National Tsing Hua University

To provide reliable, low latency, high data rate connections for 5G mobile networks and beyond, millimeter-wave spectrum is adopted due to its wide available bandwidth. For the mobile fronthauling, a seamless combination of optical fiber and wireless link is an attractive solution. In this article, a millimeter-wave-over-fiber system is demonstrated based on an analog radio-over-fiber architecture that accommodates a variety of modulation formats for different wireless applications and, hence, can be practically transparent to the wireless signal formats. This millimeter-wave-over-fiber heterogeneous system is an efficient remote antenna prototype with a small form factor supporting future wireless networks. The core of the system features high-speed optical components, fiber array, wireless electronic front-end ICs, and patch antenna array. A high-speed Ge photodetector and a SiGe electro-absorption modulator are bonded to a CMOS chip consisting of a transimpedance amplifier, a power amplifier, a low-noise amplifier, and a modulator driver. Considering 28-GHz 16-QAM OFDM signals, the opto-electronic transmitter and receiver achieve an error vector magnitude of 10% and 7.5%, respectively, satisfying 5G requirements.

Identifying deposits of modern/historical and prehistorical tsunamis in Taiwan has been successful in the past two decades and has substantially increased the extant tsunami catalogs, which have been limited in the past four centuries due to scarce and ambiguous historical accounts. In this review, the initiation of the investigation is briefly discussed, partly in response to the latest catastrophic tsunamis in the Indo-Pacific and the stimulated public concern in Taiwan. Major developments and results of the investigation include the onset of the first stage before 2010, with findings in Keelung, the eastern coast, and Lanyu Island, and the second/ongoing stage after 2013, with findings in the northern and eastern coasts and Penghu Islands. These findings contributed to validating the debated historical events, expanding the event number and time span of the tsunami catalog, and elaborating on tsunami processes, which collectively enabled the delineation of the recurrence time intervals between events. Limitations, uncertainty, further contributions, and feedback are discussed including insights into the regional western Pacific hazards of tsunamis, earthquakes, and volcanic eruptions; the principles of recognizing tsunami deposits and processes; and the propositions of future studies and hazard mitigations in Taiwan.

Free oligosaccharides in human milk have many biological functions for infant health. The reducing end of most human milk oligosaccharides is lactose, and caprine milk was reported to contain oligosaccharides structurally similar to those present in human milk. The structures of oligosaccharides were traditionally determined using nuclear magnetic resonance spectroscopy or enzyme digestion followed by various detection methods, e.g., liquid. Mass spectrometry has much higher sensitivity than nuclear magnetic resonance spectroscopy and enzyme digestion. However, conventional mass spectrometry methods only determine part of the structures of oligosaccharides, i.e., compositions and linkage positions. In this study, we used the latest developed mass spectrometry method, namely logically derived sequence tandem mass spectrometry, to determine the complete structures (i.e., composition, linkage positions, anomericities, and stereoisomers) of free neutral trisaccharides in caprine colostrum and mature milk. The high sensitivity of mass spectrometry enables us to discover oligosaccharides of low abundance. Isomers of (Hex)2HexNAc, (Hex)3, and (Hex)2Fuc which have not been reported before were identified. Many of them do not have lactose at the reducing end. Instead, the reducing end is either Glcβ-(1–4)-Glc or Glcβ-(1–4)-GlcNAc. These unusual oligosaccharides are higher in concentration and more structurally diverse in caprine colostrum than that in caprine mature milk and human milk. The structural diversity indicates more complicated biosynthetic pathways of caprine milk compared to that of human milk.

We improve upon previous low-cost radio scintillation monitor designs by implementing onboard processing and display capabilities to produce a more comprehensive device. The proposed design uses a U-blox EVK-6T’s GPS receiver, providing data updates at a rate of 1Hz, and a Raspberry Pi 4. The RPi calculates the S4c index with both a three-minute and a five-minute duration, then saves all the raw data and S4c values onto a locally stored CSV file. The proposed design has been successfully implemented, having been able to log all raw and calculated data and display the three-minute and five-minute S4c values on a plot with a temporal range from 00 UT to 24 UT. In addition to these features, we successfully validate the accuracy of our low-cost radio scintillation monitor by finding a high correlation between its C/No output to that of a high-rate receiver when performing simultaneous measurements from the same receiving antenna. The S4c values calculated from these C/No measurements also proved to be accurate as they demonstrated moderate correlation on non-scintillated days and high correlation on scintillated days.

