Fair Facial Attribute Classification via Causal Graph-Based Attribute Translation.

Recent studies have raised concerns regarding racial and gender disparity in facial attribute classification performance. As these attributes are directly and indirectly correlated with the sensitive attribute in a complex manner, simple disparate treatment is ineffective in reducing performance disparity. This paper focuses on achieving counterfactual fairness for facial attribute classification. Each labeled input image is used to generate two synthetic replicas: one under factual assumptions about the sensitive attribute and one under counterfactual. The proposed causal graph-based attribute translation generates realistic counterfactual images that consider the complicated causal relationship among the attributes with an encoder-decoder framework. A causal graph represents complex relationships among the attributes and is used to sample factual and counterfactual facial attributes of the given face image. The encoder-decoder architecture translates the given facial image to have sampled factual or counterfactual attributes while preserving its identity. The attribute classifier is trained for fair prediction with counterfactual regularization between factual and corresponding counterfactual translated images. Extensive experimental results on the CelebA dataset demonstrate the effectiveness and interpretability of the proposed learning method for classifying multiple face attributes.

Effect of Intermittent Hypoxia on Metabolic Syndrome and Insulin Resistance in the General Male Population.

Background and objectives: Obstructive sleep apnea (OSA) is closely associated with insulin resistance (IR) and is an independent risk factor for incident type 2 diabetes mellitus (T2DM). Most studies evaluate the correlation between OSA and IR in only obese or T2DM patients. Therefore, we tried to investigate the effect of OSA on metabolic syndrome and IR in the general healthy male population. Materials and Methods: 184 subjects who visited a preventive health examination program were recruited for this study. All subjects received overnight polysomnography by a portable device (Watch-PAT 200). We examined several metabolic parameters and a homeostasis model of assessment for insulin resistance index (HOMA-IR). The subjects were divided into three groups by AHI (Apnea-hyponea index): normal group (AHI < 5), mild OSA group (5 ≤ AHI < 15), and moderate-severe OSA group (AHI ≥ 15). They were also divided into two groups according to minimum oxygen saturation: low group, Min-SpO2 < 88%; and high group, Min-SpO2 ≥ 88%. Results: Parameters of metabolic syndrome, including waist circumference, systolic and diastolic blood pressure, triglyceride, and high-density lipoprotein cholesterol showed significant differences among the AHI groups. Furthermore, HOMA-IR showed significant differences among the AHI groups. Those parameters, including metabolic syndrome and HOMA-IR, also showed differences between Min-SpO2 groups. Conclusions: In summary, this study helps confirm that AHI is associated with HOMA-IR in the general male population. Furthermore, the severity of AHI correlated with the parameters of metabolic syndrome. Therefore, AHI might be an indicator for evaluating both T2DM and metabolic syndrome, even in the general male population.

Craniofacial Phenotype in Obstructive Sleep Apnea and Its Impact on Positive Airway Pressure (PAP) Adherence.

Positive airway pressure (PAP) is an important treatment tool for patients with moderate and severe obstructive sleep apnea (OSA), and adherence to PAP significantly affects treatment outcomes. Disease severity, adverse effects, and psychosocial factors are known to predict medication adherence. Cephalometric parameters have been reported to positively correlate with upper airway collapse. However, research on the correlation between these cephalometric parameters and PAP adherence remains insufficient. This study aimed to identify this relationship. This study included 185 patients with OSA who were prescribed PAP. Polysomnography (PSG) was performed to diagnose OSA, and paranasal sinus computed tomography (PNS CT) was performed to check for comorbidities of the upper airway. In addition, cephalometric parameters such as the hyoid-posterior nasal spine (H-PNS), posterior nasal spine-mandibular plane (PNS-MP), and hyoid-mandibular plane (H-MP) were measured in the midsagittal and axial CT views. Adherence was evaluated 3-12 months after the PAP prescription. A total of 136 patients were PAP-adherent, and 49 were nonadherent. There were more males in the adherent group and a higher average height in the adherent group. The PSG results showed that the apnea-hypopnea index (AHI), respiratory disturbance index (RDI), oxygen desaturation index (ODI), arousal index (AI), rapid eye movement (REM) AHI, and supine AHI were significantly higher, and the lowest oxygen saturation was lower in the adherent group. In the analysis of covariance (ANCOVA) model adjusted for sex and height, among the cephalometric parameters, H-MP was significantly longer in the adherent group (p = 0.027), and H-PNS showed a longer tendency (p = 0.074). In the logistic regression analysis model, the odds ratio (OR) and 95% confidence intervals (95% CI) of adherence and severe OSA in the third tertile compared to the first tertile of H-MP were 2.93 (1.25-6.86) and 4.00 (1.87-8.56). In the case of H-PNS, they were 2.58 (1.14-5.81) and 4.86 (2.24-10.54), respectively. This study concluded that an inferiorly placed hyoid bone in adult patients is associated with PAP adherence and disease severity.

Flexible Piezoelectric Acoustic Sensors and Machine Learning for Speech Processing.

Flexible piezoelectric acoustic sensors have been developed to generate multiple sound signals with high sensitivity, shifting the paradigm of future voice technologies. Speech recognition based on advanced acoustic sensors and optimized machine learning software will play an innovative interface for artificial intelligence (AI) services. Collaboration and novel approaches between both smart sensors and speech algorithms should be attempted to realize a hyperconnected society, which can offer personalized services such as biometric authentication, AI secretaries, and home appliances. Here, representative developments in speech recognition are reviewed in terms of flexible piezoelectric materials, self-powered sensors, machine learning algorithms, and speaker recognition.

Ternary graphene quantum dot-polydopamine-Mn3O4 nanoparticles for optical imaging guided photodynamic therapy and T1-weighted magnetic resonance imaging.

Imaging-guided therapy, which bridges treatment and diagnosis, plays an important role in overcoming the limitations of classical cancer therapy. To provide a more exact location of the tumor and to reduce side effects to normal tissues, a multifunctional probe was designed to serve as both an imaging agent and a therapeutic agent. Ternary hybrid nanoparticles comprised of visible red-responsive graphene, the T1-weighted magnetic resonance imaging (MRI) agent Mn3O4 and a mussel-inspired linker polydopamine. The conjugation of graphene to Mn3O4 through polydopamine enhanced the water solubility of Mn3O4, enabling an efficient uptake by cancer cells as well as tumor accumulation when the nanoparticles were intravenously administered into mice. These nanoparticles, when localized at a tumor site, exhibited low cytotoxicity in the dark, while light irradiation of the cancer cells transfected with the nanoparticles resulted in significant phototherapeutic effects, apparently by generating toxic reactive oxygen species. These nanoparticles also allowed excellent T1-weighted MR imaging in a human lung cancer xenograft model and were successfully used for combined visible red-imaging-guided photodynamic therapy and T1-weighted MRI.