Capability Of Model Research Articles

An exploratory machine learning model for predicting advanced liver fibrosis in autoimmune hepatitis patients: A preliminary study.

Advanced fibrosis is a crucial stage in the progression of autoimmune hepatitis (AIH), where fibrosis can either regress or advance. This study aims to leverage machine learning (ML) models for the assessment of advanced liver fibrosis in AIH patients using routine clinical features. A total of 233 patients diagnosed with AIH and underwent liver biopsy were included in the discovery cohort. The dataset was randomly split into training and testing sets. Patients were categorized into groups with no/minimal/moderate fibrosis and advanced fibrosis. Six ML models were employed to identify the optimal model. Subsequently, the predictive capability of the best ML model was validated in an additional cohort (n = 33) and compared with conventional noninvasive fibrosis scores. Three key clinical features, including prothrombin time (PT), albumin (ALB), and ultrasound spleen thickness (UTST), were analyzed by least absolute shrinkage and selection operator (LASSO) regression. In the training set, the random forest (RF) model showed the highest diagnostic performance in predicting advanced fibrosis stage (AUC=0.951). In the testing cohort and validation cohort, the RF model maintained high accuracy (AUC = 0.863 and AUC = 0.843). Additionally, the random forest model outperformed the conventional noninvasive fibrosis scores. ML models, particularly the RF model, can help improve the discrimination of advanced liver fibrosis in patients with AIH.

Analyzing sorbitol biosynthesis using a metabolic network flux model of a lichenized strain of the green microalga Diplosphaera chodatii.

Diplosphaera chodatii, a unicellular terrestrial microalga found either free-living or in association with lichenized fungi, protects itself from desiccation by synthesizing and accumulating low-molecular-weight carbohydrates such as sorbitol. The metabolism of this algal species and the interplay of sorbitol biosynthesis with its growth, light absorption, and carbon dioxide fixation are poorly understood. Here, we used a recently available genome assembly for D. chodatii to develop a metabolic flux model and analyze the alga's metabolic capabilities, particularly, for sorbitol biosynthesis. The model contains 151 genes, 155 metabolites, and 194 unique metabolic reactions participating in 12 core metabolic pathways and five compartments. Both photoautotrophic and mixotrophic growths of D. chodatii were supported by the metabolic model. In the presence of glucose, mixotrophy led to higher biomass and sorbitol yields. Additionally, the model predicted increased starch biosynthesis at high light intensities during photoautotrophic growth, an indication that the "overflow hypothesis-stress-driven metabolic flux redistribution" could be applied to D. chodatii. Furthermore, the newly developed metabolic model of D. chodatii, iDco_core, captures both linear and cyclic electron flow schemes characterized in photosynthetic microorganisms and suggests a possible adaptation to fluctuating water availability during periods of desiccation. This work provides important new insights into the predicted metabolic capabilities of D. chodatii, including a potential biotechnological opportunity for industrial sorbitol biosynthesis.IMPORTANCELichenized green microalgae are vital components for the survival and growth of lichens in extreme environmental conditions. However, little is known about the metabolism and growth characteristics of these algae as individual microbes. This study aims to provide insights into some of the metabolic capabilities of Diplosphaera chodatii, a lichenized green microalgae, using a recently assembled and annotated genome of the alga. For that, a metabolic flux model was developed simulating the metabolism of this algal species and allowing for studying the algal growth, light absorption, and carbon dioxide fixation during both photoautotrophic and mixotrophic growth, in silico. An important capability of the new metabolic model of D. chodatii is capturing both linear and cyclic electron flow mechanisms characterized in several other microalgae. Moreover, the model predicts limits of the metabolic interplay between sorbitol biosynthesis and algal growth, which has potential applications in assisting the design of bio-based sorbitol production processes.

