Deep convolutional neural networks (DCNNs) have demonstrated excellent performance in object recognition and have been found to share some similarities with brain visual processing. However, the substantial gap between DCNNs and human visual perception still exists. Functional magnetic resonance imaging (fMRI) as a widely used technique in cognitive neuroscience can record neural activation in the human visual cortex during the process of visual perception. Can we teach DCNNs human fMRI signals to achieve a more brain-like model? To answer this question, this study proposed ReAlnet-fMRI, a model based on the SOTA vision model CORnet but optimized using human fMRI data through a multi-layer encoding-based alignment framework. This framework has been shown to effectively enable the model to learn human brain representations. The fMRI-optimized ReAlnet-fMRI exhibited higher similarity to the human brain than both CORnet and the control model in within- and across-subject as well as within- and across-modality model-brain (fMRI and EEG) alignment evaluations. Additionally, we conducted an in-depth analysis to investigate how the internal representations of ReAlnet-fMRI differ from CORnet in encoding various object dimensions. These findings provide the possibility of enhancing the brain-likeness of visual models by integrating human neural data, helping to bridge the gap between computer vision and visual neuroscience.

The online version contains supplementary material available at 10.1007/s11571-025-10252-y.

Gastric cancer (GC) occupies the first few places in the world among tumors in terms of incidence and mortality, causing serious harm to human health, and at the same time, its treatment greatly consumes the health care resources of all countries in the world. The diagnosis of GC is usually based on histopathologic examination, and it is very important to be able to detect and identify cancerous lesions at an early stage, but some endoscopists' lack of diagnostic experience and fatigue at work lead to a certain rate of under diagnosis. The rapid and striking development of Artificial intelligence (AI) has helped to enhance the ability to extract abnormal information from endoscopic images to some extent, and more and more researchers are applying AI technology to the diagnosis of GC. This initiative has not only improved the detection rate of early gastric cancer (EGC), but also significantly improved the survival rate of patients after treatment. This article reviews the results of various AI-assisted diagnoses of EGC in recent years, including the identification of EGC, the determination of differentiation type and invasion depth, and the identification of borders. Although AI has a better application prospect in the early diagnosis of ECG, there are still major challenges, and the prospects and limitations of AI application need to be further discussed.

Sleep apnea/hypopnea is a sleep disorder characterized by repeated pauses in breathing which could induce a series of health problems such as cardiovascular disease (CVD) and even sudden death. Polysomnography (PSG) is the most common way to diagnose sleep apnea/hypopnea. Considering that PSG data acquisition is complex and the diagnosis of sleep apnea/hypopnea requires manual scoring, it is very time-consuming and highly professional. With the development of wearable devices and AI techniques, more and more works have been focused on building machine and deep learning models that use single or multi-modal physiological signals to achieve automated detection of sleep apnea/hypopnea. This paper provides a comprehensive review of automatic sleep apnea/hypopnea detection methods based on AI-based techniques in recent years. We summarize the general process used by existing works with a flow chart, which mainly includes data acquisition, raw signal pre-processing, model construction, event classification, and evaluation, since few papers consider these. Additionally, the commonly used public database and pre-processing methods are also reviewed in this paper. After that, we separately summarize the existing methods related to different modal physiological signals including nasal airflow, pulse oxygen saturation (SpO2), electrocardiogram (ECG), electroencephalogram (EEG) and snoring sound. Furthermore, specific signal pre-processing methods based on the characteristics of different physiological signals are also covered. Finally, challenges need to be addressed, such as limited data availability, imbalanced data problem, multi-center study necessity etc., and future research directions related to AI are discussed.

Circular RNAs (circRNAs) are a unique class of covalently closed single-stranded RNA molecules that play diverse roles in normal physiology and pathology. Among the major types of circRNA, exon-intron circRNA (EIciRNA) distinguishes itself by its sequence composition and nuclear localization. Recent RNA-seq technologies and computational methods have facilitated the detection and characterization of EIciRNAs, with features like circRNA intron retention (CIR) and tissue-specificity being characterized. EIciRNAs have been identified to exert their functions via mechanisms such as regulating gene transcription, and the physiological relevance of EIciRNAs has been reported. Within this review, we present a summary of the current understanding of EIciRNAs, delving into their identification and molecular functions. Additionally, we emphasize factors regulating EIciRNA biogenesis and the physiological roles of EIciRNAs based on recent research. We also discuss the future challenges in EIciRNA exploration, underscoring the potential for novel functions and functional mechanisms of EIciRNAs for further investigation.

Pitch plays an essential role in music perception and forms the fundamental component of melodic interpretation. However, objectively detecting and decoding brain responses to musical pitch perception across subjects remains to be explored. In this study, we employed electroencephalography (EEG) as an objective measure to obtain the neural responses of musical pitch perception. The EEG signals from 34 subjects under hearing violin sounds at pitches G3 and B6 were collected with an efficient passive Go/No-Go paradigm. The lightweight modified EEGNet model was proposed for EEG-based pitch classification. Specifically, within-subject modeling with the modified EEGNet model was performed to construct individually optimized models. Subsequently, based on the within-subject model pool, a classifier ensemble (CE) method was adopted to construct the cross-subject model. Additionally, we analyzed the optimal time window of brain decoding for pitch perception in the EEG data and discussed the interpretability of these models. The experiment results show that the modified EEGNet model achieved an average classification accuracy of 77% for within-subject modeling, significantly outperforming other compared methods. Meanwhile, the proposed CE method achieved an average accuracy of 74% for cross-subject modeling, significantly exceeding the chance-level accuracy of 50%. Furthermore, we found that the optimal EEG data window for the pitch perception lies 0.4 to 0.9 s onset. These promising results demonstrate that the proposed methods can be effectively used in the objective assessment of pitch perception and have generalization ability in cross-subject modeling.

Health risks associated with phthalate esters depend on exposure level, individual sensitivities, and other contributing factors.

This study employed artificial intelligence algorithms while applying data mining techniques to identify correlations between phthalate esters [di(2-ethylhexyl) phthalate, DEHP], lifestyle factors, and disease outcomes.

We conducted exploratory analysis using demographic and laboratory data collected from the Taiwan Biobank. The study developed a prediction model to examine the relationship between phthalate esters and the risk of developing certain diseases based on various artificial intelligence algorithms, including logistic regression, artificial neural networks, and Bayesian networks.

The results indicate that phthalate esters exhibited a greater impact on bone and joint issues than heart problems. We observed that DEHP metabolites, such as mono(2-carboxymethylhexyl) phthalate, mono-n-butyl phthalate, and monoethylphthalate, leave higher residue in females than in males, with statistically significant differences. Monoethylphthalate levels were lower in individuals who exercised regularly than those who did not, indicating statistically significant differences.

This study's findings can serve as a valuable reference for clinical nursing assessments regarding diseases related to osteoporosis, arthritis, and musculoskeletal pain. Medical professionals can enhance care quality by considering factors beyond patients' essential physical assessment items.Trial Registration: This study was registered under NCT05892029 on May 5, 2023, retrospectively.

