Engineering

Abstract: Consolidated bioprocessing (CBP), where enzyme production, substrate hydrolysis, and fermentation occur in a single bioreactor, provides a promising pathway for lignocellulosic ethanol production. Nevertheless, CBP operation involves trade-offs among ethanol titer, productivity, substrate conversion, soluble sugar accumulation, batch cycle time, and the operating severity associated with temperature and pH profiles. This study introduces a feasibility-aware multi-objective dynamic optimization approach for identifying Pareto-optimal operating policies for batch CBP processes. A simplified, mechanistically driven dynamic model is developed to represent biomass growth, enzyme activity, insoluble substrate hydrolysis, soluble sugar formation and consumption, ethanol production, and inhibition under time-varying temperature and pH profiles. The multi-objective optimization simultaneously maximizes ethanol titer, productivity, and substrate conversion while minimizing sugar accumulation, operating severity, control effort, and batch time. In the main simulation run, 120,000 dynamic policies were evaluated, resulting in 5,017 feasible policies and 328 feasible Pareto-optimal policies under a minimum conversion threshold of 0.42. The optimized dynamic policy achieved an ethanol titer of 1.265 g L−1, a maximum productivity of 0.017 g L−1 h−1, and a maximum conversion of 0.440. Compared with the best static policies, the dynamic Pareto policies improved ethanol titer, productivity, and conversion by 10.6%, 8.3%, and 14.3%, respectively. The feasibility analysis showed that a conversion threshold of 0.42 is stringent but achievable, whereas thresholds of 0.44 and 0.55 were not attainable under the current dynamic model and operating range. Independent-seed repetition confirmed the existence of a consistent high-performing region across different stochastic searches. The resulting Pareto front and operating-policy charts provide a useful basis for selecting temperature and pH profiles for CBP process operation.

Abstract: The human temporomandibular joint requires stable kinematics for optimal function; however, structural anomalies such as the bifid mandibular condyle severely compromise this biomechanical harmony. This study aims to quantify the precise biomechanical behaviour and fracture susceptibility of the bifid condyle using patient-specific finite element analysis. A high-fidelity 3D computational model was constructed from the cone-beam computed tomography data of a patient presenting with a right bifid condyle and concurrent fracture. To establish a comparative baseline, a geometrically healthy control model was computationally derived. Both models were subjected to a simulated, physiological multiaxial masticatory load of 1000 N. The simulation revealed that while the healthy control safely dissipated forces (peak cortical von Mises stress of 62.49 MPa), the bifid morphology fundamentally disrupted load transfer. Extreme mechanical forces concentrated directly at the anomalous inter-condylar notch, generating peak equivalent von Mises stresses approaching 500 MPa and peak compressive stresses nearing 600 MPa. Furthermore, localised strain energy density at the notch peaked at 12 MPa. These internal stress magnitudes significantly exceed the ultimate yield strength of human cortical bone, providing a direct biomechanical rationale for the clinically observed fracture. This computational evidence establishes that the bifid condyle acts as a critical structural vulnerability and energy sink. Consequently, the identification of a bifid condyle warrants proactive clinical management, as even asymptomatic presentations are highly predisposed to structural fatigue and macroscopic failure.