Energy-based approach to the assessment of traffic flow

This article focuses on modeling vehicle acceleration noise in different road conditions, emphasizing urban, highway, and rural roads in Ukraine. Acceleration noise, which refers to the fluctuations in a vehicle's acceleration, is a critical factor in vehicle safety, fuel efficiency, and driving comfort. The research aims to improve current vehicle dynamics models by integrating multi-body dynamics and machine learning algorithms, allowing for more precise predictions of acceleration variability in real-time. The study is based on the existing literature, showing that road surface quality significantly affects acceleration noise. With frequent stop-and-go traffic, urban roads produce moderate but irregular noise patterns. Highways show stable acceleration noise at moderate speeds, but noise increases sharply as vehicles approach higher speeds due to aerodynamic forces. Rural roads, especially those in poor condition, exhibit the highest variability in acceleration noise, even at low speeds. The proposed model has been validated using real-world data. It demonstrates a strong correlation between the predictions and actual vehicle behavior on various road types. One of the key innovations in this research is the use of machine learning to adjust model parameters in real-time dynamically. This adaptive approach enhances the model’s accuracy and applicability, especially in intelligent transport systems. The model can inform traffic management strategies, allowing for real-time adjustments to speed limits, traffic signals, and routing decisions based on road conditions. This contributes to safer, more efficient, and sustainable transport systems, particularly in regions with inconsistent road infrastructure. The research concludes that integrating acceleration noise modeling into intelligent transport systems can significantly improve traffic flow and vehicle safety. Future research will expand the dataset to include a broader range of vehicle types and road conditions, further refining the model's predictive capabilities.

Development of heat exchanger modelling capability for a finite volume aeroelasticity solver

Abstract Heat exchangers are frequently used in aero engines and are known to significantly affect the surrounding steady and unsteady flow. In certain applications, they may thereby also influence the aeroelastic stability of upstream or downstream components but there is limited research on this in the public domain. This paper aims to demonstrate the influence of heat exchangers on unsteady flows relevant to aeroelastic problems. This is achieved by developing heat exchanger modelling capability for an in-house finite volume aeroelasticity solver, for which heat exchanger is represented as a porous medium, as this is the established approach in existing aerodynamic studies using commercial CFD software. The governing equations for a Darcy-Forchheimer porous media model suitable for unsteady and compressible flows are presented, which are derived by the application of volume averaging theory to the Navier-Stokes equations. The implementation of this model within the time integration method used for the solver is then described and verified by comparison of results for steady flows against an established commercial CFD solver, where close agreement between both in-house and commercial solvers has been observed. Lastly a preliminary demonstration of the capability to model unsteady heat exchanger flows is presented by application to an aeroacoustic problem, where the interaction of the pressure waves and the heat exchanger is investigated.

IA-YOLO: A Vatica Segmentation Model Based on an Inverted Attention Block for Drone Cameras

The growing use of drones in precision agriculture highlights the needs for enhanced operational efficiency, especially in the scope of detection tasks, even in segmentation. Although the ability of computer vision based on deep learning has made remarkable progress in the past ten years, the segmentation of images captured by Unmanned Aerial Vehicle (UAV) cameras, an exact detection task, still faces a conflict between high precision and low inference latency. Due to such a dilemma, we propose IA-YOLO (Inverted Attention You Only Look Once), an efficient model based on IA-Block (Inverted Attention Block) with the aim of providing constructive strategies for real-time detection tasks using UAV cameras. The working details of this paper are outlined as follows: (1) We construct a component named IA-Block, which is integrated into the YOLOv8-seg structure as IA-YOLO. It specializes in pixel-level classification of UAV camera images, facilitating the creation of exact maps to guide agricultural strategies. (2) In experiments on the Vatica dataset, compared with any other lightweight segmentation model, IA-YOLO achieves at least a 3.3% increase in mAP (mean Average Precision). Further validation on diverse species datasets confirms its robust generalization. (3) Without overloading the complex attention mechanism and deeper and deeper network, a stem that incorporates efficient feature extraction components, IA-Block, still possess credible modeling capabilities.