This paper aims to develop a three-dimensional (3D) Alzheimer's disease (AD) prediction method, thereby bettering current predictive methods, which struggle to fully harness the potential of structural magnetic resonance imaging (sMRI) data.

Traditional convolutional neural networks encounter pressing difficulties in accurately focusing on the AD lesion structure. To address this issue, a 3D decoupling, self-attention network for AD prediction is proposed. Firstly, a multi-scale decoupling block is designed to enhance the network's ability to extract fine-grained features by segregating convolutional channels. Subsequently, a self-attention block is constructed to extract and adaptively fuse features from three directions (sagittal, coronal and axial), so that more attention is geared towards brain lesion areas. Finally, a clustering loss function is introduced and combined with the cross-entropy loss to form a joint loss function for enhancing the network's ability to discriminate between different sample types.

The accuracy of our model is 0.985 for the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset and 0.963 for the Australian Imaging, Biomarker & Lifestyle (AIBL) dataset, both of which are higher than the classification accuracy of similar tasks in this category. This demonstrates that our model can accurately distinguish between normal control (NC) and Alzheimer's Disease (AD), as well as between stable mild cognitive impairment (sMCI) and progressive mild cognitive impairment (pMCI).

The proposed AD prediction network exhibits competitive performance when compared with state-of-the-art methods. The proposed model successfully addresses the challenges of dealing with 3D sMRI image data and the limitations stemming from inadequate information in 2D sections, advancing the utility of predictive methods for AD diagnosis and treatment.

Transcriptomic data analysis entails the measurement of RNA transcript (gene expression products) abundance in a cell or a cell population at a single point in time. In other words, transcriptomics as it is currently practiced is two-dimensional (2DTA). Gene expression profiling by 2DTA has proven invaluable in furthering our understanding of numerous biological processes in health and disease. That said, shortcomings including technical variability, small sample size, differential rates of transcript decay, and the lack of linearity between transcript abundance and functionality or the formation of functional proteins limit the interpretive utility and generalizability of transcriptomic data. 2DTA utility may also be constrained by its reliance on RNA extracts obtained at a single time point. In other words, much like judging a movie by a single frame, 2DTA can only provide a snapshot of the transcriptome at time of RNA extraction. Whether this perceived "temporality" problem is real and whether it has any bearing on transcriptomic data interpretation have yet to be addressed. To investigate this problem, 25 publicly available datasets relating to MCF-7 cells, where RNA extracts obtained at 12- or 48-hours post-culture were subjected to transcriptomic analysis. The individual datasets were downloaded and compiled into two separate datasets (MCF-7 U12hr and MCF-7 U48hr). To comparatively analyze the two compiled datasets, three machine learning approaches (decision trees (DT), random forests (RF), and XGBoost (Extreme Gradient Boosting)) were used as classifiers to search for genes with distinct expression patterns between the two groups. Shapley additive explanation (SHAP), an explainable AI method, was used to assess the fundamental principles of the DT, RF, and XGBoost models. Coefficient of Determination (DC), Mean Absolute Error (MAE), and Mean Squared Error (MSE) were used to evaluate the models. The results show that the two datasets exhibited very significant gene expression patterns. The XGBoost model performed better than the DT or RF models with MSE, MAE, and DC values of 0.00028, 0.00028, and 0.95778 respectively. These observations suggest that time, as a third dimension, can impact transcriptomic data interpretation and that machine learning and explainable AI are useful tools in resolving the temporality problem in transcriptomics.

Goji berry is an important economic crop, yet pest infestations pose a significant threat to its yield and quality. Traditional pest identification mainly relies on manual inspection by experts with specialized knowledge, which is subjective, time-consuming, and labor-intensive. To address these issues, this experiment proposes an improved convolutional neural network (CNN) for accurate identification of 17 types of goji berry pests. Firstly, the original data set is augmented using a multi-graph-occlusion data augmentation method. Subsequently, the augmented data set is imported into the improved CNN for training. Based on the original ResNet18 model, a new CNN, named GojiNet, is constructed by embedding multi-attention fusion modules at appropriate locations. Experimental results demonstrate that GojiNet achieves an average recognition accuracy of 95.35%, representing a 2.60% improvement over the ResNet18 network. Notably, compared to the original network, the training time of this model increases only slightly, while its size is reduced, and the recognition accuracy is enhanced. The experiment verifies the performance of the GojiNet model through a series of evaluation indicators. This study confirms the tremendous potential and application prospects of deep learning in pest identification, providing a referential solution for intelligent and precise pest identification.

Toxicity prediction is crucial in drug discovery, helping identify safe compounds and reduce development risks. However, the lack of known toxicity data for most compounds is a major challenge. Recently, data-driven models have gained attention as a more efficient alternative to traditional in vivo and in vitro experiments. In this paper, we propose a small-scale, data-driven toxicity prediction method based on Graph Neural Network (GNN). We introduce a joint learning strategy for multiple toxicity types and construct a graph-based model, JLGCN-MTT, to improve prediction accuracy. In addition, we integrate a transfer learning strategy that leverages data from multiple toxicity types, allowing the model to make reliable predictions even when data for a specific toxicity type is limited. We conducted experiments using data from 3566 compounds in the Tox21 dataset, which contains 12 types of toxicity-related bioactivity data. The experimental results show that JLGCN-MTT outperforms traditional machine learning methods and single-task GNN in all 12 toxicity prediction tasks, with AUC improving by over 10% in 11 tasks. For small-scale data with 50, 100, and 300 training samples, the AUC improved in all cases, with the highest improvement of 11% observed when the sample size was 50. These results demonstrate that the small-scale, data-driven toxicity prediction method we propose can achieve high prediction accuracy.

Drug-target affinity prediction is a fundamental task in the field of drug discovery. Extracting and integrating structural information from proteins effectively is crucial to enhance the accuracy and generalization of prediction, which remains a substantial challenge. This paper proposes a pocket-based multimodal deep learning model named PocketDTA for drug-target affinity prediction, based on the principle of "structure determines function". PocketDTA introduces the pocket graph structure that encodes protein residue features pretrained using a biological language model as nodes, while edges represent different protein sequences and spatial distances. This approach overcomes the limitations of lack of spatial information in traditional prediction models with only protein sequence input. Furthermore, PocketDTA employs relational graph convolutional networks at both atomic and residue levels to extract structural features from drugs and proteins. By integrating multimodal information through deep neural networks, PocketDTA combines sequence and structural data to improve affinity prediction accuracy. Experimental results demonstrate that PocketDTA outperforms state-of-the-art prediction models across multiple benchmark datasets by showing strong generalization under more realistic data splits and confirming the effectiveness of pocket-based methods for affinity prediction.

