Search Results (1,585)

Region-level weakly supervised hyperspectral target detection (HTD) using multiple-instance learning (MIL) reduces annotation costs but encounters challenges such as bag label ambiguity, boundary over-smoothing, and test-time computational latency. To address these issues, we propose Hyper-VMIL, a spatial–spectral topology-regularized variational hypergraph network. Hyper-VMIL formulates latent target localization as variational inference over dual-path hypergraphs: a boundary-aware spatial hypergraph modeling geometric patch continuity and a dynamic spectral-manifold hypergraph capturing non-local material similarity. Node-adaptive gating dynamically balances spatial and spectral evidence to mitigate over-smoothing near target boundaries. Furthermore, a confidence-aware continuous posterior refinement (CTPR) mechanism reduces the confirmation bias associated with conventional hard pseudo-label binarization. Finally, a teacher–student distillation strategy transfers contextual topology into a lightweight single-spectrum student detector. Benchmark experiments on simulated ASTER and airborne MUUFL Gulfport and Avon datasets show that Hyper-VMIL achieves competitive performance against 15 baseline methods. Notably, Hyper-VMIL supports dual inference modes: Context Mode provides improved detection accuracy (+4.6% average NAUC over VMIL-ECM on MUUFL), while Pixel Mode enables single-spectrum inference ($1.25\mathsf{\mu }\mathrm{s}$ single-instance latency and an amortized streaming throughput of $0.015\mathsf{\mu }\mathrm{s}$ per pixel) suitable for onboard real-time deployment. Full article

China’s pursuit of its dual-carbon targets amid rapid artificial intelligence (AI) development raises an important question: do the environmental implications of AI innovation change as regional innovation advances? Using a balanced panel of 30 Chinese provinces from 2011 to 2024, this study measured patent-based AI innovation intensity using applications identified through a strict AI patent classification. Linear and quadratic models with province and year fixed effects were estimated, and the Lind–Mehlum test was used to assess the shape of the relationship within the observed range. The preferred specification indicates an inverted-U-shaped association: carbon emission intensity initially increases with patent-based AI innovation but declines beyond an interior turning point. The negative quadratic coefficient remains stable when the emissions data source, patent classification, sample period, treatment of outliers, and timing of the AI terms are varied. Supplementary Bartik and copula-control analyses preserve the negative curvature, although their identification limitations preclude a definitive causal interpretation. A Kaya-based exact decomposition shows that the estimated curvature is concentrated in energy intensity rather than the carbonization factor. Human capital strengthens the estimated concavity, while the clearest regional contrast is observed between central and eastern China, with the strongest curvature in the central provinces. These findings suggest that greater AI patenting does not automatically reduce emissions. Its environmental implications depend on the stage of regional innovation and its interaction with energy efficiency and absorptive capacity. Policies promoting AI innovation should therefore be coordinated with cleaner energy supply, efficiency improvements, and human capital investment. More broadly, the study provides a stage-sensitive basis for evaluating the sustainability implications of patent-based AI innovation through measurable changes in carbon emission intensity. Full article

This study investigates logistics convergence, financial development, and trade competitiveness in 19 Mediterranean countries over the period 2007–2022. Unlike previous studies that primarily examine average logistics performance across countries, this study contributes by combining convergence-club analysis, logistics inequality decomposition, and second-stage ordered-logit modelling to identify the structural characteristics associated with different logistics regimes across Mediterranean economies. Based on the Logistics Performance Index (LPI), we employ descriptive statistics, sigma-convergence analysis, the Phillips–Sul club convergence approach, relative transition paths, Theil decomposition of logistics inequality, and second-stage ordered-logit estimation. The full-panel Phillips–Sul test rejects the hypothesis of overall convergence, indicating that Mediterranean economies do not converge towards a common logistics-performance equilibrium. However, the club formation procedure identifies three statistically supported logistics convergence clubs, while Libya cannot be assigned to any convergence club. The transition-path analysis reveals three distinct logistics regimes corresponding to upper, middle, and lower convergence groups. The Theil decomposition indicates that the between-club component of logistics inequality becomes increasingly dominant after 2012 and accounts for most logistics inequality during the later years of the sample. Second-stage ordered-logit estimates indicate that GDP per capita, the rule of law, and economic complexity are positively associated with membership in higher logistics convergence clubs, whereas financial development does not emerge as a statistically significant direct predictor. Taken together, the findings suggest that financial development is associated with logistics convergence primarily through broader institutional and structural channels rather than emerging as an independently significant predictor in the ordered-logit analysis. Overall, the findings provide policy implications for promoting balanced logistics development and reducing structural disparities across Mediterranean economies. Full article