Decoding Market Emotions: The Synergy of Sentiment Analysis and AI in Stock Market Predictions

The stock market is influenced not only by traditional financial metrics but also by psychological factors such as emotions, opinions, and sentiments. In recent years, the integration of sentiment analysis and artificial intelligence (AI) has transformed stock market forecasting by enabling traders and investors to interpret market behavior more effectively. Sentiment analysis, a subset of natural language processing (NLP), analyzes textual data from diverse sources like news articles and social media to gauge public sentiment—positive, negative, or neutral—toward market conditions. This analysis bridges the gap between quantitative data and investor psychology, revealing insights that traditional metrics might not capture. With the rise of social media and online forums, the volume of opinion data has surged, necessitating advanced technologies for real-time processing and interpretation. AI, mainly through machine learning and deep learning models like GPT and BERT, is crucial in efficiently analyzing vast datasets, detecting patterns, and predicting market trends. These AI-powered tools can combine sentiment data with historical market trends, providing a holistic view of market dynamics. The advanced capabilities of AI models to comprehend nuances such as sarcasm and irony further enhance sentiment detection, allowing for more accurate predictions. While integrating sentiment analysis and AI in financial markets offers numerous advantages, it also faces challenges such as algorithmic bias, data privacy issues, and the unpredictability of human emotions. This study aims to explore the integration of sentiment analysis and AI in stock market predictions, assess the accuracy of AI-driven predictions compared to traditional methods, and analyze case studies of successful applications in financial markets. Through this, the study seeks to contribute to the evolving landscape of financial forecasting by demonstrating the potential of AI and sentiment analysis in shaping market behavior understanding and decision-making processes.

Impact Assessment of Coupling Mode of Hydrological Model and Machine Learning Model on Runoff Simulation: A Case of Washington

The inherent uncertainties in traditional hydrological models present significant challenges for accurately simulating runoff. Combining machine learning models with traditional hydrological models is an essential approach to enhancing the runoff modeling capabilities of hydrological models. However, research on the impact of mixed models on runoff simulation capability is limited. Therefore, this study uses the traditional hydrological model Simplified Daily Hydrological Model (SIMHYD) and the machine learning model Long Short Term Memory (LSTM) to construct two coupled models: a direct coupling model and a dynamically improved predictive validity hybrid model. These models were evaluated using the US CAMELS dataset to assess the impact of the two model combination methods on runoff modeling capabilities. The results indicate that the runoff modeling capabilities of both combination methods were improved compared to individual models, with the combined forecasting model for dynamic prediction effectiveness (DPE) demonstrating the optimal modeling capability. Compared with LSTM, the mixed model showed a median increase of 12.8% in Nash Sutcliffe efficiency (NSE) of daily runoff during the validation period, and a 12.5% increase compared to SIMHYD. In addition, compared with the LSTM model, the median Nash Sutcliffe efficiency (NSE) of the hybrid model simulating high flow results increased by 23.6%, and compared with SIMHYD, it increased by 28.4%. At the same time, the stability of the hybrid model simulating low flow was significantly improved. In performance testing involving varying training period lengths, the DPE model trained for 12 years exhibited the best performance, showing a 3.5% and 1.5% increase in the median NSE compared to training periods of 6 years and 18 years, respectively.

Three-Dimensional Broad Learning Gravity Data Inversion Using Single-Anomaly Training Samples

Gravity data inversion is of critical importance in geophysics, encompassing a range of applications, such as the exploration of geological resources, the identification of geological structures, and the detection of groundwater resources. This study proposes a three-dimensional (3D) machine learning approach to enhance the efficiency of the aforementioned exploration tasks by leveraging gravity data. The mapping relationship between gravity data and subsurface density structures is modeled by the broad learning network, distinguished by its high training efficiency and robust modeling capability. Notably, the proposed inversion method obviates the constraints on the number of anomalies prior to the inversion process. This is achieved by setting one anomaly with varied locations for different training samples. Numerical and field data applications demonstrate the efficiency of the proposed 3D machine learning gravity data inversion method, especially in automatically determining the number of anomalies. In particular, the proposed method produced accurate density inversion results in the field application, aiding in the identification of potential oil and gas reservoirs in the target region and offering the potential for broader application in other resource exploration. The proposed inversion method can promote the construction of density structures of subsurfaces based on gravity data.

