Limited Accuracy Research Articles

MODELLING NIGERIAN STOCK RETURNS WITH ARMA-EGARCH: A VOLATILITY ANALYSIS

This study analyzes the volatility of Nigerian stock returns from January 2, 2010, to December 31, 2022, using ARMA-EGARCH models with Generalized Error Distribution (GED). Four model specifications are assessed: ARMA-EGARCH (1,1), ARMA-EGARCH (1,2), ARMA-EGARCH (2,1), and ARMA-EGARCH (2,2). The focus is on model selection, parameter estimation, and diagnostic testing to determine the best model for capturing volatility dynamics. The ARMA-EGARCH (2,2) GED model emerges as the best based on AIC, BIC, and high log-likelihood values, offering a good balance of fit and complexity. The ARMA-EGARCH (1,1) GED model is noted for effectively balancing simplicity and fit while capturing volatility and asymmetric effects. However, all models show limitations in fully capturing volatility dynamics and maintaining parameter stability, particularly concerning volatility clustering. The ARMA-EGARCH (2,2) model consistently performs best across various statistical criteria, including AIC and BIC. Although it provides a robust fit, it has some limitations in serial correlation and model stability. This indicates the need for further model refinement and exploration to enhance forecasting accuracy and address intrinsic limitations. These findings are valuable for investors and policymakers in understanding stock market volatility modelling both in Nigeria and globally.

Extreme Gradient Boosting Model with SMOTE for Heart Disease Classification

Heart disease is one of the leading causes of death worldwide. According to data from the World Health Organisation (WHO), the number of victims who die from heart disease reaches 17.5 million people every year. However, the method of diagnosing heart disease in patients is still not optimal in determining the right treatment. Along with the development of technology, various models of machine learning algorithms and data processing techniques have been developed to find models that can produce the best precision in classifying heart disease. machine learning algorithm model in classifying heart disease, so that it can improve the effectiveness of diagnosis and help in determining the right treatment for patients. This research also aims to overcome the limitations of accuracy in existing diagnosis methods by identifying models that are capable of providing the best results in processing and analysing health data, especially in terms of heart disease classification. In this study, the XGBoost model was identified as the most superior, with an accuracy of 99%. These results show that the XGBoost model has a higher accuracy rate compared to previous methods, making it a promising solution to improve the accuracy of heart disease diagnosis and classification in the future.

Development of New Equation for Predicting State of Normometabolism from Cohort of Hospitalized Patients with Obesity

Background/Objectives: Existing resting energy expenditure (REE) predictive equations, including Mifflin-St Jeor and Harris–Benedict, show limited accuracy, particularly in patients with a BMI over 35, often leading to overestimation or underestimation of REE. This study aimed to develop a new predictive equation specifically designed to identify normometabolic status in patients with obesity, enabling more precise qualitative assessments of basal metabolism through indirect calorimetry. Methods: A cohort of 89 hospitalized patients with obesity (BMI > 30) underwent REE measurement and comprehensive anthropometric assessments. Patients were classified as normometabolic if their REE was within ±10% of the Mifflin-St Jeor prediction or if their fat-free mass-specific REE fell between 23 and 30 kcal/kg. Results: The newly developed equation demonstrated high predictive accuracy (R2 = 0.923, root mean square error = 81.872 kcal/day), with a mean bias of −0.054 kcal/day and narrower limits of agreement (−156.834 to 156.725 kcal/day) compared to widely used models. Conclusions: These advancements could enhance follow-up and management of diet therapy in patients with obesity, allowing for a more tailored approach to their metabolic health over time.