Abstract: Food fraud is a persistent global threat estimated to cost the food industry over USD 30 billion annually. The integration of artificial intelligence (AI) with analytical instrumentation has generated significant research activity directed at developing detection systems capable of identifying adulteration, mislabeling, and substitution across diverse food matrices. This systematic review critically examines the extent to which AI-assisted instrumental technologies contribute to food fraud prevention, and identifies the structural limitations that constrain their real-world implementation. A systematic search of peer-reviewed literature published between 2021 and 2026 yielded 72 eligible studies after application of predefined inclusion criteria. Studies were required to report quantitative performance metrics (accuracy, R², RMSE, AUC, sensitivity, specificity), describe methodological limitations, and mention laboratory or industrial implementation contexts. Data were extracted into a structured seven-sheet workbook covering study characteristics, instrumental technologies, AI architectures, performance metrics, industrial validation status, implementation evidence, and methodological quality. The corpus reveals a systematic pattern of high reported analytical accuracy—frequently exceeding 95% and in many cases reaching 100%—under controlled laboratory conditions. However, 75% of studies (54/72) conducted no external validation, 100% of studies reported no pilot-scale or routine monitoring application, and no study achieved inter-laboratory validation. The predominant technology was NIR spectroscopy (26/72 studies, 36%), followed by gas chromatography-based systems (14/72, 19%) and electronic noses (8/72, 11%). Classical machine learning—predominantly SVM, Random Forest, and ANN—dominated methodological approaches (43/72, 60%), with deep learning architectures accounting for 26% of studies. Technology Readiness Levels were unreported in 97% of studies. Methodological quality was predominantly moderate (42/72 studies scoring 3/5), with 19 studies scoring 2/5 and only one achieving the maximum score. This review identifies a structural gap between detection and prevention as the central finding: the scientific literature consistently demonstrates high analytical precision in laboratory settings while providing minimal evidence of real-world industrial deployment, regulatory integration, or measurable impact on the prevention of food fraud events. The findings demonstrate that the limitation is not primarily technological but systemic, highlighting the need for a paradigm shift from performance-driven research toward validation-driven, deployment-oriented frameworks.

Abstract: Simultaneous electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) provide complementary views of brain activity, capturing neural dynamics across temporal and spatial scales. Integrating these modalities offers a powerful approach for studying brain function, yet remains fundamentally challenging due to differences in measurement mechanisms, temporal resolution, and neurovascular coupling. At its core, EEG–fMRI fusion can be viewed as an inverse problem: the goal is to recover latent neural processes that are only partially observed through electrophysiological and hemodynamic signals. Here, we review data-driven fusion methods developed between 2000 and 2025, focusing on approaches that aim to identify shared neural representations across modalities. We organize the existing methods according to the fusion strategy (symmetric vs. asymmetric), the methodological objective (factorization vs. translation), and the modeling assumptions (linear vs. non-linear), and discuss commonly-used evaluation metrics and visualization strategies. We further examine evaluation strategies, highlighting the lack of a universal validation standard and the challenges of interpreting latent multimodal components. Across neurological, psychiatric, and cognitive applications, EEG-fMRI fusion has revealed distributed network dynamics that are not accessible through unimodal analyses. However, key challenges remain, including temporal misalignment, noise-induced coupling, and model-dependent interpretation. We discuss emerging directions such as nonlinear modeling, flexible coupling frameworks, and large-scale group-level fusion, which may enable more robust and interpretable multimodal integration. Together, this review reframes EEG-fMRI fusion as a problem of latent neural inference and outlines a path toward more principled, scalable, and biologically grounded approaches for understanding brain function and dysfunction.

Abstract: Connexin-43 (CX43) lateralization in ventricular myocardium has been associated with abnormal impulse propagation and increased arrhythmia susceptibility. Its quantitative assessment in histological sections remains challenging because of the difficulty of segmenting individual cardiomyocytes and the reliance of previous methods on geometric rules applied to segmented cell profiles. Here, we present CLARISA, a deep learning framework for classifying CX43-positive regions as either terminal or lateralized directly from fluorescence images, without requiring cardiomyocyte segmentation. An expert-annotated dataset was generated from left-ventricular cryosections of Wistar rat hearts, in which CX43-positive regions were labeled according to their distribution pattern. A dual-stream convolutional classifier based on EfficientNetV2-S was trained to capture both the local and contextual morphology of each region. In addition, an inference module applicable to whole tissue sections was developed to generate spatial lateralization probability maps and global percent lateralization estimates consistent with expert annotation. On the test set, CLARISA achieved a ROC-AUC of 0.905 and a PR-AUC of 0.810. These results support the feasibility of automated assessment of CX43 distribution patterns without explicit cardiomyocyte segmentation. The complete codebase is publicly available, together with access to the pretrained model and the image data used in this study. The Hugging Face model card reports the same held-out test metrics and states that the checkpoint is intended to be used with the main repository.

