Accurate segmentation of zones in prostate MRI is crucial for effective prostate cancer diagnosis and planning treatments. Computer-aided segmentation is important to overcome the complexity of subtle textural contrasts and similar signal intensities found between adjacent prostate zones, which include the Peripheral Zone (PZ) and Transition Zone (TZ), making it difficult to delineate the boundary. In this paper, we present PZSNet, which is a U-shaped hybrid transformer designed for accurate prostate zone segmentation. PZSNet leverages the multiscale feature learning and boundary-aware decoding to improve segmentation performance and to guarantee an accurate border delineation among various complicated prostate zones. The model introduces three key modules: (1) Adaptive Boundary Fusion Module (ABFM) which captures spatial dependencies and recalibrates channel wise importance to handle subtle textural differences (2) Dynamic Features Injection Module (DFIM) that enhances the multiscale context for better segmentation of zones with close signal intensities like AFS and TZ while preserving spatial resolution, and (3) Feature Enhancement Module (FEM), which adaptively enhances low level and high level features to ensure effective integration of fine details and broader context. Evaluated on the ProstateX and MSD datasets, PZSNet obtains the DSC of 96.97

Driving requires complex cognitive abilities, making it a promising behavioral domain for identifying Mild Cognitive Impairment (MCI). This paper presents a pilot proof-of-concept framework deploying AutoPi telematics units across 51 older adult drivers (10 MCI, 41 cognitively unimpaired) over a 28-month observation window, yielding 20,145 trips across GPS, IMU, and OBD-II sensor streams. A multi-stage analytical pipeline, K-Means clustering for behavioral profiling, Random Forest feature ranking, Welch's t -tests with Benjamini-Hochberg correction, and L1-regularized logistic regression with participant-level leave-one-out cross-validation, achieves an AUC of 0.698 (95% CI: 0.493-0.872) with a sensitivity of 0.800. Throttle position variability and mean throttle application are the strongest sensor-derived predictors (Cohen's d = 0.86 each), reflecting impaired speed regulation consistent with executive dysfunction in MCI; however, the cohort's gender imbalance (9 of 10 MCI participants are female) means that demographic factors, particularly gender, contribute substantially to overall model discrimination. A sensitivity analysis excluding gender reduces AUC to 0.598, comparable to the telematics-only result (0.595), confirming that the driving-behavior signal is meaningful but modest when demographic confounding is removed. Cold-start analysis indicates that approximately 50 trips, roughly four months of naturalistic driving, constitutes the minimum viable observation window for reliable screening. Subgroup analyses reveal performance disparities attributable to cohort composition rather than systematic model bias. Findings support telematics-based MCI monitoring as a promising framework warranting validation in larger, gender-balanced cohorts before clinical deployment.

This study provides a bibliometric overview of the first decade of the Journal of Big Data (JBD) since its launch in 2014. Using bibliographic records for 1,023 articles and reviews indexed in Scopus (2014–2024) and 917 records in Web of Science, the analysis combines performance indicators with science-mapping techniques, including co-citation, bibliographic coupling, and keyword co-occurrence implemented in VOSviewer and bibliometrix. The results document rapid growth in output, a Scopus/SCImago h-index of 91 by 2024, and strong international contributions led by the USA, China, India, and European countries. Citation and topical structures reveal three main thematic cores centred on big data infrastructures (e.g., Hadoop, MapReduce, Apache Spark), machine and deep learning (e.g., convolutional and recurrent neural networks), and application domains such as cybersecurity, healthcare analytics, and sentiment analysis. The findings characterize JBD’s position within the wider data science ecosystem, showing how its publications concentrate influence in a small number of highly cited surveys and frameworks while supporting an increasingly diverse long tail of topics and emerging research fronts.