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.

On the efficacy of sparse representation approaches for determining nonlinear structural system equations of motion

Abstract A sparsity-based optimization approach is presented for determining the equations of motion of stochastically excited nonlinear structural systems. This is done by utilizing measured excitation-response realizations in the formulation of the related optimization problem, and by considering a library of candidate functions for representing the system governing dynamics. Note that a novel aspect of the approach relates to treating, also, systems endowed with fractional derivative elements. Clearly, this is of significant importance to a multitude of diverse applications in engineering mechanics taking into account the enhanced modeling capabilities of fractional calculus. Further, the fundamental theoretical and computational aspects of various representative, state-of-the-art, numerical schemes for solving the derived sparsity-based optimization problem are reviewed and discussed. A Bayesian compressive sampling approach that exhibits the additional advantage of quantifying the uncertainty of the estimates is considered as well. Furthermore, comparisons and a critical assessment of the employed numerical schemes are provided with respect to their efficacy in determining the nonlinear structural system equations of motion. In this regard, two illustrative numerical examples are considered pertaining to a nonlinear tuned-mass-damper-inerter vibration control system, and to a nonlinear electromechanical energy harvester, both endowed with fractional derivative elements.

Advanced State-of-Health Estimation for Lithium-Ion Batteries Using Multi-Feature Fusion and KAN-LSTM Hybrid Model

Accurate assessment of battery State of Health (SOH) is crucial for the safe and efficient operation of electric vehicles (EVs), which play a significant role in reducing reliance on non-renewable energy sources. This study introduces a novel SOH estimation method combining Kolmogorov–Arnold Networks (KAN) and Long Short-Term Memory (LSTM) networks. The method is based on fully charged battery characteristics, extracting key parameters such as voltage, temperature, and charging data collected during cycles. Validation was conducted under a temperature range of 10 °C to 30 °C and different charge–discharge current rates. Notably, temperature variations were primarily caused by seasonal changes, enabling the experiments to more realistically simulate the battery’s performance in real-world applications. By enhancing dynamic modeling capabilities and capturing long-term temporal associations, experimental results demonstrate that the method achieves highly accurate SOH estimation under various charging conditions, with low mean absolute error (MAE) and root mean square error (RMSE) values and a coefficient of determination (R2) exceeding 97%, significantly improving prediction accuracy and efficiency.

Probing the limits and capabilities of diffusion models for the anatomic editing of digital twins

Numerical simulations of cardiovascular device deployment within digital twins of patient-specific anatomy can expedite and de-risk the device design process. Nonetheless, the exclusive use of patient-specific data constrains the anatomic variability that can be explored. We study how Latent Diffusion Models (LDMs) can edit digital twins to create digital siblings. Siblings can serve as the basis for comparative simulations, which can reveal how subtle anatomic variations impact device deployment, and augment virtual cohorts for improved device assessment. Using a case example centered on cardiac anatomy, we study various methods to generate digital siblings. We specifically introduce anatomic variation at different spatial scales or within localized regions, demonstrating the existence of bias toward common anatomic features. We furthermore leverage this bias for virtual cohort augmentation through selective editing, addressing issues related to dataset imbalance and diversity. Our framework delineates the capabilities of diffusion models in synthesizing anatomic variation for numerical simulation studies.

Intersection of Performance, Interpretability, and Fairness in Neural Prototype Tree for Chest X-Ray Pathology Detection: Algorithm Development and Validation Study.