Abstract: This systematic review provides a comprehensive and quantitatively grounded synthesis of machine learning (ML) approaches for electrocardiography (ECG)-based detection of dysglycemia, with a specific focus on translational readiness for clinical screening. A structured literature search across PubMed, Scopus, Web of Science, and IEEE Xplore (February 2025) identified 183 records, of which 17 studies met predefined inclusion criteria following PRISMA-guided screening. The included studies demonstrate substantial heterogeneity in dataset size (ranging from <50 to >25,000 subjects), ECG acquisition modalities (single-lead, 12-lead, wearable), feature representations (raw signals, heart rate variability, engineered features), and ML strategies (classical algorithms, deep learning, and multimodal models). Reported model performance is generally high, with accuracy values frequently exceeding 0.85 and area under the curve (AUC) ranging from 0.78 to 0.99. Smaller experimental studies often report inflated performance (up to 96–99% accuracy), whereas large-scale population-based investigations demonstrate more moderate but clinically plausible results (AUC ≈ 0.80–0.85). External validation, a key requirement for clinical applicability, was performed in only a limited subset of studies (approximately 12%). From a physiological perspective, ML models exploit ECG alterations associated with dysglycemia, including reduced heart rate variability, QT interval prolongation, and changes in ventricular depolarization and repolarization dynamics. However, the relationship between metabolic dysfunction and ECG signals remains indirect. A key finding of this review is the mismatch between reported predictive performance and model maturity. The majority of studies (≈65–70%) are classified as early-stage (Level 1–2 or 2–3), relying on small, single-center datasets and internal validation. Only a minority of studies achieve near-translational maturity (Level 4), characterized by large-scale datasets and external validation. ECG-based dysglycemia detection represents a promising non-invasive and scalable screening paradigm. However, its clinical translation is constrained by the lack of standardized ECG acquisition protocols, limited dataset diversity, insufficient external validation, and fragmented methodological frameworks. Future research should prioritize large multi-center datasets, standardized feature extraction pipelines, hybrid interpretable models, and prospective validation to enable robust, generalizable, and clinically deployable screening systems.

Abstract: Background: Accurate assessment of neonatal sleep is critical for monitoring brain development and identifying potential neurological disorders, yet manual scoring of multi-channel EEG recordings is labor-intensive and prone to variability. Methods: To address this, we propose a lightweight temporal-spatial feature fusion network for automatic neonatal sleep staging. The model employs a dual-branch architecture to separately capture temporal dependencies and spatial correlations in EEG signals, which are then integrated via an adaptive fusion module to obtain comprehensive feature representations while maintaining low computational complexity. Results: The framework was evaluated on a clinical neonatal dataset (CHFD) for tasks including sleep–wake classification, quiet sleep detection, and three-stage sleep staging, achieving superior performance compared with several state-of-the-art methods. Additional experiments on the MASS-S3 adult dataset demonstrate that the model retains competitive accuracy and F1-score, indicating strong generalization across populations. Conclusions: These results suggest that jointly modeling temporal and spatial features enables robust and efficient automatic sleep staging. The proposed approach offers a practical solution for clinical applications and edge deployment, providing reliable, multi-dimensional assessment of neonatal brain activity and laying the groundwork for future studies integrating larger datasets or multimodal physiological signals.

Abstract: Precision livestock farming (PLF) is hindered by high costs, infrastructure demands, and complex deployment. To address these barriers, we developed a practical, open-hardware wearable system for real-time movement tracking and behavior classification in pasture-based livestock. The collar-mounted device integrates a 6-axis IMU, GPS, and a low-power microcontroller within a modular architecture, and uses a novel routerless communication protocol to transmit data directly to a base station—eliminating reliance on network infrastructure. The system supports two operational modes: (1) synchronized data logging for video annotation, or (2) real-time embedded behavior classification. A year-long field trial confirmed robust performance, minimal animal disturbance, and resilience under harsh conditions. Remarkably, the system achieved near-perfect true positive rates (>0.95) for both basic and subtle behaviors using only one hour of annotated video for training, drastically reducing labeling effort. All design assets—CAD files, schematics, firmware, bill of materials, and protocols—are openly released to ensure full reproducibility. This work delivers a validated, scalable, and accessible tool that lowers entry barriers for researchers and developers, enabling rapid deployment and community-driven adaptation for diverse livestock applications.