Recently, artificial intelligence (AI) has emerged as a transformative tool, enhancing the speed, accuracy, and scalability of bacterial diagnostics. This review explores the role of AI in revolutionizing bacterial detection and antimicrobial susceptibility testing (AST) by leveraging machine learning models, including Random Forest, Support Vector Machines (SVM), and deep learning architectures such as Convolutional Neural Networks (CNNs) and transformers. The integration of AI into these methods promises to address the current limitations of traditional techniques, offering a path toward more efficient, accessible, and reliable diagnostic solutions. In particular, AI-based approaches have demonstrated significant potential in resource-limited settings by enabling cost-effective and portable diagnostic solutions, reducing dependency on specialized infrastructure, and facilitating remote bacterial detection through smartphone-integrated platforms and telemedicine applications. This review highlights AI's transformative role in automating data analysis, minimizing human error, and delivering real-time diagnostic results, ultimately improving patient outcomes and optimizing healthcare efficiency. In addition, we not only examine the current advances in machine learning and deep learning but also review their applications in plate counting, mass spectrometry, morphology-based and motion-based microscopic detection, holographic microscopy, colorimetric and fluorescence detection, electrochemical sensors, Raman and Surface-Enhanced Raman Spectroscopy (SERS), and Atomic Force Microscopy (AFM) for bacterial diagnostics and AST. Finally, we discuss the future directions and potential advancements in AI-driven bacterial diagnostics.

Trityl OXO71-based pulse electron paramagnetic resonance imaging (EPRI) is an excellent technique to obtain partial pressure of oxygen (pO2) maps in tissues. In this study, we used deep learning techniques to denoise 3D EPR amplitude and pO2 maps.

All experiments were performed using a 25 mT EPR imager, JIVA-25®. The MONAI implementation of four neural networks (autoencoder, Attention UNet, UNETR, and UNet) was tested, and the best model (UNet) was then enhanced with joint bilateral filters (JBF). The training dataset was comprised of 227 3D images (56 in vivo and 171 in vitro), 159 images for training, 45 for validation, and 23 for testing. UNet with 1, 2, and 3 JBF layers was tested to improve image SNR, focusing on multiscale structural similarity index measure and edge sensitivity preservation. The trained algorithm was tested using acquisitions with 15, 30, and 150 averages in vitro with a sealed deoxygenated OXO71 phantom and in vivo with fibrosarcoma tumors grown in a hind leg of C3H mice.

We demonstrate that UNet with 2 JBF layers (UNet+JBF2) provides the best outcome. We demonstrate that using the UNet+JBF2 model, the SNR of 15-shot amplitude maps provides higher SNR compared to 150-shot pre-filter maps, both in phantoms and in tumors, therefore, allowing 10-fold accelerated imaging. We demonstrate that the trained algorithm improves SNR in pO2 maps.

We demonstrate the application of deep learning techniques to EPRI denoising. Higher SNR will bring the EPRI technique one step closer to clinics.

A paradigm of chemical risk assessment is continuously extending from focusing on 'single substances' to more comprehensive approaches that examines the combined toxicity among different components in 'mixtures.' This change aims to account for the cocktail effect arising from the toxicological interactions in mixtures, which can lead to increased risks. More than 1000 potential endocrine-disrupting chemicals (EDCs) have been reported, and they can be included in different industrial and consumer chemical products and released to the environment as pollutants of emerging environmental concern. Although extensive studies involving both experiments and predictions have investigated individual EDCs, predictions of their synergistic effects are still relatively lacking, an area that requires further investigation. In this study, we extensively investigated substances in consumer products, mainly marketed in South Korea, that might exhibit estrogenic activity or reproductive toxicity. A high throughput screening (HTS) assay based on OECD Test Guideline 455 for hERαHeLa-9903 cells was constructed, and 435 substances were screened using the HTS. Thirty-five (potential) estrogenic agonists were selected, and their 1412 binary mixtures that could be prepared in four different ratios were systematically tested, considering the available effective concentrations of substances and the solubility of their resulting mixtures. The best empirical dose-response curves of 35 substances and 917 mixtures were derived in this study. Based on the HTS data, a deep neural network model was developed (area under the curve (AUC): 0.837-0.881) and compared with a random forest model (AUC: 0.656-0.829) to screen for the synergistic estrogenic activity of binary mixtures.

Timely and accurate outcome prediction is essential for clinical decision-making for ischemic stroke patients in the intensive care unit (ICU). However, the interpretation and translation of predictive models into clinical applications are equally crucial. This study aims to develop an interpretable machine learning (IML) model that effectively predicts in-hospital mortality for ischemic stroke patients.

In this study, an IML model was developed and validated using multicenter cohorts of 3225 ischemic stroke patients admitted to the ICU. Nine machine learning (ML) models, including logistic regression (LR), K-nearest neighbors (KNN), naive Bayes (NB), decision tree (DT), support vector machine (SVM), random forest (RF), XGBoost, LightGBM, and artificial neural network (ANN), were developed to predict in-hospital mortality using data from the MIMIC-IV and externally validated in Shanghai Changhai Hospital. Feature selection was conducted using three algorithms. Model's performance was assessed using area under the receiver operating characteristic (AUROC), accuracy, sensitivity, specificity and F1 score. Calibration curve and Brier score were used to evaluate the degree of calibration of the model, and decision curve analysis were generated to assess the net clinical benefit. Additionally, the SHapley Additive exPlanations (SHAP) method was employed to evaluate the risk of in-hospital mortality among ischemic stroke patients admitted to the ICU.

Mechanical ventilation, age, statins, white blood cell, blood urea nitrogen, hematocrit, warfarin, bicarbonate and systolic blood pressure were selected as the nine most influential variables. The RF model demonstrated the most robust predictive performance, achieving AUROC values of 0.908 and 0.858 in the testing set and external validation set, respectively. Calibration curves also revealed a high consistency between observations and predictions. Decision curve analysis showed that the model had the greatest net benefit rate when the prediction probability threshold is 0.10 ∼ 0.80. SHAP was employed to interpret the RF model. In addition, we have developed an online prediction calculator for ischemic stroke patients.

This study develops a machine learning-based calculator to predict the probability of in-hospital mortality among patients with ischemic stroke in ICU. The calculator has the potential to guide clinical decision-making and improve the care of patients with ischemic stroke by identifying patients at a higher risk of in-hospital mortality.

This study presents a novel deep learning approach for surface electromyography (sEMG) gesture recognition using stacked autoencoder neural network (SAE)s. The method leverages hierarchical representation learning to extract meaningful features from raw sEMG signals, enhancing the precision and robustness of gesture classification.•Feature Extraction and Classification MODWT Decomposition: The sEMG signals were decomposed using the MODWT DECOMPOSITION(Maximal Overlap Discrete Wavelet Transform) to capture various frequency components.•Time Domain Parameters: A total of 28 features per subject were extracted from the time domain, including statistical and spectral features.•Classifier Evaluation: Initial evaluations involved Autoencoder and LDA (Linear Discriminant Analysis) classifiers, with Autoencoder achieving an average accuracy of 77.96 % ± 1.24, outperforming LDA's 65.36 % ± 1.09.Advanced Neural Network Approach: Stacked Autoencoder Neural Network: To address challenges in distinguishing similar gestures within grasp groups, a Stacked Autoencoder Neural Network was employed. This advanced neural network architecture improved classification accuracy to over 100 %, demonstrating its effectiveness in handling complex gesture recognition tasks. These findings emphasize the significant potential of deep learning models in enhancing prosthetic control and rehabilitation technologies. . To verify these findings, we developed a 3d hand module in ADAMS software that is simulated using Matlab-ADAMS cosimulation.