The first written mention of computing based on mapping of topological structures and mathematical algorithms onto graphs to be built into hardware, for graph-based computing, dates back to January 1984, when one of the authors of this text, as a Purdue University faculty member, submitted a related proposal to the Defense Department of RCA, Camden, New Jersey, USA. A preliminary study concluded that the optimal structure for mapping algorithms onto hardware is not square, but of the hexagonal shape referred to as “honeycomb”. That is how the term honeycomb was created in the context of mapping structures and algorithms onto hardware. According to the open literature the first structures ever mapped onto the honeycomb architecture are interconnection networks and the first algorithms ever mapped onto the honeycomb architecture are neural networks. If a structure or an algorithm is mapped onto a graph to be built into the hardware, conditions are met for data to flow from inputs to outputs, driven by the voltage difference between inputs and outputs. If that goal is accomplished, potentials are generated for data to move faster, which means that the algorithms get executed with a speed-up. If the data movement is accomplished with a slower clock, less energy is consumed during data propagation through structures and the execution of an algorithm. Consequently, since no control mechanisms have to be implemented, the volume of the engine becomes smaller. And finally, since the structure of the data path is formed at the time of mapping the graph into hardware, the width of the data path could be made as wide as needed, and no wider than needed, which leads to a lot better precision, at no cost in the domain of execution time. Of course, the described paradigm shift became utilizable in practice only after the reconfigurable hardware was invented. That is why the first data flow machines, based on the same or similar concept, appeared only after the FPGA circuits became widespread. However, the full utilization of this paradigm will be possible only after the Analog See of Gates circuits become feasible, which enables a lot lower operational frequency (a lot lower power consumption) and a lot higher speed (due to fewer obstacles on the data propagation paths). This article sheds light on the complexity and speed related to the mapping of structures and algorithms.

With a substantial increase in digital information on various platform browsing applications, recommendation systems are imperative to filter large amounts of information in more palatable ways through the organization of the most relevant topics. This can be applied to various domains, including e-commerce, education, books, movies, music, and more. As such, it is crucial to have a thorough understanding of the various generations of recommendation systems. In this study, we conduct a comprehensive examination of generations of recommendation systems using various applicable techniques. To examine the objective of recommendation systems, we introduce challenges to recommendation systems. The generations are then broken down into sub-classes for further examination. These are content-based, collaborative filtering, hybrid models, matrix factorization, web usage mining, personality-based models, collaborative filtering using deep learning techniques, deep content-based models, and combined modeling of users and items using reviews. First-generation techniques often employ rudimentary approaches, whereas second- and third-generation techniques utilize more complex models that delve deeper into the vast amount of available data. Based on this analysis, we believe that our survey enhances the understanding of generational recommendation system models and highlights the importance of selecting suitable techniques to design novel, cross-disciplinary models.

Deep neural networks drive modern machine vision but are challenging to deploy on edge devices because of their high compute demands. Traditional approaches-running the full model on device or offloading to the cloud-face tradeoffs in latency, bandwidth, and privacy. Splitting the inference workload between the edge and the cloud offers a balanced solution, but transmitting intermediate features to enable such splitting introduces new bandwidth challenges. To address this, MPEG initiated the Feature Coding for Machines (FCM) standard, establishing a bitstream syntax and codec pipeline tailored for compressing intermediate features. This article presents the design and performance of the Feature Coding Test Model, showing significant bit-rate reductions-averaging 85.14%-across multiple vision tasks while preserving accuracy. FCM offers a scalable path for efficient and interoperable deployment of intelligent features in bandwidth-limited and privacy-sensitive consumer applications.

Artificial intelligence is rapidly transforming scientific research and engineering practice. Modern AI systems can analyze technical literature, summarize complex concepts, generate alternative design ideas, improve technical writing, and accelerate many aspects of research and development. These capabilities naturally raise important questions regarding inventorship, patent validity, confidentiality, and documentation of AI-assisted inventions. Based on my experience as an inventor and author of more than one hundred patents and patent applications, I believe AI can become an extremely valuable productivity tool for inventors—provided that it is used appropriately. The purpose of this editorial is not to discuss legal doctrine, but rather to share practical recommendations for integrating AI into the inventive process while preserving genuine human inventorship.