Abstract: Breast cancer remains a leading cause of cancer-related mortality among women globally. This study makes one focused primary contribution: a formalized, physics-grounded preprocessing-to-fusion pipeline for multi-modal breast cancer classification that is rigorously validated under both centralized and federated learning conditions. Patient-wise stratified 5-fold cross-validation was applied across Ultrasound (BUSI, n=780), Dynamic Contrast-Enhanced MRI (DUKE, n=922), and Mammography (CBIS-DDSM, n=400). Per-modality models achieved 92.50±1.2%, 90.63±1.5%, and 92.00±1.3% accuracy (McNemar’s p<0.05 vs. baselines). Weighted late-fusion achieved 93.10±1.1% (p=0.031 vs. best individual modality). A five-algorithm FL comparison (FedAvg, FedProx, SCAFFOLD, FedNova, FP16-FedAvg) under IID and non-IID (Dirichlet α=0.5) conditions is provided with per-round training time, communication time, per-round latency, and cumulative bandwidth. FP16 transmission reduced bandwidth from 8.14 GB to 1.23 GB (−84.9%, p=0.74 vs. FP32). SCAFFOLD achieved the best non-IID accuracy (90.50%). All design choices are validated by ablation experiments with McNemar’s test and Cohen’s h effect sizes.

Abstract: The development of remote blood pressure (BP) measurement algorithms using remote photoplethysmography (rPPG) has significant limitations, including the small size of publicly available datasets, privacy concerns regarding facial videos, and a lack of diverse, realistic datasets associated with actual BP measurements. To address these challenges, this study aimed to provide comprehensive, simultaneous recordings of participants' faces, along with reference physiological measurements, for 300 adult participants aged 18–65 years. For each imaging session, systolic and diastolic blood pressure and reference heart rate (HR) were recorded using clinical electronic BP monitors in addition to recording illuminance (lux) values for indoor and outdoor environments. The collected data, called CLBP-300, is a crucial resource for developing and evaluating remote vital signs from facial rPPG signals. A sample of videos is publicly available to demonstrate data quality, while academic researchers can access the complete dataset under a strict data use agreement. The data and python code presented in this study are available on https://sites.google.com/view/clbp-300?usp=sharing.

Abstract: Automated cellular detection using deep learning is a key strategy for optimising cervical cancer screening by reducing the healthcare workload and inter-observer variability. However, analyzing Whole Slide Image (WSI) patches presents challenges like annotation scarcity, morphological complexity, and class imbalance. This study conducts a systematic evaluation of YOLOv11 (n, s, and m variants) to assess the impact of target variable granularity and training paradigms on performance. Four strategies were analyzed: independent and multi-class models, each evaluated at both specific cell label and diagnostic macro-group levels. To ensure clinical robustness, patient-level data partitioning was implemented to prevent information leakage. Performance was measured using precision, recall, and mAP (0.5 and 0.5:0.95). The results reveal critical trade-offs between fine-grained discrimination and model generalization when varying architectural complexity and labeling strategies. Findings indicate that diagnostic aggregation improves stability, while single-class training optimizes specialized detection. These results provide methodological guidelines for designing AI-assisted screening systems and establish a foundation for integrating YOLOv11 detectors into Multiple Instance Learning (MIL) frameworks at the WSI scale.