Purpose Glucose metabolism is associated with several endocrine disorders. Anti-diabetes drugs are crucial in controlling diabetes and its complications; nevertheless, few studies have been carried out involving endocrine function. This study aimed to investigate the association between anti-diabetes drugs and endocrine parameters. Patients and Methods We performed a study of 180 consecutive patients with type 2 diabetes who attended a medical center. Laboratory measurements of metabolic values and endocrine parameters were assessed after a stable treatment regimen of more than 12 weeks. The differences in various endocrine parameters were compared between subjects with or without certain anti-diabetes drugs, with the administrated anti-diabetes drugs being analyzed to find independent risks associated with elevated endocrine parameters. Results After maintaining stable treatment, acceptable glycemic control was noted with an average HbA1c of 7.55% in females and 7.43% in males. Participants taking sulfonylurea (55.8 vs 26.34 ng/L, P=0.043), dipeptidyl peptidase-4 inhibitor (DPP4i) (47.14 vs 32.26 ng/L, P=0.096), or sodium-glucose co-transporter 2 inhibitor (SGLT2i) (64.58 vs 28.11 ng/L, P=0.117) had higher plasma renin concentrations compared to those without this drug but the aldosterone levels did not differ, as well as for other adrenal tests and thyroid function. Under linear regression modeling, SGLT2i was found to be independently associated with a risk of high renin level (beta coefficient: 30.186, 95% confidence interval: 1.71─58.662, P=0.038), whereas sulfonylurea only had borderline associations (B: 21.143, 95% CI: −2.729─45.014, P=0.082). Additionally, renin-angiotensin-aldosterone system (RAAS) blockade (B: 36.728, 95% CI: 12.16─61.295, P=0.004) or diuretics (B: 47.847, 95% CI: 2.039─93.655, P=0.041) was also independently associated with increased renin levels. Conclusion SGLT2i was the only class of anti-diabetes drugs independently associated with elevated renin levels, with results similar to RAAS blockade and diuretics. Although SGLT2i appears to protect reno- and cardio-function, the clinical impact of increased renin warrants further precise study for verification.

Background Early detection of dementia is critical for effective management and possible mitigation of the disease’s progression. Connected speech analysis offers a promising approach to detect early cognitive impairments by evaluating subtle changes in spontaneous language production. This is particularly challenging in low‐resource languages, where linguistic barriers prevent effective communication and analysis in clinical settings. Method We utilized the BERT model, a useful machine learning algorithm, for text classification in a supervised learning setting. Initially, our dataset comprised 80 scripts in Amis, a language with limited resources spoken in rural areas of Eastern Taiwan. To augment our dataset and improve model robustness, we translated English descriptions from picture‐based narrative tasks of dementia patients and cognitively normal individuals into Amis, incorporating these into the training process. Result After incorporating the translated datasets and retraining, the model’s ability to classify dementia and non‐dementia cases from spontaneous speech in Amis improved, reaching an overall predictive accuracy of over 80%. This suggests that data augmentation with translated content can significantly enhance model performance in low‐resource language settings. Conclusion Our findings underscore the potential of using advanced text classification techniques to facilitate early dementia detection in underrepresented linguistic communities. This approach can be potentially replicated across other low‐resource languages, contributing to global health initiatives in dementia care. It highlights the advancements in utilizing machine learning for health interventions in linguistically diverse and resource‐limited populations.

