Adversarial attacks on medical deep learning models

Introduction: Adversarial attacks are a great threat to deep learning (DL) as they can generate imperceptible perturbations in images which severely affects model performance. More worryingly, recent works have shown that medical DL models are vulnerable to such attacks. The DL process flow is susceptible to various kinds of adversarial attacks. Specifically, causative attacks occur before a model is built, during training, and exploratory attacks occur after model training, during the inference phase. Furthermore, these attacks can be exploited to compromise overall model accuracy, or influence results on specific targeted classes. This research aims to study the impact of causative and exploratory attacks for non-targeted and targeted purposes, on medical DL models built for image classification tasks. Method(s): Warping Based Backdoor Attack and Universal Adversarial Pattern Attack were selected due to their superior performance in generating imperceptible adversarial samples for non-targeted and targeted attacks. DL models were produced from both original and perturbed ISIC-2019 dermoscopic and COVID-NET chest X-ray image datasets. These models were subsequently evaluated on their classification performance. Result(s): Experiments on models achieving above 90.0% accuracy revealed that both causative and exploratory attacks could lower model accuracy by at least 45.0%. In the best-case adversarial attack scenario, model accuracy was reduced by up to 99.0%. Conclusion(s): These results provide a better understanding on the damaging nature of causative and exploratory adversarial attacks as well as vulnerability of medical DL models. The findings can serve as a starting point towards building effective defence approaches that are vital for medical systems utilising DL algorithms.

Toward High-quality Face-Mask Occluded Restoration

Face-mask occluded restoration aims at restoring the masked region of a human face, which has attracted increasing attention in the context of the COVID-19 pandemic. One major challenge of this task is the large visual variance of masks in the real world. To solve it we first construct a large-scale Face-mask Occluded Restoration (FMOR) dataset, which contains 5,500 unmasked images and 5,500 face-mask occluded images with various illuminations, and involves 1,100 subjects of different races, face orientations, and mask types. Moreover, we propose a Face-Mask Occluded Detection and Restoration (FMODR) framework, which can detect face-mask regions with large visual variations and restore them to realistic human faces. In particular, our FMODR contains a self-adaptive contextual attention module specifically designed for this task, which is able to exploit the contextual information and correlations of adjacent pixels for achieving high realism of the restored faces, which are however often neglected in existing contextual attention models. Our framework achieves state-of-the-art results of face restoration on three datasets, including CelebA, AR, and our FMOR datasets. Moreover, experimental results on AR and FMOR datasets demonstrate that our framework can significantly improve masked face recognition and verification performance. © 2023 Association for Computing Machinery.

Tea Culture Tourism Perception: A Study on the Harmony of Importance and Performance

Tea culture tourism is a product of the combination of agricultural tourism and ecotourism. After the COVID-19 period, this product is more and more popular. Tourism performance is an important index for measuring the development level of tourist destinations, and research on the influencing factors of tourism performance is an important way to promote the high-quality development of tea culture tourism. Using the tea tourism town of Wushan as a case study, 452 valid questionnaires were used as research data, and exploratory factor analysis, paired sample t-test and IPA analysis were applied. The results indicate that: (1) tourism performance is mainly divided into 5 dimensions and 22 specific indicators, including service quality, resource environment, tourism transportation, tourism-supporting facilities and tea tourism products;(2) there is a significant difference between the degree of importance and performance of visitors to each indicator, and the overall tourism performance of the case sites at an average level;(3) convenient service, professional service, business management, park traffic, parking conditions, environmental design, shopping environment, tea quality, and tea culture characteristics are potential advantageous factors, and ‘service with a smile', accessibility, trail layout, overall image, air quality, natural scenery, landscape vignettes, network communication, public toilets, sanitation facilities, tourist service centers, tea travel activities, and tourism souvenirs are areas in need of improvement. © 2023 by the authors.

