National Taipei University of Business

Retinal thickness has been proposed as a non-invasive biomarker of neurodegeneration, whereas sleep disturbances represent modifiable risk factors for cognitive impairment. However, the combined and temporal associations of these factors with cognition remain unclear. This prospective cohort study (2015-2022) included 257 participants aged ≥ 65years. The retinal nerve fiber layer (RNFL) and ganglion cell-inner plexiform layer (GC-IPL) thickness, sleep variable (sleep quality and daytime sleepiness) were assessed at baseline. Cognitive performance, including global and domain-specific measures, was evaluated at baseline and biennially. Longitudinal associations and interaction effects were evaluated using multilevel generalized linear mixed models, with particular attention to quadrant-level variations. Elevated RNFL thickness was associated with better global cognition over time ( : 0.39 × 10⁻2), with the strongest longitudinal positive associations observed in the temporal ( :17.89 × 10-2) and nasal quadrants ( :18.67 × 10-2). GC-IPL thickness showed a similar longitudinal association ( :0.60 × 10⁻2), which became more pronounced in participants with normal sleep quality ( :1.55 × 10-2) or without excessive daytime sleepiness ( :0.81 × 10-2). Furthermore, GC-IPL thickness significantly interacts with sleep quality (pinteration < 0.0001) and excessive daytime sleepiness (pinteration = 0.003). Retinal biomarkers showed quadrant-specific associations with cognition over time, and these associations were modified by sleep quality and daytime sleepiness. These findings underscore the potential of integrating retinal and sleep measures as early predictors of cognitive health, offering opportunities for targeted interventions in older adults.

While visual and textual creativity have been democratized by generative AI, auditory creation remains cognitively and technically opaque for novices. This work presents a workshop study investigating how 40 university freshmen—digital natives but sonic novices—interacted with a publicly accessible generative audio interface powered by OpenAI’s Audio API during a 6-hour hackathon. Through a qualitative analysis of 29 iterative prototyping logs and design journals from 11 teams, we identified three emergent interaction strategies that users employed to navigate the black box constraints of the AI model. First, users adopted Persona-based Control, bypassing acoustic descriptors (e.g., pitch, timbre) in favor of social archetypes. For instance, participants prompted for a Kindergarten Teacher to achieve a specific warm and lively acoustic texture, effectively using social roles as macro-parameters. Second, we observed Algorithmic Tinkering, where users exploited system loopholes to overcome linguistic limitations—most notably, utilizing Romanization (Pinyin) to force an English-optimized model to speak local dialects (Hokkien). Third, participants engaged in Meta-Prompting, utilizing Large Language Models (LLMs) to generate and refine prompts for the audio model, establishing a novice-led Chain-of-AI workflow.The primary contribution of this work is the identification of Semantic Hacking as a user-driven alignment strategy. Our findings suggest that future intelligent audio interfaces should move beyond parameter-tuning to explicitly support metaphorical and persona-based controls, thereby bridging the gap between novice mental models and black-box algorithms.

Extreme rainfall events have become more frequent and intense under climate change, presenting increasing challenges for hydrological monitoring and flood risk management. High-resolution rainfall observations are essential for capturing the spatial and temporal variability of storm events, yet conventional rain-gauge networks suffer from limited spatial coverage and cannot resolve rapidly evolving convective structures. Moreover, high-intensity rainfall events are inherently rare in natural settings, resulting in data gaps in upper rainfall categories. To address this limitation, we integrate natural rainfall observations with controlled artificial rainfall experiments to construct a comprehensive and balanced multi-class dataset covering 0–70 mm/hr at 5 mm/hr intervals. We develop a multimodal deep learning framework that jointly leverages rainfall imagery and acoustic measurements for rainfall-intensity estimation. The two sensing modalities provide complementary physical information: imagery captures streak morphology, drop density, and spatial distribution patterns, while acoustics encode drop momentum, kinetic energy, and impact signatures. Neither modality alone fully characterizes rainfall processes across all intensity ranges; by combining them, the model benefits from richer and more discriminative features. Two-second audio segments are converted into log-mel spectrograms, and a Cross-Attention fusion mechanism enables the network to selectively emphasize the most informative cues from each modality for different rainfall categories. Image-based data augmentation such as horizontal flipping further expands the training space and improves model generalization.Compared with previous studies that relied on single-modality inputs or coarse categorical schemes, our framework achieves a substantially finer classification resolution (0–70 mm/hr in 5-mm/hr bins) and exhibits improved discrimination between adjacent intensity levels. The multimodal architecture consistently outperforms single-modality baselines, with the performance gains being particularly notable in the moderate-to-heavy rainfall range, where the model achieves higher classification accuracy, highlighting the benefits of true cross-modal complementarity. The integration of artificial and natural rainfall further produces a balanced and physically representative dataset that captures both controlled high-intensity scenarios and real-world variability.Overall, this study demonstrates the potential of multimodal sensing and deep learning to advance rainfall monitoring capabilities. The proposed non-contact, low-cost, and high-resolution approach offers a promising pathway for enhancing rainfall observation in regions with sparse gauge coverage, strengthening flood early warning systems, and supporting real-time hydrological applications under a changing climate.