Background: Research to identify changes in driving behavior that occur with the onset of Pre-MCI and MCI is an emerging area with many gaps still to be addressed. These gaps include limited use of objective, continuous measurement of driver behavior in real-life traffic conditions and comprehensive, biomarker-validated, cognitive evaluation based upon both testing and clinical ratings. Using these strategies, the questions addressed in this exploratory study are whether or not differences in driving behavior are indicative of Pre-MCI/MCI and which behaviors are most predictive of Pre-MCI/MCI. Methods: As part of a naturalistic longitudinal study, older drivers with a Montreal Cognitive Assessment score ≥ 19 had telematic sensors installed in their vehicles and underwent comprehensive cognitive assessment quarterly for three years. Thirty-six participants were classified as Unimpaired (n = 23) or Pre-MCI/MCI (n = 10/3) based upon a neuropsychological battery and diagnostic algorithm. A penalized generalized linear mixed-effects model (GLMM) with a logistic link and LASSO regularization was used to model Pre-MCI/MCI group membership vs. unimpaired as a function of ten trip-level telematic features (trip distance, hard acceleration, hard braking, hard turns, speed average, maximum speed, RPM average, fuel level, throttle average, and throttle variability) at the end of their first 12 months in the study. Results: Higher RPM, shorter average trips, and greater throttle variability predicted higher odds of Pre-MCI/MCI, while more frequent hard braking, hard turns, higher mean speed, and lower average throttle (steadier pedal control) predicted lower odds of Pre-MCI/MCI. Conclusions: The model clearly distinguished unimpaired older drivers from those with MCI or Pre-MCI, suggesting that distinct patterns of driver behavior may be related to levels of cognitive function.

Given GPS points on a transportation network, the goal of the Quad-tree Based Driver Classification (QBDC) problem is to identify whether drivers have Mild Cognitive Impairment (MCI). The QBDC problem is challenging due to the large volume and complexity of the data. This paper proposes a quad-tree based approach to the QBDC problem by analyzing driving patterns using a real-world dataset. We propose a geo-regional quad-tree structure to capture the spatial hierarchy of driving trajectories and introduce new driving features representation for input into a convolutional neural network (CNN) for driver classification. The experimental results demonstrate the effectiveness of the proposed algorithm, achieving an F1 score of 95% that significantly outperforms the baseline models. These results highlight the potential of geo-regional quad-tree structures to extract interpretable features and describe complex driving patterns. This approach offers significant implications for driver classification, with the potential to improve road safety and cognitive health monitoring.

As numerous edge devices start implementing intelligent components, the challenges of energy consumption, bandwidth efficiency, and privacy gain significance. One proposed solution relies on the paradigm of split inference, which optimizes the delegation of the computational load between edge and remote devices. We developed and implemented the standard-compliant split inference system with an encoder and decoder capable of real-time streaming and processing. Our system outperforms state-of-the-art video compression implementations by an average of 83% bitrate reduction, while preserving privacy. We demonstrate the system's real-time performance on consumer devices, with interactive visualizations of object detection and segmentation, incorporating real-time metrics. Demo video: https://youtu.be/bmCbUo_ZWWU

Background/Objectives: Accurate patient weight estimation is critical for safe and effective drug dosing in emergency and critical care settings. Inaccurate estimates exceeding a 10% deviation from true weight can result in significant dosing errors in time-sensitive treatments such as thrombolysis for stroke or urgent sedation. In situations where direct weight measurement is impractical, reliable alternative estimation methods are essential. Methods: We propose a three-dimensional (3D) depth-camera system that employs a convolutional neural network (CNN) pipeline to automatically estimate total body weight (TBW), ideal body weight (IBW), and lean body weight (LBW) from volumetric features derived from a single supine patient image. Our approach was evaluated in a prospective pilot study to assess feasibility and accuracy. CNNs were selected because of their ability to extract spatial features from complex image data, outperforming regression and tree-based models in preliminary comparisons. Results: The results demonstrated that our 3D camera system was more accurate than conventional techniques, including clinician visual estimation (Mean Absolute Percentage Error [MAPE]: 12%), tape-based methods (±8.5%), and anthropometric formulas (±9.2%), achieving a mean error of ±5.4%. Conclusions: Future work will extend this technology to pediatric populations, support integration with automated dosing systems, and explore prehospital applications to further reduce medication errors and enhance patient safety.