Objectives: To determine whether the superior anatomical control achieved with a treat-and-extend (T&E) anti-VEGF regimen translates into better five-year visual outcomes than a pro re nata (PRN) regimen in treatment-naïve center-involving diabetic macular edema (CI-DME), and to characterize the anatomical–functional relationship. Methods: In this retrospective propensity-score–matched survivor cohort, one eye per patient was analyzed. One-to-one nearest-neighbor matching balanced 14 baseline covariates. Longitudinal best-corrected visual acuity (BCVA) and central subfield thickness (CST) were analyzed using linear mixed-effects models with matched-pair clustering. Absence of a clinically meaningful visual difference was evaluated using two one-sided tests (TOST) with a prespecified ±5-letter equivalence margin. The anatomical–functional relationship was assessed by segmented regression. Sensitivity analyses included inverse probability of treatment weighting, doubly robust estimation, inverse probability of censoring weighting, best-/worst-case imputation, interval-censored recurrence modeling, and E-value analysis. Results: Both regimens improved BCVA, with no clinically meaningful difference at five years (T&E +6.2 ± 14.6 vs. PRN +6.9 ± 14.0 letters; difference −0.7 letters; 90% CI, −2.4 to 1.0; TOST p < 0.001). T&E achieved greater CST reduction (−172.3 vs. −114.2 µm; difference −58.1 µm; p < 0.001), higher dry-macula rates (73.8% vs. 59.5%; p < 0.001), fewer recurrences (2.2 vs. 3.9; p < 0.001), and fewer monitoring-only visits (14.6 vs. 28.4; p < 0.001), but required 53% more injections (25.1 vs. 16.4; p < 0.001). Segmented regression identified a breakpoint at 148 µm CST reduction; additional thinning beyond this threshold was not associated with further visual improvement. Baseline BCVA, ellipsoid zone disruption, and diabetic retinopathy severity, but not treatment regimen, independently predicted five-year vision. Results were consistent across sensitivity analyses. Conclusions: Among patients completing five years of therapy, additional anatomical drying beyond an exploratory, cohort-specific breakpoint of approximately 150 µm CST reduction was not associated with further measurable visual gain in this observational cohort. Regimen selection should therefore reflect treatment burden, monitoring requirements, and patient preference rather than anticipated visual superiority. Because the cohort included only patients completing five years of follow-up, these findings apply to adherent patients and should not be generalized to unselected populations. Full article

This study presents a comparative evaluation of the wear performance of three hot-work tool steels—WCLV (1.2344), Uddeholm Unimax, and Uddeholm QRO 90 Supreme—for die forging applications. The materials were characterized in terms of hardness and bulk chemical composition using Vickers hardness testing and X-ray fluorescence spectroscopy. Tribological behavior was investigated using ball-on-disc tests, while industrial performance was assessed by analyzing forging punches after the production of 16,250 components. Surface degradation was quantified using optical profilometry and three-dimensional roughness parameters. Measured hardness values were 591 HV for WCLV, 622 HV for QRO 90 Supreme, and 639 HV for Uddeholm Unimax. The average friction coefficients were 0.88, 0.92, and 0.77, respectively. Unimax also exhibited the lowest volumetric wear, reaching 0.04683 mm³ (R19 mm) and 0.03384 mm³ (R22 mm), compared with 0.08646–0.13095 mm³ for WCLV and 0.09598–0.13635 mm³ for QRO 90 Supreme. This corresponds to approximately 45–70% lower wear relative to the other steels. Industrial trials confirmed improved surface stability of Unimax punches after service. The observed trends are consistent with differences in alloying content and the expected microstructural response associated with chromium and molybdenum additions. Overall, Uddeholm Unimax demonstrated the most favorable balance of hardness, friction behavior, and wear resistance. Full article