Immunotherapy has become an option for the first-line therapy of advanced gastric cancer (GC), with improved survival. Our study aimed to investigate unresectable GC from an imaging perspective combined with clinicopathological variables to identify patients who were most likely to benefit from immunotherapy.

Patients with unresectable GC who were consecutively treated with immunotherapy at two different medical centers of Chinese PLA General Hospital were included and divided into the training and validation cohorts, respectively. A deep learning neural network, using a multimodal ensemble approach based on CT imaging data before immunotherapy, was trained in the training cohort to predict survival, and an internal validation cohort was constructed to select the optimal ensemble model. Data from another cohort were used for external validation. The area under the receiver operating characteristic curve was analyzed to evaluate performance in predicting survival. Detailed clinicopathological data and peripheral blood prior to immunotherapy were collected for each patient. Univariate and multivariable logistic regression analysis of imaging models and clinicopathological variables was also applied to identify the independent predictors of survival. A nomogram based on multivariable logistic regression was constructed.

A total of 79 GC patients in the training cohort and 97 patients in the external validation cohort were enrolled in this study. A multi-model ensemble approach was applied to train a model to predict the 1-year survival of GC patients. Compared to individual models, the ensemble model showed improvement in performance metrics in both the internal and external validation cohorts. There was a significant difference in overall survival (OS) among patients with different imaging models based on the optimum cutoff score of 0.5 (HR = 0.20, 95 % CI: 0.10-0.37, P < 0.001). Multivariate Cox regression analysis revealed that the imaging models, PD-L1 expression, and lung immune prognostic index were independent prognostic factors for OS. We combined these variables and built a nomogram. The calibration curves showed that the C-index of the nomogram was 0.85 and 0.78 in the training and validation cohorts.

The deep learning model in combination with several clinical factors showed predictive value for survival in patients with unresectable GC receiving immunotherapy.

Herein, we report a sodium dodecyl-sulfate polyacrylamide gel electrophoresis (SDS-PAGE) method featuring dual imaging and signal-fusion deep learning for specific identification and analysis of glycomacropeptide (GMP) in milk sample. Conventional SDS-PAGE methods lack specificity because of the signle staining of protein bands, and the overlap between GMP and β-lactoglobulin (βLg). Our dual imaging method generated a pair of complementary detection signals by recruiting intrinsic fluorescence imaging (IFI) and silver staining. Comparing the IFI image with the staining image highlighted the presence of GMP and differentiated it from βLg. Additionally, we trained a signal-fusion deep learning model to improve the quantitative performance of our method. The model fused the features extracted from the paired detection signals (IFI and staining) and accurately classified them into different mixing ratios (proportion of GMP-containing whey in the sample), indicating the potential for quantitative analysis on the mixing ratios of GMP added into whey sample. The developed method has the merits of specificity, sensitivity and simplilcity, and has potential to analysis of protein/peptides with unique IFI properties in food safety, basic research and biopharming etc.

In this paper, we propose a new super-expressive activation function called the Parametric Elementary Universal Activation Function (PEUAF). We demonstrate the effectiveness of PEUAF through systematic and comprehensive experiments on various industrial and image datasets, including CIFAR-10, Tiny-ImageNet, and ImageNet. The models utilizing PEUAF achieve the best performance across several baseline industrial datasets. Specifically, in image datasets, the models that incorporate mixed activation functions (with PEUAF) exhibit competitive test accuracy despite the low accuracy of models with only PEUAF. Moreover, we significantly generalize the family of super-expressive activation functions, whose existence has been demonstrated in several recent works by showing that any continuous function can be approximated to any desired accuracy by a fixed-size network with a specific super-expressive activation function. Specifically, our work addresses two major bottlenecks in impeding the development of super-expressive activation functions: the limited identification of super-expressive functions, which raises doubts about their broad applicability, and their often peculiar forms, which lead to skepticism regarding their scalability and practicality in real-world applications.

The surge in medical waste has highlighted the urgent need for cost-effective and advanced management solutions. In this paper, a novel medical waste management approach, "MedBin," is proposed for automated sorting, reusing, and recycling. A comprehensive medical waste dataset, "MedBin-Dataset" is established, comprising 2,119 original images spanning 36 categories, with samples captured in various backgrounds. The lightweight "MedBin-Net" model is introduced to enable detection and instance segmentation of medical waste, enhancing waste recognition capabilities. Experimental results demonstrate the effectiveness of the proposed approach, achieving an average precision of 0.91, recall of 0.97, and F1-score of 0.94 across all categories with just 2.51 M parameters (where M stands for million, i.e., 2.51 million parameters), 5.20G FLOPs (where G stands for billion, i.e., 5.20 billion floating-point operations per second), and 0.60 ms inference time. Additionally, the proposed method includes a World Health Organization (WHO) Guideline-Based Classifier that categorizes detected waste into 5 types, each with a corresponding disposal method, following WHO medical waste classification standards. The proposed method, along with the dedicated dataset, offers a promising solution that supports sustainable medical waste management and other related applications. To access the MedBin-Dataset samples, please visit https://universe.roboflow.com/uob-ylti8/medbin_dataset. The source code for MedBin-Net can be found at https://github.com/Wayne3918/MedbinNet.

To improve the storage capacity of the Hopfield model, we develop a version of the dreaming algorithm that perpetually reinforces the patterns to be stored (as in the Hebb rule), and erases the spurious memories (as in dreaming algorithms). For this reason, we called it Daydreaming. Daydreaming is not destructive and it converges asymptotically to stationary retrieval maps. When trained on random uncorrelated examples, the model shows optimal performance in terms of the size of the basins of attraction of stored examples and the quality of reconstruction. We also train the Daydreaming algorithm on correlated data obtained via the random-features model and argue that it spontaneously exploits the correlations thus increasing even further the storage capacity and the size of the basins of attraction. Moreover, the Daydreaming algorithm is also able to stabilize the features hidden in the data. Finally, we test Daydreaming on the MNIST dataset and show that it still works surprisingly well, producing attractors that are close to unseen examples and class prototypes.