Echocardiographic feature analysis for differentiating myocardial hypertrophy etiologies using artificial intelligence

Abstract Background Left ventricular hypertrophy (LVH) is a common finding in cardiovascular diseases, necessitating precise etiological differentiation for tailored treatment strategies. Echocardiography, crucial for detecting LVH, often exhibits limited accuracy in differentiating LVH etiology. Artificial Intelligence (AI) holds promise in augmenting echocardiographic feature analysis, thereby enhancing LVH assessment. Objectives We aimed to develop an explainable AI algorithm detecting LVH and differentiating its common etiologies, such as hypertrophic cardiomyopathy (HCM), cardiac amyloidosis (CA), or hypertensive heart disease (HHD), based on echocardiographic images. Methods The developmental datasets were sourced from diverse medical centers (867 subjects), while an independent external validation set was obtained from a single tertiary medical center (619 subjects). Utilizing the four fundamental echocardiographic views, conventional and harmonization-driven myocardial textures and geographic features were extracted, and the explainable classification algorithm was developed. (Figure 1) Results The developmental dataset (mean age, 52.0±17.9 years; 60.3% male) comprised with LVH present in 67.6% of the training set and 63.2% of the internal validation set (P=0.323). The external validation set (mean age, 51.3±14.6; 48.6% male) showed a significantly lower LVH prevalence than the internal validation dataset (33.1% vs. 63.2%; P <0.001). In internal validation, the classification model accurately detected LVH with an area under the curve (AUC) of 1.00 (95% confidence interval [CI], 1.00–1.00). For HCM, CA, and HHD classification, it achieved AUCs of 0.97 (95% CI, 0.94–0.99), 0.95 (95% CI, 0.90–0.99), and 0.86 (95% CI, 0.78–0.93), respectively. In external validation, the model maintained impressive performance: LVH, 1.00 (95% CI, 0.99–1.00); HCM, 0.96 (95% CI, 0.92–0.98); CA, 0.89 (95% CI, 0.83–0.93); and HHD, 0.86 (95% CI, 0.81–0.91). (Figure 2) Overall, the AI-enhanced echocardiographic feature analysis successfully differentiated the causes of LVH, demonstrating high accuracy in internal validation and external validation cohorts (89.6 and 92.7%, respectively). Harmonization-driven textures were pivotal in LVH detection and HCM differentiation, while conventional textures and myocardial thickness were influential in CA and HHD differentiation. Conclusion This study showed that AI-enhanced echocardiographic feature analysis accurately identified LVH and its etiologies, highlighting the value of AI-driven texture and spatial analysis in LVH evaluation. Process of Echo feature-extraction AUC of Model in LVH differentiation

Clinical and Pathological Staging Discrepancies in Laryngeal Cancer: A Systematic Review

Background/Objectives: Laryngeal squamous cell carcinoma (LSCC) is one of the most prevalent and challenging malignancies of the head and neck. Clinical staging (cTNM) plays a pivotal role in therapeutic decision-making. However, current imaging modalities often fall short, resulting in discrepancies between cTNM and pathological staging (pTNM). This systematic review aimed to critically evaluate the existing literature on the concordance between clinical and pathological staging of LSCC, quantifying staging inaccuracies and highlighting the prevalence of both under- and overstaging at diagnosis. Methods: A comprehensive search of the English-language literature was conducted across multiple databases, including PubMed, Embase, Scopus, the Cochrane Library, and Web of Science. Eligibility was limited to retrospective case series and observational studies reporting sufficient data to directly correlate individual patients’ cTNM and pTNM classifications. Results: Thirty-one studies comprising 7939 patients met the inclusion criteria. The overall concordance rate between cT and pT was approximately 86.43%. The concordance rates between cT and pT were 82.41%, 82.03%, 78.14%, and 89.64% for cT1, cT2, cT3, and cT4, respectively. Most discordant cases in cT2 and cT3 involved understaging at clinical diagnosis. Conclusions: The limited accuracy of clinical staging in reflecting the true extent of disease remains a critical challenge in the management of LSCC. The inability of current imaging techniques to reliably detect the subtle invasion of key anatomical structures contributes to both under- and overstaging, with significant clinical implications. For patients undergoing non-surgical organ-preservation strategies, these inaccuracies may adversely affect oncologic outcomes.