Against the background of global production-network restructuring, low-carbon transition, and rising external uncertainty, this study examines the effect of green infrastructure investment on urban industrial chain resilience. Using panel data for 285 Chinese prefecture-level cities from 2012 to 2024, industrial chain resilience is measured from the dimensions of industrial diversification and urban innovation capacity. Double machine learning is employed for baseline estimation, supplemented by mediation analysis, threshold regression, spatial econometric analysis, and a series of robustness tests. The results show that green infrastructure investment significantly enhances industrial chain resilience, and the finding remains robust to alternative model specifications, cross-fitting settings, generalized propensity score weighting, continuous-treatment entropy balancing, winsorization, and the exclusion of pandemic-period observations. Resource allocation efficiency plays a partial mediating role in this relationship. The threshold analysis identifies a significant nonlinear effect associated with energy consumption intensity, with the positive effect of green infrastructure investment being stronger below the estimated threshold and weakening above it. Spatial analysis further shows significant spatial dependence in both green infrastructure investment and industrial chain resilience, together with positive spillover effects on neighboring cities. These findings highlight the importance of improving green infrastructure investment efficiency, strengthening factor allocation, and promoting regional coordination in enhancing urban industrial chain resilience. Full article

Feature engineering has become a common step in AI-based HVAC forecasting, often involving variables calculated from raw building management system (BMS) measurements, such as temperature differences, setpoint tracking deviations, airflow balance indicators, rolling statistics, and temporal or seasonal descriptors. Accurate short-term forecasting can provide a baseline of expected operation for anomaly and fault detection and can support control optimization and operator decision making. However, real-world deployment is complicated due to differences in BMS sensor availability and data quality, as well as the preprocessing and maintenance burden associated with complex feature sets. The actual contribution of these features to the performance of AI forecasting remains underexplored, particularly for short-term prediction of air handling unit (AHU) operation. This study evaluates the impact of features on short-term AHU forecasting using three deep learning (DL) architectures: Temporal Convolutional Networks (TCNs), Long Short-Term Memory (LSTM) networks, and a hybrid CNN–LSTM model. An actual operational AHU dataset from a BMS was used to predict key operational variables, including supply and extract air temperatures, supply and extract fan operating signals, and supply air temperature setpoint-tracking error. Fan signal balance was additionally evaluated as a derived indicator calculated from the two predicted fan signals. Four input configurations were evaluated: (i) full (74 inputs), containing raw BMS measurements, short-cycle temporal variables, engineered and dynamic features, and annual-calendar information; (ii) no annual calendar (68 inputs), identical to full but excluding annual-calendar variables; (iii) raw + short-cycle temporal (20 inputs); and (iv) raw-only (12 inputs). The models used a 60-min input history to forecast the following 30-min at one-minute resolution. Persistence and Ridge models were included as reference baselines. All models were trained and tested on identical data splits and forecasting horizons to ensure a fair comparison. Each DL experiment was repeated across five independent runs, and performance was evaluated using MAE, RMSE, and R². The TCN showed the strongest overall DL performance. Raw-only achieved the highest mean R² in 11 of 15 architecture–target comparisons using just 12 inputs. The best mean DL R² ranged from 0.916 for the fan signals to 0.993 for extract air temperature. Annual-calendar features improved the TCN results but provided no consistent benefit for the LSTM or CNN–LSTM. Ridge slightly outperformed the best DL configurations for temperature-related targets, reflecting the strong short-term continuity of these signals. These findings show that recent raw BMS measurements contain most of the information needed for accurate 30-min AHU forecasting, while explicit seasonal and engineered features provide limited additional value. The resulting simpler models may in the future be used as forecasting components in predictive control and fault detection systems. However, their control and energy-saving benefits must be tested separately. Full article