Impact, which may occur during manufacturing, serving and maintaining, is a significant threat to in-service composite structures, e.g. wind turbine blades. It calls for developing a method for assessment and localization of impact. In this paper, a single-sensor impact localization method based on deep learning is proposed. Specifically, a multiscale feature fusion convolutional neural network is designed, which, in combination with a convolutional block attention module, adaptively extracts features from single-sensor signals to achieve accurate region-level source localization. Complete ensemble empirical mode decomposition with adaptive noise is employed to reduce noise and extract intrinsic mode functions from acoustic emission signals, enabling more effective feature extraction. The decomposed signals are then converted into grayscale images, forming a dataset for the deep learning model. This approach allows for the extraction of rich feature information. A steel ball drop experiment is conducted to simulate the low-speed impact response of the wind turbine blade spar. The experimental results show significant advantages in localization accuracy. This study offers a promising solution for acoustic emission source region localization in complex composite structures.

This study introduces a hybrid approach that combines unsupervised self-organizing maps (SOM) with a supervised convolutional neural network (CNN) to enhance model accuracy in vector-borne disease modeling. We applied this method to predict insecticide resistance (IR) status in key malaria vectors across Africa. Our results show that the combined SOM/CNN approach is more robust than a standalone CNN model, achieving higher overall accuracy and Kappa scores among others. This confirms the potential of the SOM/CNN hybrid as an effective and reliable tool for improving model accuracy in public health applications.•The hybrid model, combining SOM and CNN, was implemented to predict IR status in malaria vectors, providing enhanced accuracy across various validation metrics.•Results indicate a notable improvement in robustness and predictive accuracy over traditional CNN models.•The combined SOM/CNN approach demonstrated higher Kappa scores and overall model accuracy.

Humans are remarkably efficient at decision making, even in "open-ended" problems where the set of possible actions is too large for exhaustive evaluation. Our success relies, in part, on processes for calling to mind the right candidate actions. When these processes fail, the result is a kind of puzzle in which the value of a solution would be obvious once it is considered, but never gets considered in the first place. Recently, machine learning (ML) architectures have attained or even exceeded human performance on open-ended decision making tasks such as playing chess and Go. We ask whether the broad architectural principles that underlie ML success in these domains generate similar consideration failures to those observed in humans. We demonstrate a case in which they do, illuminating how humans make open-ended decisions, how this relates to ML approaches to similar problems, and how both architectures lead to characteristic patterns of success and failure.

This work lays the groundwork for creating an automated system for the inventory of urban waste elements. Our primary contribution is the development of, to the best of our knowledge, the first re-identification system for urban waste elements that uses Artificial Intelligence and Computer Vision, incorporating information from a classification module and geolocation context to enhance post-processing performance. This re-identification system helps to create and update inventories by determining if a new image matches an existing element in the inventory based on visual similarity or, if not, by adding it as a new identity (new class or new identity of an existing class). Such a system could be highly valuable to local authorities and waste management companies, offering improved facility maintenance, geolocation, and additional applications. This work also addresses the dynamic nature of urban environments and waste management elements by exploring Continual Learning strategies to adapt pretrained systems to new settings with different urban elements. Experimental results show that the proposed system operates effectively across various container types and city layouts. These findings were validated through testing in two different Spanish locations, a "City" and a "Campus", differing in size, illumination conditions, seasons, urban design and container appearance. For the final re-identification system, the baseline system achieves 53.18 mAP (mean Average Precision) in the simple scenario, compared to 21.54 mAP in the complex scenario, with additional challenging unseen variability. Incorporating the proposed post-processing techniques significantly improved results, reaching 74.14 mAP and 71.75 mAP in the simple and complex scenario respectively.

Primary Central Nervous System Lymphoma is an aggressive central nervous system neoplasm with poor response to pharmacological treatment, partially due to insufficient drug delivery across blood-brain barrier. In this study, we developed a novel therapy for this lymphoma by combining a targeted nanopolymer treatment with an immune checkpoint inhibitor antibody (anti-PD-1). A N-(2-hydroxypropyl)methacrylamide copolymer-based nanoconjugate was designed to block tumor cell c-Myc oncogene expression by antisense oligonucleotide. Angiopep-2 peptide was conjugated to the copolymer to facilitate nanodrug crossing of the blood-brain barrier. Systemically administered polymeric nanodrug, alone or in combination with immune checkpoint inhibitor antibody anti-PD-1, was tested in syngeneic mouse model of A20 intracranial brain lymphoma. There was no significant survival difference between saline- and free anti-PD-1-treated groups. However, significant survival advantage vs. saline was observed upon treatment with nanodrug bearing Angiopep-2, H6 (6 histidines for endosome escape), and c-Myc antisense alone and especially when it was combined with anti-PD-1 antibody. Animal survival after combined treatment was also significantly increased vs. free anti-PD-1. Artificial Intelligence-assisted analysis of gene expression database after RNA-seq of tumors was used to find novel immune pathways, molecular targets and the most effective multifunctional drugs together with future drug prediction for brain lymphoma in vivo model. Spectral flow cytometry and RNA-seq analysis revealed a robust activation of tumor infiltrating T lymphocytes with enhanced interferon γ signaling and polarization to M1-type macrophages in treated tumors, which was confirmed by immunofluorescence staining. In summary, a new effective blood-brain barrier crossing nano immuno therapeutic system was developed that effectively blocked tumor c-Myc acting in combination with immune checkpoint inhibitor anti-PD-1 to treat primary brain lymphoma. The treatment improved survival of tumor-bearing animals through activation of both the adaptive and innate immune responses.

Targeted protein degradation (TPD) techniques, particularly proteolysis-targeting chimeras (PROTAC) and molecular glue degraders (MGD), have offered novel strategies in drug discovery. With rapid advancement of computer-aided drug design (CADD) and artificial intelligence-driven drug discovery (AIDD) in the biomedical field, a major focus has become how to effectively integrate these technologies into the TPD drug discovery pipeline to accelerate development, shorten timelines, and reduce costs. Currently, the main research directions for applying CADD and AIDD in TPD include: 1) ternary complex modeling; 2) linker generation; 3) strategies to predict degrader targets, activities and ADME/T properties; 4) In silico degrader design and discovery. Models developed in these areas play a crucial role in target identification, drug design, and optimization at various stages of the discovery process. However, the limited size and quality of datasets related to TPD present challenges, leaving room for further improvement in these models. TPD involves the complex ubiquitin-proteasome system, with numerous factors influencing outcomes. Most current models adopt a static perspective to interpret and predict relevant tasks. In the future, it may be necessary to shift toward dynamic approaches that better capture the intricate relationships among these components. Furthermore, incorporating new and diverse chemical spaces will enhance the precision design and application of TPD agents.

Accurate prediction of protein-ligand binding affinity plays a crucial role in drug discovery. However, determining the binding affinity of protein-ligands through biological experimental approaches is both time-consuming and expensive. Although some computational methods have been developed to predict protein-ligands binding affinity, most existing methods extract the global features of proteins and ligands through separate encoders, without considering to extract the local pocket interaction features of protein-ligand complexes, resulting in the limited prediction accuracy. In this work, we proposed a novel Protein-Ligand binding Affinity prediction method (named PLAGCA) by introducing Graph Cross-Attention mechanism to learn the local three-dimensional (3D) features of protein-ligand pockets, and integrating the global sequence/string features and local graph interaction features of protein-ligand complexes. PLAGCA uses sequence encoding and self-attention to extract the protein/ligand global features from protein FASTA sequences/ligand SMILES strings, adopts graph neural network and cross-attention to extract the protein-ligand local interaction features from the molecular structures of protein binding pockets and ligands. All these features are concatenated and input into a multi-layer perceptron (MLP) for predicting the protein-ligand binding affinity. The experimental results show that our PLAGCA outperforms other state-of-the-art computational methods, and it can effectively predict protein-ligand binding affinity with superior generalization capability. PLAGCA can capture the critical functional residues that are important contribution to the protein-ligand binding.