A Coarse- and Fine-Grained Co-Exploration Approach for Optimizing DNN Spatial Accelerators: Improving Speed and Performance

The rapid advancement of deep neural networks has significantly increased demands for computational complexity and data volume. This trend is especially evident with the emergence of large language models, which have rendered traditional architectures such as CPUs and GPGPUs insufficient in meeting performance and energy efficiency requirements. Spatial accelerators present a promising solution by optimizing on-chip compute, storage, and communication resources. In exploring spatial accelerator design spaces, analytical model-based simulators and cycle-accurate simulators are commonly employed, each offering distinct advantages: high computational speed and superior simulation accuracy, respectively. However, the limited accuracy of analytical models and the slow simulation speed of cycle-accurate simulators impede the achievement of globally optimal solutions during design space exploration. Therefore, effectively leveraging the strengths of both simulator types while mitigating their inherent trade-offs is a critical challenge in designing customized spatial accelerators. In this work, we introduce a novel co-exploration methodology that integrates both coarse-grained and fine-grained approaches to navigate design and mapping spaces effectively. We utilize the rapid simulation capabilities of analytical models to perform coarse-grained global exploration, quickly eliminating designs and mapping configurations with inferior performance. Building on the results of this initial exploration, we employ cycle-accurate simulators to conduct fine-grained local exploration within the identified promising regions of the design and mapping spaces. This dual-phase approach aims to identify optimal hardware designs and dataflow mapping strategies that enhance performance and energy efficiency. The experimental results demonstrate that, compared to state-of-the-art methods, our approach reduces the number of exploration points by up to 99%, while achieving a 17.9% reduction in latency, a 2.5% decrease in energy consumption, and a 30.3% improvement in throughput.

A New Hybrid Improved Kepler Optimization Algorithm Based on Multi-Strategy Fusion and Its Applications

The Kepler optimization algorithm (KOA) is a metaheuristic algorithm based on Kepler’s laws of planetary motion and has demonstrated outstanding performance in multiple test sets and for various optimization issues. However, the KOA is hampered by the limitations of insufficient convergence accuracy, weak global search ability, and slow convergence speed. To address these deficiencies, this paper presents a multi-strategy fusion Kepler optimization algorithm (MKOA). Firstly, the algorithm initializes the population using Good Point Set, enhancing population diversity. Secondly, Dynamic Opposition-Based Learning is applied for population individuals to further improve its global exploration effectiveness. Furthermore, we introduce the Normal Cloud Model to perturb the best solution, improving its convergence rate and accuracy. Finally, a new position-update strategy is introduced to balance local and global search, helping KOA escape local optima. To test the performance of the MKOA, we uses the CEC2017 and CEC2019 test suites for testing. The data indicate that the MKOA has more advantages than other algorithms in terms of practicality and effectiveness. Aiming at the engineering issue, this study selected three classic engineering cases. The results reveal that the MKOA demonstrates strong applicability in engineering practice.

“Omics” and Postmortem Interval Estimation: A Systematic Review

Postmortem interval (PMI) estimation is a challenge of utmost importance in forensic daily practice. Traditional methods face limitations in accuracy and reliability, particularly for advanced decomposition stages. Recent advances in “omics” sciences, providing a holistic view of postmortem biochemical changes, offer promising avenues for overcoming these challenges. This systematic review aims at investigating the role of mass-spectrometry-based “omics” approaches in PMI estimation to elucidate molecular mechanisms underlying predictable time-dependent biochemical alterations occurring after death. A systematic search was performed, adhering to PRISMA guidelines, through “free-text” protocols in the databases PubMed, SCOPUS and Web of Science. The inclusion criteria were as follows: experimental studies analyzing, as investigated samples, animal or human corpses in toto or in parts and estimating PMI through MS-based untargeted omics approaches, with full texts in the English language. Quality assessment was performed using STROBE and ARRIVE critical appraisal checklists. A total of 1152 papers were screened and 26 included. Seventeen papers adopted a proteomic approach (65.4%), nine focused on metabolomics (34.6%) and two on lipidomics (7.7%). Most papers (57.7%) focused on short PMIs (<7 days), the remaining papers explored medium (7–120 days) (30.77%) and long PMIs (>120 days) (15.4%). Muscle tissue was the most frequently analyzed substrate (34.6% of papers), followed by liver (19.2%), bones (15.4%), cardiac blood and leaking fluids (11.5%), lung, kidney and serum (7.7%), and spleen, vitreous humor and heart (3.8%). Predictable time-dependent degradation patterns of macromolecules in different biological substrates have been discussed, with special attention to molecular insights into postmortem biochemical changes.