While deep learning classifiers have shown remarkable results in detecting chest X-ray (CXR) pathologies, their adoption in clinical settings is often hampered by the lack of transparency. To bridge this gap, this study introduces the neural prototype tree (NPT), an interpretable image classifier that combines the diagnostic capability of deep learning models and the interpretability of the decision tree for CXR pathology detection. This study aimed to investigate the utility of the NPT classifier in 3 dimensions, including performance, interpretability, and fairness, and subsequently examined the complex interaction between these dimensions. We highlight both local and global explanations of the NPT classifier and discuss its potential utility in clinical settings. This study used CXRs from the publicly available Chest X-ray 14, CheXpert, and MIMIC-CXR datasets. We trained 6 separate classifiers for each CXR pathology in all datasets, 1 baseline residual neural network (ResNet)-152, and 5 NPT classifiers with varying levels of interpretability. Performance, interpretability, and fairness were measured using the area under the receiver operating characteristic curve (ROC AUC), interpretation complexity (IC), and mean true positive rate (TPR) disparity, respectively. Linear regression analyses were performed to investigate the relationship between IC and ROC AUC, as well as between IC and mean TPR disparity. The performance of the NPT classifier improved as the IC level increased, surpassing that of ResNet-152 at IC level 15 for the Chest X-ray 14 dataset and IC level 31 for the CheXpert and MIMIC-CXR datasets. The NPT classifier at IC level 1 exhibited the highest degree of unfairness, as indicated by the mean TPR disparity. The magnitude of unfairness, as measured by the mean TPR disparity, was more pronounced in groups differentiated by age (chest X-ray 14 0.112, SD 0.015; CheXpert 0.097, SD 0.010; MIMIC 0.093, SD 0.017) compared to sex (chest X-ray 14 0.054 SD 0.012; CheXpert 0.062, SD 0.008; MIMIC 0.066, SD 0.013). A significant positive relationship between interpretability (ie, IC level) and performance (ie, ROC AUC) was observed across all CXR pathologies (P<.001). Furthermore, linear regression analysis revealed a significant negative relationship between interpretability and fairness (ie, mean TPR disparity) across age and sex subgroups (P<.001). By illuminating the intricate relationship between performance, interpretability, and fairness of the NPT classifier, this research offers insightful perspectives that could guide future developments in effective, interpretable, and equitable deep learning classifiers for CXR pathology detection.

Feature selection and response prediction on a suspension bridge due to wind effect by machine learning

This research introduces an efficient machine learning approach for predicting the structural health of long-span suspension bridges, focusing on their responses to dynamic environmental load like earthquakes and wind. Unlike traditional analytical methods that rely on simplifications, this study utilizes real-world data from anemometers and accelerometers to enhance prediction accuracy and efficiency. Also instead of using whole signals, the study uses fewer time frame signal. The main contribution of this paper is the application of support vector regression (SVR), optimized through Observer-Teacher-Learner-Based-Optimization (OTLBO) for parameter tuning. The optimization process is further refined using Multi-Objective Observer-Teacher-Learner-Based-Optimization (MOOTLBO), targeting feature selection and dimension reduction to improve model performance significantly. Hardanger Bridge serves as a case study, demonstrating the model's capability to accurately predict acceleration responses to wind. By inputting wind sensor data and analyzing acceleration outputs, the enhanced SVR model, through OTLBO and MOOTLBO, shows remarkable predictive accuracy, validated against six performance indices. This methodological advancement suggests that the machine learning-based model could potentially supplant traditional finite element models, offering a more adaptable, efficient, and accurate tool for structural health monitoring (SHM). This advancement addresses critical discrepancies in SHM systems, such as data gaps or sensor malfunctions, by providing a reliable prediction method that ensures the ongoing safety and integrity of bridge structures.

