transformer were used for both pretraining approaches. All models were fine-tuned to segment NSCLC (n = 377 training dataset) and tested on two separate datasets containing early (public n = 156) and advanced stage (internal n = 196) NSCLC. Models were evaluated in terms of: (a) accuracy, (b) robustness to image differences from contrast, slice thickness, and reconstruction kernels, and (c) impact of pretext tasks for pretraining. Feature reuse was evaluated using centered kernel alignment. Wild-pretrained Swin models resulted in higher feature reuse at earlier level layers and increased feature differentiation close to output. Wild-pretrained Swin outperformed self-pretrained models for analyzed imaging acquisitions. Neither ViT nor CNN showed a clear benefit of wild-pretraining compared to self-pretraining. Masked image prediction pretext task that forces networks to learn the local structure resulted in higher accuracy compared to contrastive task that models global image information. Wild-pretrained Swin networks were more robust to analyzed CT imaging differences for lung tumor segmentation than self-pretrained methods. ViT and CNN models did not show a clear benefit for wild-pretraining overself-pretraining.

A novel convolutional neural network for enhancing the continuity of pavement crack detection

Pavement cracks affect the structural stability and safety of roads, making accurate identification of crack for assessing the extent of damage and evaluating road health. However, traditional convolutional neural networks often struggle with issues such as missed detection and false detection when extracting cracks. This paper introduces a network called CPCDNet, designed to maintain continuous extraction of pavement cracks. The model incorporates a Crack align module (CAM) and a Weighted Edge Cross Entropy Loss Function (WECEL) to enhance the continuity of crack extraction in complex environments. Experimental results show that the proposed model achieves mIoU scores of 77.71%, 80.36%, 91.19%, and 71.16% on the public datasets CFD, Crack500, Deepcrack537, and Gaps384, respectively. Compared to other networks, the proposed method improves the continuity and accuracy of crack extraction.