A Novel Virtual Planned-Orthodontic-Surgical Approach for Proportional Condylectomy in Condylar Hyperplasia

Background/Objectives: Condylectomy is a delicate and intricate procedure commonly employed in the management of temporomandibular joint (TMJ) disorders, osteochondromas, condylar hyperplasia, hemimandibular hyperplasia, and other pathologies affecting the condylar region. The advent of surgical cutting guides has introduced a new dimension to condylectomy procedures as they enable surgeons to plan and execute precise cuts with a heightened level of accuracy. In the literature already exists cases of cutting guide-based condylectomy, but they only depend on the mere mirroring procedure in virtual planning, which has accuracy limitations because it does not consider asymmetry of peri-condylar structures at the level of the ramus, body, and mandibular angle. Methods: CAD-CAM orthodontic preparation through the NEMOFAB Software was performed to correct the canting of the occlusal plane, following the “orthodontic first” technique. The same software was used for VSP of the surgical cutting guide to perform the condylectomy, basing not to the mere mirroring of the opposite side but considering the whole condylar-TMJ-glenoid fossa structure. Results: At 6 months follow-up, the patient showed good occlusion and an almost totally recovered lower third symmetry as median-upper and lower interincisive lines coincide with each other and with the chin median. A good occlusal and masticatory outcome was obtained. The joint structure was preserved with remodeling of the glenoid cavity caused by the presence of the joint disc, which was preserved during surgery. Conclusions: The goal of this study is to propose a method of therapeutic management of condylar hyperplasia that benefits from accurate pre-operative orthodontic treatment (orthodontics first) to maximize the results of proportional condylectomy, reducing post-operative orthodontic care as well as any need for any adjuvant orthognathic surgery. A new virtual surgical planning method is also proposed for creating a cutting guide that not only takes advantage of the mirroring technique to accurately calculate the amount of condyle to be cut but also considers the entire condyle–TMJ complex to perform a condylectomy that is more precise.

Cross-Point Ferroelectric Hf0.5Zr0.5O2 Capacitors for Remanent Polarization-Driven In-Memory Computing.

Ferroelectric Hf0.5Zr0.5O2 (HZO) capacitors have been extensively explored for in-memory computing (IMC) applications due to their nonvolatility and back-end-of-line (BEOL) compatible process. Several IMC approaches using resistance and capacitance states in ferroelectric HZO have been proposed for vector-matrix multiplication (VMM), but previous approaches suffer from limited accuracy and reliability. In this work, we propose a promising approach centered on the remanent polarization (Pr) switching of binary ferroelectric HZO capacitor synapses. We experimentally demonstrate a simple pattern recognition task showing that the voltage readout of Pr switching provides excellent accuracy due to its high on/off ratio and consequent reliability. We also performed large-scale simulations on a complex inference task, achieving high accuracy and immunity to device variations. We therefore believe that our proposed paradigm is promising for near-term neuromorphic IMC.

A digital twin solution for fault detection in time-critical IIoT applications

ABSTRACT IIoT sensor data plays a pivotal role in monitoring the industrial system’s health and identifying potential faults. However, traditional fault detection approaches often face challenges such as network latency, limited accuracy, and resource-intensive processing. This paper introduces an end-to-end Digital Twin solution that enhances fault detection for IIoT systems. The solution is powered by two key innovations: the integration of a Digital Twin architecture that leverages a collaborative cloud-edge approach for real-time monitoring, and the use of a lightweight two-phased machine-learning ensemble model optimized for resource-constrained environments. The great performance achieved across various fault scenarios demonstrates the effectiveness of the proposed approach. The model provides an average accuracy of 99.71% with a mere 4.8 ms of average estimation delay. These advancements ensure both high accuracy and rapid response times, providing a robust solution for proactive fault detection in dynamic industrial environments.