Background Early detection is crucial for the timely intervention and management of dementia, potentially slowing its progression. Early stages of dementia might only subtly affect communication, yet connected speech analysis can detect these minor anomalies. Cognitive tests involving connected speech, like the Boston Diagnostic Aphasia Examination’s (BDAE) “cookie‐theft” picture description task, are pivotal in detecting dementia. However, previous research typically treats their responses as a whole. Method This study preprocesses connected speech from the “cookie theft” task to facilitate AI‐driven detection improvements. We analyzed narratives from 54 normal subjects, 26 with MCI, and 38 with Alzheimer’s Disease (AD), focusing on nine major concepts to quantify accuracy across groups. Result Differences in concept accuracy were significant even among normal individuals. A detailed analysis revealed that concepts 3 and 1 were particularly effective in differentiating between normal individuals and those with cognitive impairments, and between MCI and AD. Concepts 2, 3, and 7 were crucial for distinguishing between normal and MCI subjects. Concepts 4, 5, 8, and 9 showed no discriminatory power. Conclusion Advances in technology, particularly in natural language processing (NLP), allow for the automated analysis of connected speech. It can provide quick and reliable assessments. By categorizing and preprocessing speech data before AI analysis, our approach enhances the model’s predictive accuracy for early‐stage dementia detection. It can also help for developing more diagnostic images and cognitive training exercises. Furthermore, it does not require expensive equipment or invasive procedures, making it accessible for widespread use in various settings, including community centers and primary care, and particularly beneficial for regular monitoring.

Background As global populations age, dementia prevalence is increasing, with projections suggesting significant growth in the number of affected individuals and their caregivers. In Taiwan, family caregivers provide substantial support, often facing intense burdens due to prolonged caregiving duties. This study aims to assess the psychological health of these caregivers to better understand and address their needs. Method This research involved 20 long‐term family caregivers of dementia patients. We used several psychometric instruments to assess their psychological status: the Pittsburgh Sleep Quality Index (PSQI), the Perceived Stress Scale (PSS), and Beck’s Depression Inventory (BDI). Measurements were taken twice, one month apart, to ensure consistency over time. Result The findings indicate severe psychological impacts among caregivers. Ninety percent of participants reported chronic insomnia as indicated by a PSQI score above 5. The PSS results showed that 65% of caregivers experienced abnormal stress levels. Depression was prevalent, with 60% of caregivers showing depressive symptoms; 30% were mildly depressed, 20% moderately, and 10% severely. Conclusion The study highlights the critical psychological toll on family members providing long‐term care to dementia patients. The severity of mental health issues among these caregivers underscores the urgent need for comprehensive support systems that can alleviate their burden and improve their quality of life.

Potassium metal batteries are emerging as a promising high‐energy density storage solution, valued for their cost‐effectiveness and low electrochemical potential. However, understanding the role of potassiphilic sites in nucleation and growth remains challenging. This study introduces a single‐atom iron, coordinated by nitrogen atoms in a 3D hierarchical porous carbon fiber (Fe─N‐PCF), which enhances ion and electron transport, improves nucleation and diffusion kinetics, and reduces energy barriers for potassium deposition. Molten potassium infusion experiments confirm the Fe─N‐PCF's strong potassiphilic properties, accelerating adsorption kinetics and improving potassium deposition performance. According to the Scharifker‐Hills model, traditional carbon fiber substrates without potassiphilic sites cause 3D instantaneous nucleation, leading to dendritic growth. In contrast, the integration of single‐atom and hierarchical porosity promotes uniform 3D progressive nucleation, leading to dense metal deposition, as confirmed by dimensionless i²/imax² versus t/tmax plots and real‐time in situ optical microscopy. Consequently, in situ X‐ray diffraction demonstrated stable potassium cycling for over 1900 h, while the Fe─N‐PCF@K||PTCDA full cell retained 69.7% of its capacity after 2000 cycles (72 mAh g⁻¹), with a low voltage hysteresis of 0.876 V, confirming its strong potential for high energy density and extended cycle life, paving the way for future advancements in energy storage technology.