Abstract: This observational study evaluated changes in selected performance parameters of 15 new high-speed dental handpieces after eight months of routine clinical use in a routine educational undergraduate environment (two 4h daily clinical shifts, five days per week, with repeated sterilization cycles). All handpieces underwent routine cleaning, lubrication, and autoclave sterilization as instructed. The turbine components from the handpieces were disassembled and examined by stereomicroscopy before and after use, while free-running speed and bur-tube friction grip force were assessed at the same intervals. Two handpieces were no longer operational at follow-up due to ball bearing failure. Among the remaining handpieces, statistically significant reductions were observed in both free-running speed and friction grip force (p < 0.01). Microscopic examination of the rotors revealed surface alterations consistent with corrosion and wear. Within the limitations of this study, routine clinical use over an eight month period was associated with measurable changes in key performance characteristics of high-speed dental handpieces in educational clinical settings.

Abstract: Conductive intracardiac communication (CIC) is one of the most innovative and promising communication technologies in multi-point cardiac pacing schemes that utilize the heart as the transmission channel in recent years. Current research predominantly focuses on the static channel characteristics, with only limited investigations into the dynamic responses of amplitude-frequency and amplitude-time behaviors. Designing CIC systems solely on the basis of static properties can result in inaccurate channel estimation, distorted channel state information (CSI), elevated bit error rate (BER), and overall degradation of system communication performance. To solve the problems of dynamic channel measurement and modeling of the heart, this paper for the first time proposes a dynamic channel modeling method for CIC based on sinusoidal response and impulse response. Firstly, we develop a physical simulation and miniaturized measurement setup to measure dynamic cardiac channel, and analyze the amplitude-frequency characteristics and amplitude-time characteristics. The influence of factors such as instrument differences, heart rate, flow rate and comparative experiments of free electrodes and fixed electrodes on the channel characteristics are also discussed. Secondly, we systematically analyze the path loss, shadowing effect, multipath effect and Doppler effect of the CIC channel. Combined with the dynamic channel characteristics and parameters, we propose a composite fading dynamic channel model and analyze the BER performance of baseband signals transmission and On-Off Keying (OOK) modulation systems. We can conclude that (1) The CIC channel exhibits capacitive characteristics. The fluctuation of the channel gain of the free electrode is mainly caused by motion artifacts. The fixed electrode can effectively suppress this interference. (2) The dynamic channel gain of CIC varies periodically with the heartbeat, and the fluctuation range of the signal is less than 1-2 dB. This is due to the length change of the myocardial tissue. (3) The CIC channel still presents extremely weak shadow fading, no significant multipath, and no measurable Doppler characteristics under dynamic conditions, belonging to an extremely slow fading channel. This work provides effective dynamic channel measurements approach and parameter basis for the transceiver design of CIC and a reliable model for the simulation for CIC systems.

Abstract: Myocardial stiffness is a critical determinant of cardiac function and disease, influencing ventricular filling, contractility, mechanotransduction, and the progression of conditions such as hypertrophic cardiomyopathy, myocardial infarction, and heart failure with preserved ejection fraction. Over the past two decades, research in cardiac biomechanics has advanced from conventional ex vivo tissue characterization to multiscale experimental investigation, sophisticated constitutive modelling, and patient-specific computational inference based on imaging modalities such as magnetic resonance imaging and echocardiography.Despite these advances, the field remains fragmented across experimental biomechanics, computational modelling, and clinical imaging. Experimental studies commonly focus on isolated tissue characterization using biaxial testing, indentation, and rheological methods, whereas computational studies increasingly employ inverse finite element frameworks to estimate myocardial stiffness in vivo. At the same time, growing evidence indicates that myocardial viscoelasticity and other time-dependent mechanical behaviours play an important role in cardiac function, although these features are still insufficiently incorporated into many constitutive models.This review synthesises current knowledge on passive and viscoelastic myocardial stiffness across scales by integrating experimental methods, constitutive modelling strategies, and image-informed computational approaches. It examines the influence of myocardial microstructure, fibre architecture, extracellular matrix remodelling, and fibrosis on tissue stiffness, and reviews emerging techniques for non-invasive estimation of myocardial mechanical properties. The review also considers the potential of patient-specific cardiac digital twins for clinical decision support. Finally, it identifies key methodological challenges, unresolved questions, and future opportunities for advancing standardised mechanical characterisation and the clinical translation of cardiac biomechanics.