Enhanced Multi-Model Deep Learning for Rapid and Precise Diagnosis of Pulmonary Diseases Using Chest X-Ray Imaging

Background: The global burden of respiratory diseases such as influenza, tuberculosis, and viral pneumonia necessitates rapid, accurate diagnostic tools to improve healthcare responses. Current methods, including RT-PCR and chest radiography, face limitations in accuracy, speed, accessibility, and cost-effectiveness, especially in resource-constrained settings, often delaying treatment and increasing transmission. Methods: This study introduces an Enhanced Multi-Model Deep Learning (EMDL) approach to address these challenges. EMDL integrates an ensemble of five pre-trained deep learning models (VGG-16, VGG-19, ResNet, AlexNet, and GoogleNet) with advanced image preprocessing (histogram equalization and contrast enhancement) and a novel multi-stage feature selection and optimization pipeline (PCA, SelectKBest, Binary Particle Swarm Optimization (BPSO), and Binary Grey Wolf Optimization (BGWO)). Results: Evaluated on two independent chest X-ray datasets, EMDL achieved high accuracy in the multiclass classification of influenza, pneumonia, and tuberculosis. The combined image enhancement and feature optimization strategies significantly improved diagnostic precision and model robustness. Conclusions: The EMDL framework provides a scalable and efficient solution for accurate and accessible pulmonary disease diagnosis, potentially improving treatment efficacy and patient outcomes, particularly in resource-limited settings.

A Bayesian-optimized 1D CNN-based outlier detection approach for wireless sensor networks

Wireless sensor networks (WSNs) have recently emerged as a critical technology in various applications, including industrial automation, building monitoring, and military. However, the data generated by these networks are often prone to outliers, which can compromise sensor data quality and reliability. Detecting outliers is paramount to ensure proper network functioning. Traditional detection techniques pose several challenges, such as weak adaptability to the increasing complexity and dynamic environmental changes, limited accuracy, and higher computation costs. To address these challenges, this paper proposes an optimized one-dimensional convolutional neural networks (1D CNN)-based outlier detection approach for WSNs. This approach comprises two key modules: a predictor module and an outlier detector. The predictor module employs a 1D CNN model to forecast forthcoming sensor measurements based on historical data. Bayesian optimization is used to enhance the 1D CNN model’s accuracy. The outlier detector identifies outliers based on the Euclidean distance between the predicted measurements and their corresponding actual values. Experiments on synthetic and real-world datasets reveal that our proposed approach outperforms other existing deep learning-based frameworks in terms of accuracy, F1 score, and false alarm rates.

DenseNet-Enabled Intelligent Ship Energy Consumption Prediction Method with Navigation Information and Environmental Factors

As a crucial procedure, ship energy consumption prediction is of great significance for the implementation of navigation planning, energy efficiency improvement, and emission reduction strategies. Currently, various studies have been conducted, primarily focusing on container ships, cargo ships, bulk carriers or tankers, with relatively fewer studies dedicated to research vessels. Regarding the limitations of low accuracy and poor effectiveness in ship energy consumption prediction, caused by the diverse voyages and specialized operational missions of research vessels, this paper proposes a DenseNet-enabled energy consumption prediction model considering the complex navigation information and diverse environmental factors. By leveraging the high parameter efficiency of DenseNet, effective feature extraction and strong generalization capabilities can be achieved, which provide more reliable and accurate predictions. Firstly, due to the fact that the real-time fuel consumption is influenced by a variety of internal and external factors, multi-source monitoring data are obtained and analyzed. After data analysis and feature extraction, the processed data and features are utilized to establish the prediction model. Through validations on the actual voyage data, it is demonstrated that the DenseNet-based prediction model outperforms the ResNet, DBCNN, and FCNN models in terms of both accuracy and predictive reliability. Therefore, the proposed method is capable of accurately predicting ship energy consumption under diverse shipping conditions and provides valuable guidance for the development of intelligent shipping.

