Terry B. J. Kuo

Empirical mode decomposition (EMD) is an adaptive signal analysis method requiring no predefined basis functions, capable of handling nonlinear and non-stationary signals. However, its high computational cost has long prevented real-time operation on embedded systems, particularly at audio frequencies (8–96 kHz). To the best of our knowledge, this study presents the first real-time audio-frequency EMD system running entirely on a commercial microcontroller (ARM Cortex-M7, STM32H730). The system captures audio via a MEMS microphone, decomposes signals into intrinsic mode functions (IMFs) using fast EMD, applies independent gain to each IMF, and reconstructs the output in real time. Systematic evaluation shows that: (1) the maximum processing rate increases approximately in proportion to the MCU clock frequency, achieving 100 kHz (one IMF) or 40 kHz (four IMFs) at 300 MHz; (2) processing latency is only approximately 1.6 ms (16 kHz, 8 ms window, four IMFs + residual), increasing approximately in proportion to window length; and (3) each IMF corresponds to a distinct frequency range, enabling data-driven frequency-selective processing. Built entirely from commercial off-the-shelf components and powered by two 1.5 V batteries, the system provides initial evidence for transitioning fast EMD from offline analysis to real-time embedded signal processing.

Background: Diabetic retinopathy (DR) is the most common microvascular complication of type 2 diabetes mellitus (T2DM), yet DR screening coverage remains inadequate worldwide. Diabetic autonomic neuropathy typically precedes DR, and heart rate variability (HRV) provides a non-invasive autonomic measure; however, whether HRV can screen for DR remains unexplored. Methods: This pilot study investigated whether sleep-onset minute-scale HRV indices—SDANN1 (standard deviation of 1-min segment means of normal-to-normal intervals) and SDANN2 (2-min segments)—serve as candidate screening biomarkers. Using a public 24-h Holter ECG dataset of T2DM inpatients (N = 39; 11 DR-positive, 28 DR-negative), sleep-onset time-points from cardiopulmonary coupling analysis were used to extract 30 min of post-onset ECG. Results: SDANN1 was significantly lower in the DR group (median 23.6 vs. 44.8 ms, p = 0.028, d = 0.74), with an area under the receiver operating characteristic curve of 0.731 (95% CI: 0.530–0.901). At the optimal threshold of ≤ 25 ms, the odds ratio was 8.1 (95% CI: 1.7–38.4), with DR prevalence of 58% in the high-risk versus 12.5% in the low-risk group. Among 12 candidates, only the SDANN family reached significance. Conclusions: sleep-onset minute-scale HRV shows promise as a candidate risk-stratification biomarker for DR; prospective longitudinal validation in larger, sex-balanced cohorts is required before any screening application.

UNSTRUCTURED Artificial intelligence cannot infer what a clinical environment does not measure. Acute inpatient psychiatry exposes this neglected infrastructure problem because safety, adherence, consent, and workflow constraints exclude much of the conventional sensing design space. We distinguish information handling from information creation and propose intentional sufficiency: first test whether routine care already creates information adequate for the intended use; when it does not, add only the sensing capability needed to create it. We illustrate both routes using an unusual routine nursing sleep record and a ward-specific sensing architecture under development.

Background: Heart rate variability (HRV) is a valuable indicator for assessing vascular health, but keeping clinical decision support systems (CDSSs) aligned with the rapidly evolving literature remains challenging. This study aimed to develop an LLM-assisted literature synthesis framework to support transparent HRV-based risk stratification, enabling systematic extraction and organization of HRV evidence from published studies. Methods: An LLM-driven framework was developed to extract HRV parameters from 140 medical abstracts. The system simulated step-by-step human reasoning to identify key HRV indicators and group patient data using predefined statistical thresholds derived from the literature. System performance was evaluated using ECG-derived HRV features as a feasibility evaluation of literature-guided HRV classification. Results: The proposed framework demonstrated an accuracy of 86% in literature-guided HRV classification, with a sensitivity of 81% and a specificity of 87%. Compared with traditional machine learning approaches, the LLM-assisted system provided transparent, literature-grounded reasoning and could be readily updated as new studies became available. Conclusions: Large language models can support evidence-guided parameter selection and feasibility-level HRV-based risk stratification, rather than serving as predictive classifiers. This approach reduces manual effort, enhances transparency, and addresses common “black box” concerns associated with AI-assisted CDSS development in clinical practice.