Abstract: Mental stress is a common issue in demanding occupational setups, such as smart industrial settings, particularly from working with robots, being one of the primary reasons for decreased performance and productivity. Quantifying and evaluating stress are critical for worker safety, performance, and overall well-being. A novel integrated framework is proposed in this research for digitising and assessing cognitive stress that combines neuroimaging (EEG and fNIRS), gaze tracking and machine learning. A factory workers’ stress assessment experiment is designed and implemented, which employs physiological, behavioural and subjective measures to assess mental stress from different perspectives. Physiological features extracted from multimodal data are used for training supervised classification and regression models. To further optimise the pipeline, multiple feature selection algorithms are tested, followed by ensemble learning approaches, and the best one is chosen for stress prediction. After implementing the novel stress quantification framework for the factory workers' stress assessment experiment, the ensemble learning approach produced the best results for both regression (RMSE: 10.86) and classification (accuracy: 84.1%) techniques using the STAI score as the target. The behavioural and subjective measures demonstrate the effect of varying process variables (light, noise, task speed, and complexity) during the experiment. Multimodal data, machine learning, and other computational approaches are integrated in this study to objectively quantify cognitive stress, utilising the novel stress quantification framework presented in this research, thereby bridging the gap between research and practical application. This study proposes a multi-domain framework for measuring stress, providing a promising solution for worker well-being in occupational setups.

Abstract: This paper proposes a two-stage hybrid framework for biosignal quality validation that produces beat-level and segment-level labels for real-time filtering and offline dataset curation. The framework is designed for non-stationary periodic biomedical time-series signals including electrocardiography (ECG), photoplethysmography (PPG), impedance cardiography (ICG), phonocardiography (PCG), electromyography (EMG), and electroencephalography (EEG) and is demonstrated and evaluated primarily on ECG. A prerequisite is synchronized acquisition of the primary biosignal together with inertial motion sensing (IMU/accelerometer) and electrode impedance or lead-off status, with the IMU positioned near the sensing electrodes. The first stage performs sensor-integrity gating to reject intervals corrupted by motion or poor electrode contact. The second stage applies software signal-quality indices to the remaining beats, including physiological plausibility constraints (R to R peaks analysis), DTW-based morphological consistency against adaptive templates, frequency-domain SNR estimation, and baseline-wander quantification. This study systematically evaluates and compares the classification performance of six complementary sensor-level and software-based signal quality assessment methods. When integrated within the proposed hybrid framework, validation against expert-annotated ECG quality labels demonstrates high performance, achieving approximately 98% accuracy, 98% F1-score, 99% sensitivity, and 97% specificity. This modular, extensible approach enhances the trustworthiness of downstream analytics by preventing contaminated segments from entering feature extraction and model training pipelines, enabling more stable physiological monitoring in free-living conditions, reducing false alarms in continuous monitoring applications, and generating higher-quality datasets for AI-based diagnostic systems.

Abstract: Wearable lower-limb exoskeletons can enhance mobility, reduce metabolic cost, and aid rehabilitation. Effective human-exo cooperation requires personalized assistance profiles that match biomechanical principles. Existing methods often rely on fixed curves, involve complex tuning, and lack biomechanical interpretability. To address this, we propose a “Physics-guided perception and physiology-driven optimization” approach. First, a Physics-guided Dynamic Fusion Model (PDFM) is proposed, which integrates Newton–Euler dynamics, LSTM, and NTM to estimate multi-plane hip joint moments without ground reaction forces, employing biomechanical models as complementary fusion factors rather than the embedded hard constraints used in conventional physics-informed neural networks (PINNs). Estimation accuracy reached 92.51% (sagittal), 86.86% (coronal), and 88.15% (transverse), outperforming all single-network baselines across all three anatomical planes. Second, an assistance profile derived from estimated moments is individually optimized using Bayesian optimization based on multi-muscle sEMG. Compared to no-exo walking, the optimized system reduced target muscle loading by 49.31% and metabolic cost by 14.75%; relative to the pre-optimized profile, the reductions were 23.64% and 5.74%, respectively. This work provides a validated framework for personalized hip exoskeleton assistance.