A Comparative Analysis of Instruction Fine-Tuning Large Language Models for Financial Text Classification

Large Language Models (LLMs) have demonstrated impressive capabilities across diverse Natural Language Processing (NLP) tasks, including language understanding, reasoning, and generation. However, general-domain LLMs often struggle with financial tasks due to the technical and specialized nature of financial texts. This study investigates the efficacy of instruction fine-tuning smaller-scale LLMs, including Mistral-7B, Llama3-8B, and Phi3-mini, to enhance their performance in financial text classification tasks. We fine-tuned both instruction-tuned and base models across four financial classification tasks, achieving significant improvements in task-specific performance. Furthermore, we evaluated the zero-shot capabilities of these fine-tuned models on three unseen complex financial tasks, including argument classification, deal completeness classification, and causal classification. Our results indicate while base model fine-tuning led to greater degradation, instruction-tuned models maintained more robust performance. To address this degradation, we employed model merging techniques, integrating single-task domain-specific fine-tuned models with the base model. Using this merging method resulted in significant enhancements in zero-shot performance, even exceeding the original model’s accuracy on certain datasets. Our findings underscore the effectiveness of instruction fine-tuning and model merging for adapting LLMs to specialized financial text classification tasks.

A Centrality-Weighted Bidirectional Encoder Representation from Transformers Model for Enhanced Sequence Labeling in Key Phrase Extraction from Scientific Texts

Deep learning approaches, utilizing Bidirectional Encoder Representation from Transformers (BERT) and advanced fine-tuning techniques, have achieved state-of-the-art accuracies in the domain of term extraction from texts. However, BERT presents some limitations in that it primarily captures the semantic context relative to the surrounding text without considering how relevant or central a token is to the overall document content. There has also been research on the application of sequence labeling on contextualized embeddings; however, the existing methods often rely solely on local context for extracting key phrases from texts. To address these limitations, this study proposes a centrality-weighted BERT model for key phrase extraction from text using sequence labelling (CenBERT-SEQ). The proposed CenBERT-SEQ model utilizes BERT to represent terms with various contextual embedding architectures, and introduces a centrality-weighting layer that integrates document-level context into BERT. This layer leverages document embeddings to influence the importance of each term based on its relevance to the entire document. Finally, a linear classifier layer is employed to model the dependencies between the outputs, thereby enhancing the accuracy of the CenBERT-SEQ model. The proposed CenBERT-SEQ model was evaluated against the standard BERT base-uncased model using three Computer Science article datasets, namely, SemEval-2010, WWW, and KDD. The experimental results show that, although the CenBERT-SEQ and BERT-base models achieved higher and close comparable accuracy, the proposed CenBERT-SEQ model achieved higher precision, recall, and F1-score than the BERT-base model. Furthermore, a comparison of the proposed CenBERT-SEQ model to that of related studies revealed that the proposed CenBERT-SEQ model achieved a higher accuracy, precision, recall, and F1-score of 95%, 97%, 91%, and 94%, respectively, than related studies, showing the superior capabilities of the CenBERT-SEQ model in keyphrase extraction from scientific documents.

A fine‐grained image classification method based on information interaction

AbstractTo enhance the accuracy of fine‐grained image classification and address challenges such as excessive interference factors within the dataset, inadequate extraction of local key features, and insufficient channel semantic association, a dual‐branch information interaction model that integrates convolutional neural networks (CNN) with Vision Transformers is proposed. This model leverages the Vision Transformer branch to extract global features, which are subsequently combined with the CNN branch to further augment the model's capability for local information extraction. In order to enhance the ability of the CNN branch to extract global information and reduce the loss of feature information, a feature enhancement module is added to the CNN branch. Since the Vision Transformer branch directly convolves with the convolution kernel will result in the inability to learn the underlying features of the image, a shallow feature extraction module is proposed, and the CNN and Vision Transformer branches interact with the information of the dual branches through the down‐sampling Down module and the up‐sampling UP module. The accuracy of the improved method on CUB‐200‐2011, Stanford Cars and FGVC‐Aircraft fine‐grained image classification datasets are 95.2%, 97.1% and 96.9%, respectively. The experimental results show that the method has good generalization on different datasets.