Objective: The aim of this study was to evaluate the efficacy and safety of endovascular and hybrid treatment approaches in patients with aortic arch aneurysms. Materials and Methods. This retrospective study included 68 male and female patients with a confirmed diagnosis of either an aortic arch aneurysm or a descending thoracic aortic aneurysm with a short proximal landing zone (<1.5 cm) who underwent either hybrid or endovascular treatment at the Department of Vascular Surgery between January 2017 and December 2024. Study outcomes included a composite measure of technical success, a composite measure of in-hospital clinical success, and a composite measure of long-term treatment outcomes, including stroke, myocardial infarction, and aortic-related mortality. Results. All 68 patients were divided into two groups: Group I comprised patients who underwent endovascular treatment, whereas Group II included patients who underwent hybrid surgical treatment. The groups were comparable with regard to demographic and anatomical characteristics, clinical presentation, and comorbidities. The composite technical success rate (defined as successful target stent-graft deployment without conversion to open surgery and absence of type I or type III endoleaks) was comparable between the groups at the intraoperative stage (p = 1.000). The composite measure of in-hospital clinical success was achieved in 33 patients (94%) in Group I and 22 patients (67%) in Group II and was significantly higher in the endovascular group (adjusted p = 0.005). This difference was primarily attributable to a higher incidence of complications in the hybrid treatment group, including stroke (9%) and peripheral nerve injury (9%), associated with the open surgical component of the procedure. The mean follow-up duration was shorter in Group I (19.3 ± 10.4 months) than in Group II (63.9 ± 29.5 months), reflecting the fact that most patients in Group I underwent treatment during the later years of the study period. Although a difference in the composite long-term outcome measure was observed before adjustment (p = 0.031), this finding did not remain statistically significant after correction for multiple testing (adjusted p = 1.000). Conclusions. In this preliminary single-centre study, endovascular and hybrid approaches showed comparable technical efficacy in the early postoperative period. However, hybrid surgical treatment was associated with a less favourable safety profile during the early postoperative period, as reflected by the significantly lower in-hospital composite clinical success rate and longer hospital stay than in the endovascular group. These findings remained robust after correction for multiple testing. Long-term results should be interpreted with caution and require confirmation in larger prospective studies with longer and balanced follow-up periods. Full article

Culicoides biting midges are important veterinary vectors, yet reported distributional patterns in Türkiye may reflect both ecological structure and uneven sampling. We synthesized 58 eligible Türkiye-specific reports (50 database-derived and eight citation-traced), consolidated them into 43 study/data-source families, and compiled 1164 taxon records representing 62 taxa. The all-explicit province dataset included 33 provinces and 45 taxa, whereas the primary province dataset included 18 provinces and 43 taxa. National taxonomic sources supported 72 species within the eligible evidence chain and 74 under a later taxonomic interpretation. Reported richness was strongly associated with sampling effort (Poisson coefficient = 0.579, 95% CI: 0.385–0.773), whereas biogeographical region was not significant after effort was accounted for (likelihood-ratio p = 0.914). Jaccard PERMANOVA identified an exploratory regional compositional pattern (F = 4.388, R² = 0.369, p = 0.001; PERMDISP p = 0.325), with turnover accounting for approximately 78% of mean between-region dissimilarity. Most richness diagnostics ranged from 54 to 64 taxa, whereas Chao2 was unstable (69.6; 95% CI: 49.5–152.0). Vector-related evidence remained insufficient for species–pathogen risk modeling. Overall, the synthesis is consistent with a Wallacean shortfall strongly associated with uneven sampling and supports standardized, biogeographically balanced surveillance to test transition-zone structure, climate-sensitive phenology and redistribution, and integrated vector-host-pathogen hypotheses under comparable sampling effort. Full article