Improved YOLOv5: Efficient Object Detection for Fire Images

Camera-based fire detection systems have been found to yield significantly superior results compared to their sensor-based counterparts, and are thus widely employed for this purpose across the globe. This study presents an effective and lightweight fire detection technique based on deep learning. A novel convolutional neural network has been created specifically for the purpose of identifying fire areas. This network utilizes the YOLOv5 algorithm. The default versions of the You Only Look Once (YOLO) technique demonstrate limited accuracy when applied to fire detection instances, even after undergoing training and testing. After training and testing in fire detection instances, the default versions of the YOLO technique exhibit a significantly low level of accuracy. We chose the YOLOv5 network to augment its capabilities and maximize its effectiveness in identifying fire disasters. The self-designed convolution module, also known as the improved convolution module, is implemented to substitute the regular convolution module in YOLOv5. By dividing the input features into two main portions, it helps to realize the lightweight of the network. Additionally, the network incorporates the attention mechanism. The testing findings clearly show that the proposed method is more effective than previous ways of identifying fire, with a precision of 91.22% and a recall of 93.78%.

A Novel Neural Network-Based Approach Comparable to High-Precision Finite Difference Methods

Deep learning methods using neural networks for solving partial differential equations (PDEs) have emerged as a new paradigm. However, many of these methods approximate solutions by optimizing loss functions, often encountering convergence issues and accuracy limitations. In this paper, we propose a novel deep learning approach that leverages the expressive power of neural networks to generate basis functions. These basis functions are then used to create trial solutions, which are optimized using the least-squares method to solve for coefficients in a system of linear equations. This method integrates the strengths of streaming PINNs and the traditional least-squares method, offering both flexibility and a high accuracy. We conducted numerical experiments to compare our method with the results of high-order finite difference schemes and several commonly used neural network methods (PINNs, lbPINNs, ELMs, and PIELMs). Thanks to the mesh-less feature of the neural network, it is particularly effective for complex geometries. The numerical results demonstrate that our method significantly enhances the accuracy of deep learning in solving PDEs, achieving error levels comparable to high-accuracy finite difference methods.

Assessment of greenness, blueness, and whiteness profiles of a validated HPLC-DAD method for quantitation of Donepezil HCl and Curcumin in their laboratory prepared co-formulated nanoliposomes

White and Green Analytical Chemistry are innovative approaches in analytical chemistry that prioritize both sustainability and efficiency. Together, these approaches aim to advance scientific research while minimizing environmental impact and enhancing safety. This integration of environmental consciousness into analytical practices represents a significant step forward in achieving sustainable scientific progress. In the present study, a sensitive eco-friendly HPLC-DAD method was carried out and validated to allow concurrent determination of Donepezil HCl (DPZ) and Curcumin (CUR) in their pure form and laboratory made nano-liposome formulation. Optimum seperation was accomplished by utilising Zobrax Eclipse Plus C18 column (4.6*100 mm,5 μm) with gradient elution of the mobile phase composed of 0.02 M phosphate buffer at pH 3.2 and ethanol at flow rate of 1.5 ml/min. A diode array detector (DAD) was implemented for detection at 273 nm and 435 nm for DPZ and CUR, respectively, with the column oven set at 40 °C. The method was validated according to ICH specifications in terms of accuracy, precision, linearity range, detection and quantification limit. The calibration plots were linear with correlation coefficients (r2) = 0.999 over the range (0.1–100 µg/ml) and (0.1–100 µg/ml) for DPZ and CUR, successively. The validated HPLC-DAD approach was adopted to analyse both medications in laboratory prepared nano-liposomal formulation in which the analytes were successfully quantified with good recovery values and no disrubtion from the added excipients. The investigation of whiteness, blueness, and greenness metrics revealed a major benefit of the suggested approach over previous reported ones.