Antibacterial, efficient and sustainable CS/PVA/GA electrospun nanofiber membrane for air filtration

Since the outbreak of the COVID-19 epidemic, a large number of disposable protective masks have been manufactured and used, and the abandonment of masks has caused enormous pollution. In this paper the chitosan (CS), polyvinyl alcohol (PVA), and water were used as raw materials and the nanofiber membranes were prepared by electrostatic spinning. The CS/PVA fiber membranes were crosslinked by glutaraldehyde hydrochloric acid vapour. The fiber morphology, hydrolysis resistance, antibacterial properties, chemical structure, thermal stability and filtration performance of nanofiber membranes were characterized. Results shows that the antibacterial performance of the crosslinked composite nanofibers exceeds 97%, the thermal stability is improved, and the fiber morphology is not destroyed. The hybridized fiber membrane has high filtration performance, excellent antibacterial and hydrolysis resistance, which broadening the PVA fiber membrane application. It is expected to replace traditional protective materials and relieve environmental pressure.

THE USE OF INPATIENT VIRTUAL SHARED MEDICAL APPOINTMENTS FOR PATIENTS WITH ACUTE CEREBROVASCULAR PATHOLOGIES AND THEIR CAREGIVERS: A PILOT STUDY TO IMPROVE HEALTH LITERACY, PATIENT SELF-EFFICACY AND REDUCE PROVIDER BURNOUT

Objective Admission to the hospital for an acute cerebrovascular condition such as stroke or brain hemorrhage can be a traumatic and disorienting experience for patients and their family members. The COVID-19 pandemic has further intensified this experience in addition to exacerbating clinician and resident burnout. To ameliorate some of these concerns, a team of resident and medical student trainees implemented a virtual shared medical appointment (vSMA) program for inpatients with acute cerebrovascular disorders and their caregivers. This study hypothesized that an early intervention vSMA improves patient and caregiver health literacy and preparedness, simultaneously educates trainees on effective communication skills, and reduces clinician burnout. Methods Patients and caregivers of admitted patients were screened through the neurosurgery, neurocritical care, and neurology electronic medical record census. A weekly 60- minute secure virtual session consisted of introductions, a 10- minute standardized presentation on cerebrovascular disease management, followed by participant-guided discussion. Participants completed pre- and post-surveys. We report data on this feasibility study and present challenges, both expected and unforeseen. Results A total of 170 patients were screened;13 patients and 26 caregivers participated in at least 1 session. A total of 6 different healthcare providers facilitated sessions. The vSMA program received overwhelmingly positive feedback from caregivers. Surveys demonstrated 96.4% of caregivers and 75% of patients were satisfied with the session. 96.4% of caregivers and 87.5% of patients would recommend this type of appointment to a friend or family member. 88.8% of providers felt validated by conducting the session. The participant group had a 20% greater percentage of patients discharged home without home needs compared to the non-participant group. The primary obstacle encountered included technological frustrations with the consent process and the sessions themselves. Conclusions Implementation of a vSMA program at a tertiary care center during a pandemic was feasible. Themes caregivers expressed on the post-survey included better understanding of caring for a stroke patient, and coping with the unpredictability of a patient's prognosis. The pandemic has precipitated shifts towards telehealth, but our study highlights the importance of avoiding marginalization of the elderly and less technologically inclined populations (Table Presented).

Policies Implemented in Beijing for Guaranteeing Healthcare for Community-dwelling Patients with Noncommunicable Diseases during the COVID-19 Pandemic

The COVID-19 pandemic brings about influence and challenge for ensuring healthcare services for non-communicable diseases. To guarantee the healthcare services for community-living patients with non-communicable diseases and to meet their healthcare needs, the Beijing municipal government issued a series of policies and relevant supporting measures, including five parts: promoting the implementation of the extended prescription policy, providing Internet-based medical services, further implementing the hierarchical medical system, giving full play to the role of family doctors, and carrying out the service of doorstep delivery of medicines. We reviewed and summarized policies and corresponding measures implemented in Beijing for guaranteeing healthcare for community-dwelling non-communicable disease patients during COVID-19 early response period and ongoing containment period. By evaluating the implementation effect of the policies and comparing with those at home and abroad, it is found that the community chronic disease management under the continuous epidemic situation can be further optimized in the future from the aspects of strengthening the training of grass-roots medical personnel, paying attention to the monitoring of chronic diseases and their risk factors, accurate health management, continuing to implement the hierarchical diagnosis and treatment system, and exploring the whole cycle health management of chronic diseases. Copyright © 2022 by the Chinese General Practice.