Green mergers and acquisitions (M&A) may enable firms to acquire external environmental technologies, assets, and organizational capabilities. However, whether green M&A is associated with subsequent green innovation, how this association evolves across innovation types and time horizons, and whether prior carbon information disclosure conditions this process remain unclear. Using 40,923 firm year observations of Chinese A-share-listed firms from 2012 to 2024, this study combines licensed green M&A data from Zhixing Data Analytics, green patent data from CNRDS, and carbon disclosure, financial, and corporate governance data from CSMAR. The baseline treatment identifies firm years in which at least one green M&A transaction first announced during the year was subsequently recorded as completed. The analysis employs firm and year fixed-effects models, common-sample distributed-lag specifications, formal cross-type coefficient comparisons, forward-outcome tests, propensity score matching, entropy balancing, and alternative measures and specifications. Green M&A is positively associated with total green patenting. The baseline coefficient of 0.045 implies an approximately 4.65% increase in one plus the number of total green patent applications. The contemporaneous association is stronger for green utility model patenting than for green invention patenting, whereas the association with invention patenting becomes more evident in subsequent periods. Prior carbon information disclosure positively moderates the association between green M&A and one-year-ahead invention patenting (β = 0.220, p = 0.018), while the corresponding moderation estimates for total and utility model patenting are not statistically significant. The findings are supported by observable selection adjustments and several alternative measurements and specifications, although fixed-effects PPML estimates using the original patent counts are not statistically significant. This study provides an integrated framework for understanding how innovation timing and prior information governance shape the green M&A dilemma. The results suggest that regulators and investors should assess green acquisitions using credible pre-acquisition carbon disclosure and post-acquisition innovation trajectories rather than relying on environmental transaction labels alone. Full article

Targeted CSN2 codon-67 genotyping supports A2-oriented breeding and herd management. We compared Kompetitive Allele-Specific PCR (KASP) with a locally implemented gel-based tetra-primer amplification refractory mutation system PCR (T-ARMS PCR) workflow in a balanced, non-population-representative Holstein–Friesian panel using re-audited Sanger-supported reference classifications. Both assays distinguish His67-associated (A1-type) from Pro67-associated (A2-type) β-casein classes rather than complete CSN2 alleles. The paired analysis included 96 independent samples. KASP produced 90 reference-concordant, four reference-discordant, and two no-call outcomes (93.8% all-record reference-concordant yield), whereas T-ARMS PCR produced 77 reference-concordant, six reference-discordant, and 13 ambiguous outcomes (80.2%). Concordance among callable/evaluable results was 95.7% and 92.8%, respectively. In the primary comparison, KASP showed a 13.5 percentage point higher yield (95% CI, 4.2–22.9 percentage points; p = 0.015), mainly because it generated fewer non-interpretable outcomes. All callable KASP discordances involved Pro67/Pro67 reference samples called His67/Pro67. Because no independent DNA dilution, new DNA extraction, repeat KASP run, or new Sanger sequencing was performed, their cause and reproducibility remain unresolved. Formal repeatability and between-run reproducibility were not systematically evaluated; therefore, this comparison is not a formal assay validation and does not define T-ARMS PCR performance beyond the tested local configuration. Discordant, ambiguous, or high-impact classifications require independent repeat testing or sequencing confirmation. Full article