Automated 3D semantic segmentation of PCB X-ray CT images and netlist extraction

Printed Circuit Board (PCB) design reconstruction is essential for addressing part obsolescence, intellectual property recovery, compliance, quality assurance, and enhancing national capabilities. Traditional methods for PCB design extraction, both non-geometry-based and geometry-based, have limitations in accuracy, efficiency, and scalability. This paper presents an automated approach, combining image processing and machine learning, to achieve 3D semantic segmentation of PCB X-ray Computed Tomography (X-ray CT) images and subsequent netlist extraction. By employing a 3D U-Net architecture with a ResNet-18 backbone and training on synthetic data, we introduce a first-of-its-kind method for direct 3D semantic segmentation, significantly improving over previous efforts. Our approach eliminates the need for extensive labeled datasets by using inherently labeled synthetic data. Further, this method enhances ease of segmentation by significantly reducing or eliminating the preprocessing effort required for 2D image stacks. It also improves universality by expanding the scope of application beyond images with specific 2D stack criteria, segmenting the 3D image in its entirety. Additionally, this method enables the processing of images of PCBs that have undergone bending, which is common among PCBs with a thickness below a certain threshold. The implications of this approach extend beyond PCBs, finding applications in various physical and biological sciences where 3D image segmentation is crucial. This methodology includes high-resolution 3D imaging, watershed segmentation, machine learning-based semantic segmentation, and netlist extraction. Validation with both synthetic and real-world PCB datasets shows high accuracy and robustness, offering a scalable solution for PCB design reconstruction.

Evaluation of Advanced Artificial Intelligence Algorithms' Diagnostic Efficacy in Acute Ischemic Stroke: A Comparative Analysis of ChatGPT-4o and Claude 3.5 Sonnet Models.

Background/Objectives: Acute ischemic stroke (AIS) is a leading cause of mortality and disability worldwide, with early and accurate diagnosis being critical for timely intervention and improved patient outcomes. This retrospective study aimed to assess the diagnostic performance of two advanced artificial intelligence (AI) models, Chat Generative Pre-trained Transformer (ChatGPT-4o) and Claude 3.5 Sonnet, in identifying AIS from diffusion-weighted imaging (DWI). Methods: The DWI images of a total of 110 cases (AIS group: n = 55, healthy controls: n = 55) were provided to the AI models via standardized prompts. The models' responses were compared to radiologists' gold-standard evaluations, and performance metrics such as sensitivity, specificity, and diagnostic accuracy were calculated. Results: Both models exhibited a high sensitivity for AIS detection (ChatGPT-4o: 100%, Claude 3.5 Sonnet: 94.5%). However, ChatGPT-4o demonstrated a significantly lower specificity (3.6%) compared to Claude 3.5 Sonnet (74.5%). The agreement with radiologists was poor for ChatGPT-4o (κ = 0.036; %95 CI: -0.013, 0.085) but good for Claude 3.5 Sonnet (κ = 0.691; %95 CI: 0.558, 0.824). In terms of the AIS hemispheric localization accuracy, Claude 3.5 Sonnet (67.2%) outperformed ChatGPT-4o (32.7%). Similarly, for specific AIS localization, Claude 3.5 Sonnet (30.9%) showed greater accuracy than ChatGPT-4o (7.3%), with these differences being statistically significant (p < 0.05). Conclusions: This study highlights the superior diagnostic performance of Claude 3.5 Sonnet compared to ChatGPT-4o in identifying AIS from DWI. Despite its advantages, both models demonstrated notable limitations in accuracy, emphasizing the need for further development before achieving full clinical applicability. These findings underline the potential of AI tools in radiological diagnostics while acknowledging their current limitations.

Prediction of Spalling and Roughness in the Mechanistic-Empirical Pavement Design Guide

This paper documents an assessment of the prediction of joint spalling and International Roughness Index (IRI) for jointed plain concrete pavements (JPCP) in the Mechanistic- Empirical Design Guide (MEPDG) report and software. The role of joint sealant type in the prediction of spalling is examined, as well as the sensitivity of predicted IRI to predicted spalling and other factors. Limitations of the predictive accuracy of the MEDG spalling and IRI models are detailed. Users of the MEPDG and its accompanying software should be aware of the inference space of the data from which these spalling and IRI models were developed, and the practical implications of using the MEPDG models to predict the effect of joint sealant type, joint spalling, climate, and other factors on roughness in jointed plain concrete pavements. The MEPDG software predicts essentially no spalling for joints with preformed sealant, but substantially more spalling – and exactly the same amount of spalling – for unsealed joints, hotpour- sealed joints, and silicone-sealed joints.