Sensors

Generative artificial intelligence (GenAI) is increasingly used in advertising, yet its effects on visual attention and preference remain insufficiently characterized. This study compared human-created (H), AI-generated (AI), and AI-refined human (HAI) advertisements, the latter obtained by submitting human-created advertisements to a generative model with a standardized refinement prompt, using eye tracking and a forced-choice task. Thirty-nine young adults (18–26 years) with high generative-AI experience completed 858 trials involving 22 advertisement pairs while gaze was recorded at 60 Hz. Fixation duration, fixation count, time to first fixation, saccadic magnitude, and pupil diameter were compared using repeated-measures analyses of participant-level means with Holm correction, and choices were analyzed using binomial models with participant-clustered standard errors. AI-generated advertisements were selected more frequently than human-created advertisements (58.97% of trials; odds ratio = 1.44, $p=0.0015$), whereas HAI advertisements did not differ from human-created advertisements in choice frequency (49.88%; $p=0.978$). Relative to human-created advertisements, both AI and HAI stimuli were associated with higher fixation counts and smaller saccadic magnitudes, with these effects surviving Holm correction. Time to first fixation was statistically equivalent across conditions within the smallest effect detectable by the design, pupil diameter did not differ, and differences in fixation duration did not survive correction. The higher fixation counts reflected a larger share of fixation time on AI and HAI advertisements rather than a higher fixation rate. Although the gaze pattern was similar for AI and HAI stimuli while only AI stimuli were preferred, the difference between the two choice effects was not statistically reliable. The observed gaze differences cannot be uniquely attributed to creative origin because AI and HAI stimuli also exhibited higher edge density than human-created stimuli. These findings support the use of eye tracking as a sensor-based method for characterizing visual responses to AI-assisted advertising while highlighting the need for tighter stimulus control and more diverse samples in future studies.

Myotonometry provides a non-invasive assessment of resting muscle mechanical properties, but its reliability and functional relevance in Parkinson’s disease (PD) remain insufficiently established. This study examined the within-session intra-rater reliability of MyotonPRO measurements and explored their associations with clinical motor severity, gait, functional mobility, and manual dexterity in 24 people with PD assessed in the ON-medication state. Frequency, stiffness, logarithmic decrement, relaxation time, and creep were measured bilaterally in the biceps brachii, triceps brachii, tibialis anterior, and medial gastrocnemius. Reliability between two sets of five impulses was assessed using ICC(3,2), and associations were examined using Spearman correlations with Benjamini–Hochberg false discovery rate correction. Under the standardized same-session, same-rater protocol, all parameters showed good-to-excellent reliability point estimates (ICC = 0.823–0.988), although the precision of some estimates was limited by relatively wide confidence intervals. No significant associations were found with MDS-UPDRS III scores, Hoehn and Yahr stage, or Timed Up and Go performance. Several moderate nominal correlations were observed between lower-limb myotonometric parameters and preferred gait performance, and between triceps brachii parameters and Grooved Pegboard completion time (|r_s| = 0.410–0.488); however, none remained significant after false discovery rate correction. Thus, the reliability findings apply to the present standardized protocol, whereas the associations with motor performance remain exploratory and hypothesis-generating.

The rise of advanced deepfake technologies presents new challenges to media authenticity, making it harder to spot manipulated videos through traditional approaches. Current approaches have a quadratic time complexity $\mathcal{O}\left(n^2\right)$, poor cross-dataset generalization with 20 to 35% accuracy degradation, and inadequate audio–visual fusion mechanisms. This paper presents Mamba-AVGuard, a novel audio–visual deepfake detection framework leveraging Mamba Vision Transformers with selective state-space modeling to achieve linear $\mathcal{O}\left(n\right)$ complexity within the Mamba encoder streams. The novel framework combines a 12-layer visual Mamba encoder for spatial and temporal facial inconsistency analysis with an 8-layer lightweight audio Mamba encoder for prosodic inconsistency analysis, fused via a cross-modal attention block with ±33 ms temporal synchronization. Experiments conducted on two large-scale audio–visual deepfake benchmarks, FakeAVCeleb and AV-Deepfake1M, demonstrate that Mamba-AVGuard outperforms compared methods with 99.92% in-dataset accuracy and 98.47% cross-dataset accuracy on these two benchmarks. The system achieves a processing speed of less than 1.8 s for 30-s videos, with 56% less memory than the best competing transformer-based method, facilitating a throughput of more than 1200 videos per hour for social media verification and forensic workstations.