Background: Obstructive sleep apnea (OSA) affects millions in Taiwan, but common screening tools, such as oximeters and ApneaLink ® , may reduce sleep quality and have limited accuracy. Methods: We enrolled 277 adults with self-or family-observed snoring. All underwent home sleep apnea testing (HSAT) via ApneaLink ® and simultaneous overnight monitoring with a patch-type heart rate variability (HRV) analyzer. After strict data quality control, 86 subjects remained. HRV indices from ECG signals were processed using time-, frequency-, and nonlinear-domain analyses. An artificial intelligence (AI) model, incorporating a novel Cardiovascular Hypopnea Index (CVHI), was developed using leave-one-out validation. Results: The AI model achieved 81.4% accuracy, outperforming demographic-based (73%) and previous ECG-based (70.6%) screening. At an apnea-hypopnea index (AHI) cutoff of 15, it showed strong classification for moderate-to-severe OSA (AUC >0.8). Conclusion: The patch-type HRV analyzer with AI analysis provides accurate, low-interference OSA screening, suitable for largescale clinical and home use.

Objective This study investigates the relationship between depression, autonomic dysfunction, inhibitory control (IC), and reaction time by analyzing heart rate variability (HRV) during a cognitive task.Methods A total of 29 healthy males and 25 males diagnosed with depression (aged 20–35 years) participated. HRV data were recorded during the Conners Continuous Performance Test-II (CCPT-II) in each group. HRV parameters, including mean RR intervals, standard deviation of normal-to-normal heartbeats (SDNN), low-frequency power with logarithm (lnLF), and high-frequency power with logarithm (lnHF), were analyzed and correlated with IC (d’) and reaction time.Results The depression group exhibited significantly lower lnHF values compared to the healthy group. SDNN and lnLF decreased in both groups during CCPT-II. In the healthy group, d’ correlated significantly with SDNN, lnLF, and lnHF at t14 and across the test duration. However, in the depression group, only RR intervals correlated with d’. A significant correlation between reaction time and HRV was noted at t14 in the healthy group, suggesting autonomic nervous system (ANS) involvement in cognitive performance.Conclusion Reaction time in healthy individuals correlated with ANS function during later stages of CCPT-II, whereas depression disrupted this association. The lower d’ in the depression group was not due to a speed-accuracy trade-off but rather a more pronounced neural network impairment. These findings suggest that depression impairs both IC and autonomic regulation.

We proposed a set of quantitative, inexpensive, and non-invasive methods for screening dementia based on resting EEG. The methods used here included ensemble Holo spectrum (eHolo), ensemble intrinsic Probability Density function (eiPDF), and ensemble intrinsic Multiscale Entropy (eiMSE). These methods were developed originally for physical sciences, specifically to study the wave-turbulence interaction phenomena. Applications, however, were found in neuroscience and bioengineering. Tested on mostly retrospective open-source EEG data, the results measured by receiver operating characteristic (ROC) curves and area under the curve (AUC) values based on micro- and macro-average methods all passed the clinically acceptable threshold of 70%. Specifically, the AUC scores are these: eHolo above 0.89, eiPDF at 0.82, and eiMSE above 0.87, respectively. If we use all three methods in combination, the AUC score is above 0.91. Emulating the bone-density measurement, we have also established a quantitative Z-score to measure the relative standing of each subject. Encouragingly, the Z-score shows a moderate correlation with the widely scattered CDR score, with RHO values between 0.26 and 0.46. We strongly suggest that large-scale clinical trials be organized so that the method can be validated and ready for the WHO and Chinese calls for screening of the general population by 2030.

Objective Executive function correlates with the parasympathetic nervous system (PNS) based on static heart rate variability (HRV) measurements. Our study advances this understanding by employing dynamic assessments of the PNS to explore and quantify its relationship with inhibitory control (IC).Methods We recruited 31 men aged 20–35 years. We monitored their electrocardiogram (ECG) signals during the administration of the Conners’ Continuous Performance Test-II (CCPT-II) on a weekly basis over 2 weeks. HRV analysis was performed on ECG-derived RR intervals using 5-minute windows, each overlapping for the next 4 minutes to establish 1-minute intervals. For each time window, the HRV metrics extracted were: mean RR intervals, standard deviation of NN intervals (SDNN), low-frequency power with logarithm (lnLF), and high-frequency power with logarithm (lnHF). Each value was correlated with detectability and compared to the corresponding baseline value at t0.Results Compared with the baseline level, SDNN and lnLF showed marked decreases during CCPT-II. The mean values of HRV showed significant correlation with d’, including mean SDNN (R=0.474, p=0.012), mean lnLF (R=0.390, p=0.045), and mean lnHF (R=0.400, p=0.032). In the 14th time window, the significant correlations included SDNN (R=0.578, p=0.002), lnLF (R=0.493, p=0.012), and lnHF (R=0.432, p=0.031). Significant correlation between d’ and HRV parameters emerged only during the initial CCPT-II.Conclusion A significant correlation between PNS and IC was observed in the first session alone. The IC in the repeated CCPT-II needs to consider the broader neural network.