STCM-GCN: a spatial-temproal prediction method for traffic crashes under road network constraints

Existing research faces challenges in accurately predicting crashes due to the unreasonable selection of spatial units, biased crash data collection, and insufficient integration of multi-source data. To address these issues, Graph Neural Networks (GNNs) for node classification are employed to predict crashes at macroscopic road level. Crash alarm data are incorporated as a supplement to official archive data to ensure the spatial–temporal distribution's authenticity and to mitigate data sparsity. Additionally, traffic violation data are included as a feature to enrich risk information. Finally, a multi-graph deep learning framework (STCM-GCN) with spatial, temporal, and spatial–temporal modules has been developed. Data from Shenzhen, China, demonstrates that the STCM-GCN outperforms baseline models and has a reasonable structure. The inclusion of crash alarm data and traffic violations contributes to performance improvement. Additionally, the model exhibits spatial–temporal robustness, and the analysis of computational efficiency provides comprehensive insights into the model's capabilities.

Large Language Models with Vision on Diagnostic Radiology Board Exam Style Questions

Rationale and ObjectivesThe expansion of large language models to process images offers new avenues for application in radiology. This study aims to assess the multimodal capabilities of contemporary large language models, which allow analysis of image inputs in addition to textual data, on radiology board style examination questions with images. Materials and Methods280 questions were retrospectively selected from the AuntMinnie public test bank. The test questions were converted into three formats of prompts; 1) Multimodal, 2) Image only, and 3) Text only input. Three models, GPT-4V, Gemini 1.5 Pro, and Claude 3.5 Sonnet, were evaluated using these prompts. The Cochran Q test and pairwise McNemar test were used to compare performances between prompt formats and models. ResultsNo difference was found for the performance in terms of % correct answers between the text, image, and multimodal prompt formats for GPT-4V (54%, 52%, and 57%, respectively; p=.31) and Gemini 1.5 Pro (53%, 54%, and 57%, respectively; p=.53). For Claude 3.5 Sonnet, the image input (48%) significantly underperformed compared to the text input (63%, p<.001) and the multimodal input (66%, p<.001), but no difference was found between the text and multimodal inputs (p=.29). Claude significantly outperformed GPT and Gemini in the text and multimodal formats (p<.01). ConclusionVision capable large language models cannot effectively use images to increase performance on radiology board-style examination questions. When using textual data alone, Claude 3.5 Sonnet outperforms GPT-4V and Gemini 1.5 Pro, highlighting the advancements in the field and its potential for the use in further research.

Advances in Artificial Intelligence for automated knee osteoarthritis classification using the IKDC system.

Knee osteoarthritis is one of the most prevalent and debilitating musculoskeletal diseases, with a high incidence among the elderly population. Early detection and accurate classification can improve clinical outcomes for affected patients. This study investigates the use of artificial intelligence (AI) and computer vision for automated detection and classification of knee osteoarthritis using the IKDC classification system. The aim was to develop an automated system for this purpose and evaluate its accuracy in classifying disease severity. A public dataset containing radiographic knee images with varying degrees of osteoarthritis, previously classified according to the IKDC scale, was utilized. Images were processed using LandingLens software, an advanced computer vision platform facilitating AI model development and implementation. A machine learning model based on the ConvNext architecture-a convolutional neural network-was trained on 1901 images and evaluated using 380 test images. The model demonstrated an overall accuracy of 95.16% in classifying knee osteoarthritis according to the IKDC scale, with a sensitivity of 95.11%. Class-specific accuracies were 92.40% for class A, 93.20% for class B, 98.45% for class C, and 95.69% for class D. These results highlight the model's capability to distinguish between different severity grades of osteoarthritis with high accuracy. This study underscores the efficacy of AI and computer vision in automating knee osteoarthritis detection, providing a precise and reliable tool for physicians in disease diagnosis. Integrating these technologies into clinical practice has the potential to enhance efficiency and consistency in patient evaluation, potentially leading to improved clinical outcomes and more personalized medical care. Level III.