Background: Falls among older adults are linked to central nervous system deterioration and executive function deficits. Objectives: This study evaluates the feasibility of a gamified EEG-based neurofeedback intervention combined with adaptive working-memory training, targeting neural processes associated with attentional inhibition and working memory in older adults, and characterises, as exploratory secondary outcomes, the oscillatory and functional measures that change over the training period. Methods: Twenty healthy older adults ($65.0\pm 3.3$ years) completed a 24-session longitudinal training programme. The intervention combined real-time alpha-band neurofeedback (NF), targeting left-frontal alpha activity associated with inhibitory control, with an adaptive N-back task engaging theta-band working memory processes. Behavioural outcomes included the Colour Trail Making Test A and B (CTMT-A/B), the Berg Balance Scale short form (BBS-3P), and the four-item Dynamic Gait Index (DGI-4). EEG was recorded at the Early, Middle, and Later stages to characterise training-related neural and behavioural changes. Results: The programme proved highly deliverable: adherence was 100% across all 24 sessions, no session was terminated early, and no severe adverse events occurred. Participants acquired control of the trained signal, with left-frontal alpha power rising across training at the neurofeedback target site, and frontal theta increased bilaterally under adaptive working memory load, consistent with the reduced hemispheric asymmetry characteristic of this age group. Scores changed in the direction of improvement on all four behavioural measures; however, functional changes were small, and cognitive changes cannot be separated from repeated-testing effects. Of sixteen candidate EEG–behaviour associations, four showed moderate-to-large participant-level coefficients and were retained as candidate associations: right-frontal alpha with balance ($r=-0.64$) and with gait adaptability ($r=-0.48$), and bilateral frontal theta with processing speed ($r=-0.53$ and $-0.51$). Several associations that appeared strong when observations were pooled across stages did not survive participant-level modelling, indicating that repeated-measures methods are needed in this literature. All associations are exploratory, uncorrected for multiplicity, and reported with effect sizes and confidence intervals. Conclusions: The gamified neurofeedback and working-memory training programme was feasible, well tolerated, and fully adhered to in a supervised experimental setting by community-dwelling older adults and was accompanied by measurable modulation of the targeted frontal rhythms. This study delivers a shortlist of candidate EEG features, with the effect-size estimates needed to power a confirmatory trial. Because the design is single-arm, both functional scales approached ceiling, and the same test forms were repeated at each stage, practice effects cannot be separated from intervention effects, and these associations are hypothesis-generating; a sham-controlled trial in a fall-prone population is the appropriate next step. Full article

Environmental chemical exposures are increasingly recognized as contributors to metabolic dysfunction, particularly when they occur as chronic, low-dose mixtures rather than as isolated high-dose toxicants. However, current approaches often focus on exposure intensity, single-compound hazard or isolated biomarker associations, and provide limited explanation for why individuals with comparable exposure profiles may develop markedly different metabolic outcomes. This semi-systematic review proposes an Exposure–Adaptive Capacity (EAC) framework to interpret the metabolic consequences of environmental chemical mixtures through the interaction between exposure burden and host adaptive capacity. A structured literature search covered PubMed/MEDLINE, Scopus and Web of Science records published through 30 June 2026; a reviewer-triggered PubMed/MEDLINE update was executed on 30 July 2026 using harmonized British and American dyslipidaemia/dyslipidemia terms, explicit eligibility domains and evidence-mapping procedures. The review integrates epidemiological, mechanistic, toxicological and translational evidence related to environmental chemicals, metabolic dysfunction, mitochondrial impairment, oxidative stress, inflammation, endocrine disruption, metabolic resilience and biomarkers. The evidence indicates that several chemical classes, including per- and polyfluoroalkyl substances, bisphenols, phthalates, pesticides, persistent organic pollutants and selected metals, converge on mitochondrial bioenergetics, redox regulation, inflammatory signalling, endocrine and nuclear receptor activity, nutrient-sensing networks and adipose tissue function. The EAC framework defines exposure burden as the cumulative biological pressure imposed by chemical mixtures and adaptive capacity as the organism’s functional ability to buffer, compensate for or recover from exposure-induced metabolic stress. To make the framework empirically testable, we specify measurable domains for exposure burden, adaptive capacity and EAC mismatch, and distinguish biomarkers of exposure, early biological effect, adaptive capacity, metabolic dysfunction and vulnerability. A quotient-based expression is retained only as a heuristic representation, while empirical testing is proposed through exposure-by-adaptive-capacity interaction models and complementary multidimensional approaches. The framework provides a structured basis for future exposomic, epidemiological and translational studies by shifting attention from exposure alone to the balance between environmental pressure and biological resilience. Full article