Over the past decade, epigenetic clocks have emerged as powerful machine learning tools, not only to estimate chronological and biological age but also to assess the efficacy of anti-ageing, cellular rejuvenation and disease-preventive interventions. However, many computational and statistical challenges remain that limit our understanding, interpretation and application of epigenetic clocks. Here, we review these computational challenges, focusing on interpretation, cell-type heterogeneity and emerging single-cell methods, aiming to provide guidelines for the rigorous construction of interpretable epigenetic clocks at cell-type and single-cell resolution.

The performance of machine learning (ML) algorithms is dependent on which dataset it has been trained on. While ML algorithms are increasingly used for lift risk assessment, many algorithms are often trained and tested on controlled simulation datasets, lacking the diversity of the lifting conditions. Consequently, concerns arise regarding their applicability in real-world scenarios characterized by substantial variations in lifting scenarios and postures. Our study investigates the impact of different lifting scenarios on the performance of ML algorithms trained on surface electromyography (sEMG) armband sensor data to classify hand-load levels (2.3 and 6.8 kg). Twelve healthy participants (6 male and 6 female) performed repetitive lifting tasks employing various lifting scenarios, including symmetric (S), asymmetric (A), and free-dynamic (F) techniques. Separate algorithms were developed using diverse training datasets (S, A, S+A, and F), ML classifiers, and sEMG features, and tested using the F dataset, representing unconstrained and naturalistic lifts. The mean accuracy and sensitivity were significantly lower in models trained on constrained (S) datasets compared to those trained on naturalistic lifts (F). The accuracy, precision, and sensitivity of models trained with frequency-domain sEMG features were greater than those trained with the time-domain features. In conclusion, ML algorithms trained on controlled symmetric lifts showed poor performance in predicting loads for dynamic, unconstrained lifts; thus, particular attention is needed when using such algorithms in real-world scenarios.

Skin diseases affect over a third of the global population, yet their impact is often underestimated. Automating the classification of these diseases is essential for supporting timely and accurate diagnoses. This study leverages Vision Transformers, Swin Transformers, and DinoV2, introducing DinoV2 for the first time in dermatology tasks. On a 31-class skin disease dataset, DinoV2 achieves state-of-the-art results with a test accuracy of 96.48 ± 0.0138% and an F1-Score of 97.27%, marking a nearly 10% improvement over existing benchmarks. The robustness of DinoV2 is further validated on the HAM10000 and Dermnet datasets, where it consistently surpasses prior models. Comparative analysis also includes ConvNeXt and other CNN architectures, underscoring the benefits of transformer models. Additionally, explainable AI techniques like GradCAM and SHAP provide global heatmaps and pixel-level correlation plots, offering detailed insights into disease localization. These complementary approaches enhance model transparency and support clinical correlations, assisting dermatologists in accurate diagnosis and treatment planning. This combination of high performance and clinical relevance highlights the potential of transformers, particularly DinoV2, in dermatological applications.

With its ability of joint representation learning and clustering via deep neural networks, the deep clustering have gained significant attention in recent years. Despite the considerable progress, most of the previous deep clustering methods still suffer from three critical limitations. First, they tend to associate some distribution-based clustering loss to the neural network, which often overlook the sample-wise contrastiveness for discriminative representation learning. Second, they generally utilize the features learned at a single layer for the clustering process, which, surprisingly, cannot go beyond a single layer to explore multiple layers for joint multi-layer (multi-stage) learning. Third, they typically use the convolutional neural network (CNN) for clustering images, which focus on local information yet cannot well capture the global dependencies. To tackle these issues, this paper presents a new deep clustering method called pyramid contrastive learning for clustering (PCLC), which is able to incorporate a pyramidal contrastive architecture to jointly enforce contrastive learning and clustering at multiple network layers (or stages). Particularly, for an input image, two types of augmentations are first performed to generate two paralleled augmented views. To bridge the gap between the CNN (for capturing local information) and the Transformer (for reflecting global dependencies), a mixed CNN-Transformer based encoder is utilized as the backbone, whose CNN-Transformer blocks are further divided into four stages, thus giving rise to a pyramid of multi-stage feature representations. Thereafter, multiple stages of twin contrastive learning are simultaneously conducted at both the instance-level and the cluster-level, through the optimization of which the final clustering can be achieved. Extensive experiments on multiple challenging image datasets demonstrate the superior clustering performance of PCLC over the state-of-the-art. The source code is available at https://github.com/Zachary-Chow/PCLC.

Computer-aided detection (CAD) systems have been widely used to assist medical professionals in interpreting medical images, aiding in the detection of potential diseases. Despite their usefulness, CAD systems cannot yet fully replace doctors in diagnosing many conditions due to limitations in current algorithms. Cerebral microbleeds (CMBs) are a critical area of concern for neurological health, and accurate detection of CMBs is essential for understanding their impact on brain function. This study aims to improve CMB detection by enhancing existing machine learning algorithms.

This paper presents four CNN-based algorithms designed to enhance CMB detection. The detection methods are categorized into traditional machine learning approaches and deep learning-based methods. The traditional methods, while computationally efficient, offer lower sensitivity, while CNN-based approaches promise greater accuracy. The algorithms proposed in this study include a multi-channel CNN with optimized architecture and a multiscale CNN structure, both of which were designed to reduce false positives and improve overall performance.

The first CNN algorithm, with an optimized multi-channel architecture, demonstrated a sensitivity of 99.6 %, specificity of 99.3 %, and accuracy of 99.5 %. The fourth algorithm, based on a stable multiscale CNN structure, achieved sensitivity of 98.2 %, specificity of 97.4 %, and accuracy of 97.8 %. Both algorithms exhibited a significant reduction in false positives compared to traditional methods. The experiments conducted confirm the effectiveness of these algorithms in improving the precision and reliability of CMB detection.

The proposed CNN-based algorithms demonstrate a significant advancement in the automated detection of CMBs, with notable improvements in sensitivity, specificity, and accuracy. These results underscore the potential of deep learning models, particularly CNNs, in enhancing CAD systems for neurological disease detection and reducing diagnostic errors. Further research and optimization may allow these algorithms to be integrated into clinical practices, providing more reliable support for healthcare professionals.