Fuzzy-set qualitative comparative analysis (fsQCA) has gained widespread popularity in social science research. However, incorporating the concept of growth—common in business and economic studies—into fsQCA remains unintuitive and methodologically challenging. This study proposes a new approach to systematically integrate growth into fsQCA applications. The method addresses three key challenges: the naming of antecedents and outcomes, the calibration of variables, and the interpretation of results. An empirical analysis demonstrates that the proposed approach offers a more generalized and robust framework for handling both positive and negative growth values. By applying the method, antecedents and outcomes are appropriately named and calibrated, and the resulting solutions are more interpretable and self-explanatory. Ultimately, the method enhances the clarity and accuracy of fsQCA outcomes, reducing the risk of misinterpretation.

Pancreatic cancer (PC) is frequently missed on unenhanced CT examinations performed for unrelated clinical indications, where the pancreas is included incidentally and clinical suspicion is low. To develop and validate a deep learning-based tool for PC diagnosis and risk stratification on unenhanced CT. This retrospective study included 3080 unenhanced CT studies of Taiwanese patients with PC, other pancreatic diseases and normal pancreas between 2004 and 2019 from a tertiary hospital, randomly divided into training, validation, and internal test sets. Unenhanced CT studies from United States institutions were used for external testing. A hybrid convolutional neural network-transformer model was trained for PC diagnosis and risk stratification. Performance was evaluated using sensitivity, specificity, and area under the curve (AUC), with comparisons to 2 radiologists by McNemar's test and exploratory decision curve analysis. The internal dataset included 713 PCs (mean age, 64.6 ± 12.0 years; 384 men), 1661 normal pancreas and 706 other pancreatic diseases. In an exploratory comparison restricted to unenhanced CT (29 PCs, 31 controls), the sensitivity of computer-aided diagnosis (CAD) tool (89.7%, 72.6-97.8) seemed comparable with that of 1 radiologist (86.2%, 68.3-96.1) and higher than another (41.4%, 23.5-61.1); but wide confidence intervals and inter-radiologist variability warrant cautious interpretation. In the internal test set (142 PCs, 474 controls), sensitivity was 90.8% (84.9-95.0) and specificity 93.0% (90.4-95.2) (AUC: 0.98), with sensitivity comparable to radiologist reports based on enhanced and unenhanced CT (95.4%, 90.2-98.3; P = .21). In the external set (42 PCs, 22 controls), sensitivity was 76.2% (60.5-87.9) and specificity 86.4% (65.1-97.1) (AUC: 0.89). The tool stratified cases into 7 risk levels with likelihood ratios ranging from <0.01 to 173.46. Exploratory decision curve analysis suggested potential net benefit across threshold probabilities. This tool may assist in the opportunistic detection and risk stratification of PC on unenhanced CT.

Hidradenitis suppurativa (HS) is a chronic inflammatory skin disorder associated with systemic comorbidities. Its potential link with urolithiasis, a common urological disease, remains unclear. We conducted a multicenter, retrospective cohort study using the TriNetX US Collaborative Network. Adults ≥18 years with HS diagnoses and more than two visits between 2005 and 2018 were identified and matched 1:1 with non-HS controls using propensity scores. Patients with prior urolithiasis or malignancy were excluded. The final analytic cohort included 79,567 patients with HS and 79,567 controls. Hazard ratios (HRs) with 95% confidence intervals (CIs) were estimated. Sensitivity analyses applied alternative definitions, wash-out periods, and follow-up durations. Comparator groups included psoriasis, rosacea, and androgenic alopecia. Stratified analyses were performed by age and sex. HS was associated with an increased risk of urolithiasis in 15-year follow-up (HR=1.334, 95% CI=1.260-1.413). This association persisted across sensitivity analyses. Comparisons with psoriasis, rosacea, and androgenic alopecia also demonstrated significantly elevated risks (HRs ranging from 1.247 to 1.296). Stratified analyses showed consistent results in both sexes (males: HR=1.340, 95% CI=1.202-1.493; females: HR=1.326, 95% CI=1.214-1.417) and age groups (18-64 years: HR=1.317, 95% CI=1.237-1.402; ≥65 years: HR=1.308, 95% CI=1.160-1.475). Site-specific analyses demonstrated increased risks for renal, ureteric, and lower urinary tract calculi. HS is associated with elevated risk of urolithiasis, consistent across analytic models and subgroups. These findings highlight the existence of overlapping inflammatory pathways and underscore the importance of renal monitoring in HS management.

This study proposes a fast image encryption method for color images, integrating an autoencoder to compress the image and a 6D hyperchaotic system to ensure enhanced security. Initially, a hash value is obtained from the original color image. The hash value, which serves as the secret key of the proposed encryption method, is used to initialize the state variables of the hyperchaotic system, which produces six distinct pseudo-random sequences. The input image is then compressed into a latent image (lossy) using a Vision Transformer Autoencoder model. This latent image is scrambled using chaotic sequences and a Random Shuffle technique. Diffusion is achieved through the Trifid Cipher transformation, which utilizes the remaining chaotic sequences to manipulate pixel values, thereby yielding a cipher version of the latent image. The suggested technique is faster and significantly enhances security compared to the state-of-the-art methods. This method achieves an average entropy of 7.9986, a correlation coefficient close to zero ≈ 0.00004, and key sensitivity analysis gives NPCR = 99.6110% and UACI = 33.4637%. Moreover, the key space of [Formula: see text] confirms that the proposed scheme offers strong resistance against brute-force attacks.