Abstract: The production of biosolids in Brazil has increased due to the expansion of Sewage Treatment Plants, making these materials a sustainable alternative for agricultural use. Composed of high organic matter and nutrients such as nitrogen, phosphorus, calcium, and magnesium, biosolids have the potential to improve the physical, chemical, and biological properties of tropical soils, contributing to greater fertility, water retention, and microbial activity. National literature demonstrates that these materials can par-tially replace mineral fertilizers and assist in the recovery of degraded areas. On the other hand, the presence of contaminants still represents a challenge. Heavy metals such as Cd, Pb, Ni, and Hg generally appear in low concentrations, while Cu and Zn tend to approach the maximum limits established by CONAMA Resolution No. 498/2020. Regarding pathogens, the efficiency of sanitization depends on the treatment method employed. Emerging organic pollutants, including pharmaceuticals and hor-mones, have been detected, but still lack specific regulations in Brazil. Thus, although biosolids present high agronomic potential, their safe use requires adequate monitor-ing, improvement in controlling the origin of sewage, and advances in legislation, es-pecially regarding emerging organic pollutants.

Abstract: The fabrication of complex architectures remains a central challenge in 3D bioprinting, where low mechanical properties of hydrogels restrict the range of feasible geometries. Four-dimensional (4D) bioprinting can mitigate these limitations by introducing programmed structure shape-morphing in response to external stimuli. However, in most existing approaches, shape-morphing behavior is introduced after hydrogel formation, limiting the complexity of the resulting deformation. Here, a proof-of-concept strategy is presented, in which shape-morphing is directly encoded during fabrication. By modulating light exposure time layer-by-layer in vat photopolymerization, spatial variations in crosslinking density are introduced in situ within GelMA hydrogel constructs. Upon immersion in aqueous media, these variations generate differential swelling, leading to controlled bending of the printed structures. This approach enables the programming of deformation pathways at the printing stage, without requiring additional materials or post-processing steps. The morphing behavior was further supported by finite element simulations, which reproduced the experimentally observed deformation and enabled prediction of the shape change. Overall, this study demonstrates that swelling-driven actuation can be encoded during fabrication. Although demonstrated on simplified geometries, this approach provides a versatile framework for process-driven shape morphing programming and represents a step toward more spatially resolved and potentially volumetric 4D bioprinting strategies.

Abstract: Globally, the burden of breast cancer remains high as it is the most prevalent malignancy among women and a major contributor to cancer mortality, with therapeutic success often limited by drug resistance, treatment toxicity, and tumor heterogeneity. Cold atmospheric plasma (CAP), a partially ionized gas enriched in reactive oxygen and nitrogen species (RONS) electromagnetic waves and ultraviolet radiation, has emerged as a selective antitumor therapy, inducing cancer-specific cytotoxicity while sparing normal tissue. Here, we review the mechanisms of CAP action against breast cancer, including RONS-mediated oxidative stress, mitochondrial disruption, apoptosis, immunogenic cell death, and suppression of metastatic and angiogenic pathways. Notably, This approach selectively targets therapy-resistant breast cancer stem cells and sensitizes the highly aggressive forms, particularly triple-negative breast cancer (TNBC). Its synergy with drug therapy, radiotherapy, immunotherapy and surgery further broadens therapeutic potential. Advances in delivery platforms, such as plasma-activated media, nanoparticles, and hydrogels, address CAPs instability and enhance tumor penetration. Despite promising preclinical results, clinical translation faces barriers such as the short half-life of RONS, device standardization, and unresolved immunomodulatory effects. Overcoming these challenges through interdisciplinary collaboration and optimized protocols may unlock the potential of CAP for precision oncology.