Artificial Neural Networks (ANNs) aim at mimicking information processing in biological networks. In cognitive neuroscience, graph modeling is a powerful framework widely used to study brain structural and functional connectivity. Yet, the extension of graph modeling to ANNs has been poorly explored especially in terms of functional connectivity (i.e. the contextual change of the activity's units in networks). In the perspective of designing more robust and interpretable ANNs, we study how a brain-inspired graph-based approach can be extended and used to investigate ANN properties and behaviors. We focus our study on different continual learning strategies inspired by the biological mechanisms and modeled with ANNs. We show that graph modeling offers a simple and elegant framework to deeply investigate ANNs, compare their performances, and explore deleterious behaviors such as catastrophic forgetting.

Histopathology image analysis plays a pivotal role in disease diagnosis and treatment planning, relying heavily on accurate nuclei segmentation for extracting vital cellular information. In recent years, artificial intelligence (AI) and in particular deep learning models have been applied successfully in solving computational pathology image analysis tasks. The You Only Look Once (YOLO) object detection framework, which is based on a convolutional neural network (CNN) architecture has gained traction across various domains for its real-time processing capabilities. This systematic review aims to comprehensively explore and evaluate the advancements, challenges, and applications of YOLO-based methodologies in nuclei segmentation within the domain of histopathological images. The review encompasses a structured analysis of recent literature, focusing on the utilization of YOLO variants for nuclei segmentation. Key methodologies, training strategies, dataset specifics, and performance metrics are evaluated to elucidate the strengths and limitations of YOLO in this context. Additionally, the review highlights the unique characteristics of YOLO that enable efficient object detection and delineation of nuclei structures, offering a comparative analysis against traditional segmentation approaches. This systematic review underscores the promising outcomes achieved through YOLO-based architectures, emphasizing their potential for accurate and rapid nuclei segmentation. Furthermore, it identifies persistent challenges such as handling variances in nuclei appearances, optimizing model architectures for histopathological images, and improving generalization across diverse datasets. Insights derived from this review can provide a foundation for future research directions and enhancements in nuclei segmentation methodologies using YOLO within histopathology, fostering advancements in disease diagnosis and biomedical research.

To develop an interpretable deep learning (XDL) model based on superb microvascular imaging (SMI) for the noninvasive diagnosis of the degree of interstitial fibrosis (IF) in chronic kidney disease (CKD).

We included CKD patients who underwent renal biopsy, two-dimensional ultrasound, and SMI examinations between May 2022 and October 2023. Based on the pathological IF score, they were divided into two groups: minimal-mild IF (≤25%) and moderate-severe IF (>25%). An XDL model based on the SMI while establishing an ultrasound radiomics model and a color doppler ultrasonography (CDUS) model as the control group. The utility of the proposed model was evaluated using the receiver operating characteristic curve (ROC) and decision curve analysis.

In total, 365 CKD patients were included herein. In the validation group, AUCs of the ROC curves for the DL, ultrasound radiomics, and CDUS models were 0.854, 0.784, and 0.745, respectively. In the test group, AUCs of the ROC curve for the DL ultrasound radiomics, and CDUS models were 0.824, 0.792, and 0.752, respectively. The pie chart and heat map based on Shapley additive explanations (SHAP) provided substantial interpretability for the model.

Compared with the ultrasound radiomics and CDUS models, the DL model based on the SMI had higher accuracy in the noninvasive judgment of the degree of IF in CKD patients. Pie and heat maps based on Shapley can explain which image regions are helpful in diagnosing the degree of IF.

Deep learning, a leading technology in artificial intelligence (AI), has shown remarkable potential in revolutionizing nuclear medicine.

This review presents recent advancements in deep learning applications, particularly in nuclear medicine imaging, lesion detection, and radiopharmaceutical therapy.

Leveraging various neural network architectures, deep learning has significantly enhanced the accuracy of image reconstruction, lesion segmentation, and diagnosis, improving the efficiency of disease detection and treatment planning. The integration of deep learning with functional imaging techniques such as positron emission tomography (PET) and single-photon emission computed tomography (SPECT) enable more precise diagnostics, while facilitating the development of personalized treatment strategies. Despite its promising outlook, there are still some limitations and challenges, particularly in model interpretability, generalization across diverse datasets, multimodal data fusion, and the ethical and legal issues faced in its application.

As technological advancements continue, deep learning is poised to drive substantial changes in nuclear medicine, particularly in the areas of precision healthcare, real-time treatment monitoring, and clinical decision-making. Future research will likely focus on overcoming these challenges and further enhancing model transparency, thus improving clinical applicability.

Diseases that develop necrotizing vasculitis of cutaneous muscular arteries include cutaneous arteritis (CA) and polyarteritis nodosa (PAN). It is difficult to distinguish them based on skin biopsy findings alone. This study demonstrated that artificial intelligence (AI) can discriminate them based on skin biopsy findings and revealed where AI focuses on the image. Ninety-three hematoxylin-and-eosin images of CA and 19 PAN images were used. Among them, 85 CA and 17 PAN images were used to train AI; thereafter, AI was challenged to classify the remaining images. The same test images were evaluated by 26 pathologists with different years of experience. AI accuracy was 75.2 %, whereas that of pathologists was 42.8 %. Gradient-weighted class activation mapping (Grad-CAM) indicated that AI focused on connective tissues around the affected vessels rather than the affected vessels. Twenty-two of the 26 pathologists were randomly divided into two groups of 11 each, one of which referred to Grad-CAM images and was challenged in the second-round test of images different from the first round. The accuracy significantly improved after referring to Grad-CAM images, whereas it was equivalent to the first round without referring to Grad-CAM images. In the survey after the second-round test, pathologists who referred to Grad-CAM images suggested that inflammation and fibrosis in the surrounding connective tissues in PAN might be abundant compared to CA. AI may be useful for histological differentiation between CA and PAN and can help pathologists improve the ability of discriminating CA and PAN based on histological findings of skin biopsy specimens.

Medical Visual Question Answering (Med-VQA) is a critical multimodal task with the potential to address the scarcity and imbalance of medical resources. However, most existing studies overlook the limitations of the inconsistency in information density between medical images and text, as well as the long-tail distribution in datasets, which continue to make Med-VQA an open challenge. To overcome these issues, this study proposes a Language-Guided Progressive Fusion Network (LGPFN) with three key modules: Question-Guided Progressive Multimodal Fusion (QPMF), Language-Gate Mechanism (LGM), and Triple Semantic Feature Alignment (TriSFA). QPMF progressively guides the fusion of visual and textual features using both global and local question representations. LGM, a linguistic rule-based module, distinguishes between Closed-Ended (CE) and Open-Ended (OE) samples, directing the fused features to the appropriate classifiers. Finally, TriSFA captures the rich semantic information of OE answers and mine the underlying associations among fused features, predicted answers, and ground truths, aligning them in a ternary semantic feature space. The proposed LGPFN framework outperforms existing state-of-the-art models, achieving the best overall accuracies of 80.39%, 84.07%, 75.74%, and 70.60% on the VQA-RAD, SLAKE, PathVQA, and VQA-Med 2019 datasets, respectively. These results demonstrate the effectiveness and generalizability of the proposed model, underscoring its potential as a medical Artificial Intelligent (AI) agent that could benefit universal health coverage.