As consumer devices become increasingly intelligent and interconnected, efficient data transfer solutions for machine tasks have become essential. This paper presents an overview of the latest Feature Coding for Machines (FCM) standard, part of MPEG-AI and developed by the Moving Picture Experts Group (MPEG). FCM supports AI-driven applications by enabling the efficient extraction, compression, and transmission of intermediate neural network features. By offloading computationally intensive operations to base servers with high computing resources, FCM allows low-powered devices to leverage large deep learning models. Experimental results indicate that the FCM standard maintains the same level of accuracy while reducing bitrate requirements by 75.90

This paper presents a content-adaptive feature layer filtering method for intermediate feature compression in split inference systems using multi-scale neural networks. The proposed encoder-side optimization removes redundant feature layers based on object size information derived from the input image. Early layers, which contain high spatial resolution within feature maps are suited for detecting small objects. These early layers are then pruned when large objects dominate the scene and their contribution becomes negligible. This reduces redundancy and improves compression efficiency. The method requires no retraining of the task network and remains compatible with conventional codecs by spatially packing the retained features. Aligned with the MPEG Feature Coding for Machines (FCM) framework, this approach enables more efficient collaborative intelligence by reducing bandwidth during intermediate feature transmission. Experimental results on object detection and segmentation tasks show up to a 43% bitrate reduction without compromising task accuracy.

Dataflow computing has proved to be more efficient for certain high-performance computing algorithms. The prerequisites are that there is enough parallel calculation to cover the overhead of executing instructions on the dataflow hardware until the first result is ready. This is often true with algorithms that work with big data and that can process multiple iterations independently, e.g., while simulating certain phenomena in many elementary volumes. However, dataflow hardware runs typically at an order of magnitude lower frequencies compared to the control-flow processor. From a programmer’s point of view, programming dataflow architectures is considerably harder than programming control-flow architectures. As a result, it is not always obvious whether programming dataflow architectures for certain algorithms is worth the effort needed. Therefore, there is a need for a programmer to be able to predict the outcome of programming for dataflow architectures in terms of accelerating program execution and power savings. This article presents a newly developed tool that a programmer can use for profiling control-flow algorithms and estimating the acceleration possibilities using the dataflow hardware.

Social media is proving to be a game changer in the recruitment of older adults for research. Given the success of our initial Facebook outreach to older adults in South Florida, we expanded it to include older Spanish-speaking adults. Neither of these initial campaigns yielded Spanish-speaking participants. In our second campaign thus far, follow-up telephone outreach yielded successful contact with 11 of 39 (28%) leads. Of those 11, 2 were excluded due to distance from the testing sites, 3 did not want further information, and 6 declined after receiving more information. Many reported difficulty fully understanding the content and intent of the posted Facebook ads. Our more traditional recruitment efforts utilized personal referrals and health fairs to foster a critical personal connection, which appeared to be lacking on the social media platform. This disconnect highlighted the need to provide a more complete description of the participant’s role in the study and achieve relatability with the target audience in the ads. We concluded that translation into Spanish was insufficient to engage potential participants, and an additional step is necessary, i.e., incorporating personal experiences of Spanish-speaking participants already enrolled in the study and a voiceover that clearly articulates the study’s purpose, benefits, and requirements. By fostering meaningful connections and providing appropriate messaging, researchers can attract individuals genuinely interested in participation. Ultimately, adapting recruitment strategies to reflect participants’ values and preferences can increase the likelihood of successful engagement and improve the effectiveness of participant recruitment efforts.

This paper introduces ROI-Packing, an efficient image compression method tailored specifically for machine vision. By prioritizing regions of interest (ROI) critical to end-task accuracy and packing them efficiently while discarding less relevant data, ROI-Packing achieves significant compression efficiency without requiring retraining or fine-tuning of end-task models. Comprehensive evaluations across five datasets and two popular tasks-object detection and instance segmentation-demonstrate up to a 44.10

Changes in the driving behavior of older drivers can be indicative of conditions of mild cognitive impairment (MCI), which affect their memory and recognition skills on the road. Traditional clinical evaluations cover only a limited subset of cognitively impaired drivers, prompting the need for innovative technologies to monitor the cognitive status of older drivers routinely. In this study, we developed in-vehicle sensing devices capable of capturing vehicular data streams that reveal older drivers' driving patterns. Using K-means clustering on preprocessed and scaled data, we identified four distinct driver profiles characterized by trip frequency, driving style, and demographic factors. These profiles ranged from active, frequent travelers to sedentary, cautious drivers, with significant differences in trip duration, distance, and vehicle operation metrics such as speed and engine load. A developed random forest model further identified peak hour trips, age, gender, and ambient temperature as significant predictors of MCI, highlighting the complex interplay between lifestyle, driving behaviors, and demographics.