Pancreatic cystic changes in adults are increasingly identified through advanced cross-sectional imaging. However, the impact of initial/intra-lobular epithelial remodeling on the local β-cell population remains unclear. In this study, we examined 10 human cadaveric donor pancreases (tail and body regions) via integration of stereomicroscopy, clinical H&E histology, and 3D immunohistochemistry, identifying 36 microcysts (size: 1.22±0.56 mm) alongside 54 low-grade pancreatic intraepithelial neoplasias (positive control of epithelial remodeling; size: 2.42±1.05 mm). Both conditions exhibited significant increases in CK7 and insulin immunoreactive signals compared with normal lobules. Importantly, despite luminal contents of microcysts causing false positives (autofluorescence) in fluorescence imaging, the defined cystic epithelium showed distinct duct-β-cell associations—including β-cells in the epithelium and duct-β-cell clusters—visualized via antifade 3D/Airyscan super-resolution imaging in the high-refractive-index polymer. The peri-luminal β-cells displayed insulin+ vesicles residing near the basal domain, while the CK7+ cytokeratins in duct cells accumulated in the apical domain, underlining polarized tissue and cellular organizations. Overall, in microcyst formation, we demonstrate local and associated pancreatic exocrine and endocrine tissue remodeling. Because artifacts are a concern in β-cell investigation in a novel environment, our work using 3D-labeled human pancreas with cytokeratin and vesicle resolving powers provides a robust approach for characterizing the duct-β-cell association in a clinically relevant setting.

Tensor transposition is a fundamental operation in tensor calculations with various applications. However, a naive implementation that copies each element from the source tensor to the transposed position in the target tensor requires double space, making it unsuitable for large-scaled tensors on memory-limited accelerators, like Graphic Processing Units (GPUs). In this paper, we propose an algorithm and its implementation, called EITHOT, for In-place Transposition of High Order Tensors on GPUs, which requires only 5% additional memory at most for large high order tensors. To achieve this, EITHOT uses a newly proposed method, called permutation decomposition, to factorize a transposition of a high-order tensor into a sequence of low-order tensor transpositions. Then, based on the estimated extra memory requirements, EITHOT divides a large tensor into smaller tensors and transposes each smaller tensor separately. Finally, the transposed smaller tensors are combined to form the desired result. The GPU implementation optimizes memory access performance using the cooperative groups programming model. Our experiments demonstrate that EITHOT delivers competitive performance compared to the state-of-the-art out-of-place GPU implementations. Furthermore, EITHOT can handle nearly double the size of tensors compared to out-of-place methods, making it suitable for various transpositions of N-order tensors.

Glioblastoma (GBM), a highly aggressive brain tumor, poses significant treatment challenges due to its highly immunosuppressive microenvironment and the brain immune privilege. Immunotherapy activating the immune system and T lymphocyte infiltration holds great promise against GBM. However, the brain’s low immunogenicity and the difficulty of crossing the blood-brain barrier (BBB) hinder therapeutic efficacy. Recent advancements in immune-actuated particles for targeted drug delivery have shown the potential to overcome these obstacles. These particles interact with the BBB by rapidly and reversibly disrupting its structure, thereby significantly enhancing targeting and penetrating delivery. The BBB targeting also minimizes potential long-term damage. At GBM, the particles demonstrated effective chemotherapy, chemodynamic therapy, photothermal therapy (PTT), photodynamic therapy (PDT), radiotherapy, or magnetotherapy, facilitating tumor disruption and promoting antigen release. Additionally, components of the delivery system retained autologous tumor-associated antigens and presented them to dendritic cells (DCs), ensuring prolonged immune activation. This review explores the immunosuppressive mechanisms of GBM, existing therapeutic strategies, and the role of nanomaterials in enhancing immunotherapy. We also discuss innovative particle-based approaches designed to traverse the BBB by mimicking innate immune functions to improve treatment outcomes for brain tumors. Graphical Abstract

The Corey–Chaykovsky cyclopropanation of readily available masked o‐benzoquinones (MOBs) has been investigated to generate functionalized bicyclo[4.1.0]heptane derivatives with regioselectivity. The resulting bicyclic products were subjected to a sequence of hydrolysis, BF₃ ⋅ Et₂O‐mediated ring‐opening, and then again hydrolysis to access various tropolone derivatives – a class of molecules with significant roles in synthetic, biological, and theoretical chemistry. Utilizing an improved approach based on established cyclopropanation and ring‐opening strategies, several bicyclo[4.1.0]heptanes, tropolone difluorides, and tropolones can be synthesized from readily available MOBs. Our efforts focused on modifying each step to make these reactions general and of considerable practical value.