The rapidly advancing field of artificial intelligence (AI) has transformed numerous scientific domains, including biology, where a vast and complex volume of data is available for analysis. This paper provides a comprehensive overview of the current state of AI-driven methodologies in genomics, proteomics, and systems biology. We discuss how machine learning algorithms, particularly deep learning models, have enhanced the accuracy and efficiency of embedding sequences, motif discovery, and the prediction of gene expression and protein structure. Additionally, we explore the integration of AI in the embedding and analysis of biological networks, including protein-protein interaction networks and multi-layered networks. By leveraging large-scale biological data, AI techniques have enabled unprecedented insights into complex biological processes and disease mechanisms. This work underlines the potential of applying AI to complex biological data, highlighting current applications and suggesting directions for future research to further explore AI in this rapidly evolving field.

The sensitivity of reverse-transcription polymerase chain reaction (RT-PCR) is limited for diagnosis of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). Chest computed tomography (CT) is reported to have high sensitivity; however, given the limited availability of chest CT during a pandemic, the assessment of more readily available imaging, such as chest radiographs, augmented by artificial intelligence may substitute for the detection of the features of coronavirus disease 2019 (COVID-19) pneumonia.

We trained a deep convolutional neural network to detect SARS-CoV-2 pneumonia using publicly available chest radiography imaging data including 8,851 normal, 6,045 pneumonia, and 200 COVID-19 pneumonia radiographs. The entire cohort was divided into training (n = 13,586) and test groups (n = 1510). We assessed the accuracy of prediction with independent external data.

The sensitivity and positive predictive values of the assessment by artificial intelligence were 96.8% and 90.9%, respectively. In the first external validation of 204 chest radiographs among 107 patients with confirmed COVID-19, the artificial intelligence algorithm correctly identified 174 (85%) chest radiographs as COVID-19 pneumonia among 97 (91%) patients. In the second external validation with 50 immunocompromised patients with leukemia, the higher probability of the artificial intelligence assessment for COVID-19 was correlated with suggestive features of COVID-19, while the normal chest radiographs were closely correlated with the likelihood of normal chest radiographs by the artificial intelligence prediction.

The assessment method by artificial intelligence identified suspicious lung lesions on chest radiographs. This novel approach can identify patients for confirmatory chest CT before the progression of COVID-19 pneumonia.

Generative artificial intelligence (GAI) has recently achieved significant success, enabling anyone to create texts, images, videos and even computer codes while providing insights that might not be possible with traditional tools. To stimulate future research, this work provides a brief summary of the ongoing and historical developments in GAI over the past 70 years. The achievements are grouped into four categories: (i) rule-based generative systems that follow specialized rules and instructions, (ii) model-based generative algorithms that produce new content based on statistical or graphical models, (iii) deep generative methodologies that utilize deep neural networks to learn how to generate new content from data and (iv) foundation models that are trained on extensive datasets and capable of performing a variety of generative tasks. This paper also reviews successful generative applications and identifies open challenges posed by remaining issues. In addition, this paper describes potential research directions aimed at better utilizing, understanding and harnessing GAI technologies.

Considering Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) network approaches have shown promising image classification performance, the aim of this study was to compare the performance of novel Convolutional Neural Network and Long Short-Term Memory (CNN-LSTM) architectures with a classic CNN for classification of panoramic radiographs with inflammatory periapical lesions.

A dataset of 356 panoramic radiographs with periapical lesions and 769 control images were retrospectively collected and divided into training, validation, and testing sets. Next, four different models were constructed: a classic CNN, a classic LSTM, a cascaded CNN-LSTM, and parallel CNN-LSTM architecture. In each model the CNN took the full panoramic radiograph as input while the LSTM network ran on the images divided into 6 sequential patches. Sensitivity, specificity, and Area Under the Receiver-Operating Curve (AUC) were calculated. McNemar's test compared the sensitivity and specificity between the classic CNN and the other models.

Parallel CNN-LSTM had a significantly higher sensitivity than classic CNN for detecting periapical lesions (95% vs. 81%, 95% confidence interval for the difference = 6 - 22 %, P = 0.002), while also exhibiting the best overall performance of the four models [AUC = 96% vs. 90% (classic CNN), 92% (classic LSTM), and 94% (cascaded CNN-LSTM)].

The parallel CNN-LSTM architecture outperformed the classic CNN for classification of panoramic radiographs with inflammatory periapical lesions.

Combining CNN and LSTM models improves the classification of panoramic radiographs with and without inflammatory periapical lesions.

To establish an advanced automated bone marrow (BM) segmentation model on whole-body (WB-)MRI in monoclonal plasma cell disorders (MPCD), and to demonstrate its robust performance on multicenter datasets with severe myeloma-related pathologies.

The study cohort comprised multi-vendor, multi-protocol imaging data acquired with varying field strength across 8 different centers. In total, 210 WB-MRIs of 207 MPCD patients were included. An nnU-Net algorithm was established for segmenting the individual bone marrow spaces (BMS) of the spine, pelvis, humeri and femora (advanced segmentation model). For this task, 186 T1-weighted (T1w) WB-MRIs from center 1 were used in the training set. Test sets included 12 T1w WB-MRIs from center 2 (I) and 9 T1w WB-MRIs from centers 3-8 (II). Example cases were included to showcase segmentation performance on T1w WB-MRIs with extensive tumor load. The segmentation accuracy of the advanced segmentation model was compared to a prior established basic segmentation model by calculating Dice scores and using the Wilcoxon signed-rank test.

The mean Dice score on the individual BMS was 0.89±0.13 (test set I) and 0.88±0.11 (test set II), significantly higher than the Dice scores of a prior basic model (p<0.05). Dice scores for the BMS of the individual bones ranged from 0.77 to 0.96 (test set I), and 0.81 to 0.95 (test set II). BM altered by myeloma-relevant pathologies, artifacts or low imaging quality was precisely segmented.

The advanced model performed reliable, automated segmentations, even on heterogeneously acquired multicenter WB-MRIs with severe pathologies.

Recent developments in artificial intelligence (AI) may significantly alter physiological research and healthcare delivery. Whereas AI applications in medicine have historically been trained for specific tasks, recent technological advances have produced models trained on more diverse datasets with much higher parameter counts. These new, "foundation" models raise the possibility that more flexible AI tools can be applied to a wider set of healthcare tasks than in the past. This review describes how these newer models differ from conventional task-specific AI, which relies heavily on focused datasets and narrow, specific applications. By examining the integration of AI into diagnostic tools, personalized treatment strategies, biomedical research, and healthcare administration, we highlight how these newer models are revolutionizing predictive healthcare analytics and operational workflows. In addition, we address ethical and practical considerations associated with the use of foundation models by highlighting emerging trends, calling for changes to existing guidelines, and emphasizing the importance of aligning AI with clinical goals to ensure its responsible and effective use.