The production of hot mix asphalt (HMA) is energy-intensive, resulting in carbon dioxide (CO₂) and greenhouse gas (GHG) emissions. The primary energy source (accounting for over 97%) and emissions source is the rotary drum employed for the drying and heating of the aggregates. Quantifying the CO₂ emissions associated with combustion is of crucial importance in order to facilitate a more profound comprehension of the environmental impacts of HMA production. The objectives of this study are to develop a methodological framework for the quantification of combustion-related carbon dioxide emissions in the context of asphalt production. The proposed framework investigates three complementary approaches: (i) an energy-balance-based thermal energy (TE) model, (ii) recordings of fuel consumption and (iii) direct measurement of exhaust-gas composition. By applying these methods in parallel and cross-comparing their results batch by batch, the framework enables reliable verification of actual CO₂ emissions from the module A3—production stage of asphalt manufacturing. In this stage, the predominant source of greenhouse gases is fuel combustion during aggregate drying and heating. A comprehensive set of data was collected from two HMA batch plants, each operating under distinct conditions. The parameters considered included fuel type, asphalt mixture type, asphalt production time, aggregate moisture content, mixing temperature, and production rate. The TE model demonstrated a robust linear correlation with measured energy consumption (R² = 0.97), and fuel-based CO₂ estimates exhibited minimal discrepancy compared to direct exhaust-gas measurements on average (mean difference 1.0%; t-test p = 0.674). However, systematic discrepancies were observed between the two plants (with overestimation of up to 20% at one plant (AP1) and underestimation of up to 12% at the other (AP2)). This demonstrates that energy-based CO₂ estimation methods require plant-specific calibration against direct measurement before they can be reliably applied in life cycle assessment (LCA) and environmental product declaration (EPD) practice. Measured CO₂ emission intensities ranged from 17.39 to 21.76 kg/t at AP1 and from 16.05 to 18.44 kg/t at AP2; the casing-losses factor of the TE model was calibrated to CL = 23% for the studied diesel-fired plants (mean deviation +0.4% from measured energy); and aggregate moisture content explained 74% of the variance in measured energy consumption (R² = 0.743). Full article

Background/Objectives: Histological biomarkers of therapy response remain incompletely defined in preclinical breast cancer models. We evaluated whether quantitative immunofluorescence imaging and multimodal machine learning (ML) could identify image-derived tissue biomarkers associated with response in a 4T1 murine breast cancer therapy-response setting. Methods: We analysed 1696 immunofluorescence images across nine immunofluorescence staining panels: aPDL1, alpha-smooth muscle actin-CD31, alpha-smooth muscle actin-Ki67, CD3-CD31, CD8-Ki67, collagen–hyaluronic acid (HA), granzyme B-CD8, HMGB1, and pimonidazole/hypoxia. A Python image-analysis algorithm extracted 117 RGB-channel, intensity, morphometric, and cross-channel spatial features per image. The modelling framework included image-only deep learning (DL), tabular ML on extracted histological features, and image-plus-tabular gated fusion. Models were evaluated using stratified cross-validation, train-fold-only class balancing, bootstrap confidence intervals, permutation testing, nested feature-selection sensitivity analysis, and grouped leakage controls. Results: Histological staining panels classified treatment-response status across the nine-panel benchmark. Collagen–HA was the strongest staining (AUC 0.954), followed by alpha-smooth muscle actin-Ki67 (AUC 0.851) and hypoxia (AUC 0.837). DL achieved the best performance in eight of nine stainings. In 346 collagen–HA images, combined collagen and HA features achieved AUC 0.848, collagen-only features AUC 0.833, and hyaluronic-acid-only features AUC 0.805. In 104 animals with matched hypoxia and collagen–hyaluronic-acid images, extracellular-matrix features achieved AUC 0.985, hypoxia-only features achieved AUC 0.929, and adding hypoxia did not improve extracellular-matrix-only prediction. Conclusions: Quantitative extracellular-matrix imaging provides a strong preclinical signal for therapy-response stratification in 4T1 breast cancer. The findings are hypothesis-generating and require independent preclinical and human validation before clinical translation. Full article