Human Placental Hematopoietic Stem Cell Derived Natural Killer Cells (CYNK-001) Mediate Protection Against Influenza A Viral Infection

Background: Influenza A virus (IAV) infections are associated with a high healthcare burden around the world and there is an urgent need to develop more effective therapies. Natural killer (NK) cells provide the first line of innate defense against IAV by killing infected epithelial cells, by producing antiviral cytokines and affecting adaptive immunity. Preclinical studies have demonstrated that NK cells play a pivotal role in reducing IAV-induced pulmonary infection;however, little is known about the therapeutic potential of adoptively transferred NK cells for IAV infections. Celularity Inc. is developing human placental hematopoietic stem cell-derived allogeneic, off-the-shelf NK cell therapy (CYNK-001) for the treatment of viral infections, including coronavirus disease of 2019. Here, we report the evaluation of antiviral activities of CYNK-001 against IAV infection. Methods: In vitro antiviral activities of CYNK-001 were evaluated using human alveolar epithelial cell line A549, infected with IAV strain A/PR/8/34 (H1N1) at variable multiplicity of infection (MOI). The expression of ligands for NK cell receptors was analyzed on infected A549 cells using Fc-coupled recombinant proteins. CYNK-001 was added to A549 cells 16 hours post infection. CYNK-001 degranulation was measured after 4 hours of coculture, and CYNK-001 cytotoxicity against IAV-infected A549 was measured real-time using impedance-based xCELLigence platform. In vivo antiviral and immunomodulatory activities of CYNK-001 were assessed in A/PR/8/34 (H1N1)-induced severe acute lung injury mouse model. Mice were intranasally infected with 2500 PFU IAV. PBS or 1 x 107 CYNK-001 cells were intravenously administered twice at 1 and 3 days post infection (dpi). At 6 dpi, lungs were collected for the evaluation of viral load by qPCR, lung injury and immune cell profiling by histology. Bronchoalveolar lavage fluid (BALF) was collected at 6 dpi for cytokine analysis by multiplex assays, total protein concentration by ELISA and immune cell profiling by flow cytometry. Results: In vitro, IAV infection corresponded with dose-dependent expression of ligands to NK cell-activating receptors, including NKp44, NKp46 and NKG2D. CYNK-001 cells exhibited increased IFNγ, TNFα and GM-CSF production, and elevated level of degranulation upon coculture with IAV-infected A549 cells. Cytokines in culture supernatant and CD107a expression in CYNK-001 cells were upregulated in a virus dose-dependent manner. Consistent with this finding, CYNK-001 cytotoxicity against IAV-infected A549 cells increased from 35% at 0 MOI to 50%, 60% and 75% at 0.001, 0.01 and 0.1 MOI, respectively. These data indicate that CYNK-001 cells recognize virally infected cells, resulting in specific cytotoxic elimination of the source of infection. In vivo, treatment of IAV-infected mice with CYNK-001 reduced weight loss and increased their likelihood of survival. PBS control group developed a severe disease and 37.5% mortality was observed as early as day 4. In the group treated with CYNK-001, disease onset was delayed by 2 days. qPCR analysis of viral RNA showed that CYNK-001-treated mice had lower viral load in the lung than vehicle-treated mice, demonstrating antiviral function of CYNK-001 in vivo. CYNK-001-treated mice had reduced lung injury as assessed by lower total protein concentration in BALF. Moreover, CYNK-001 reduced BALF murine cytokines and chemokines, including IFNγ (p<0.001), IL-6, TNFα, MCP-1 (p<0.05), CXCL2 and CXCL9. Lastly, immunohistochemical analysis of the lung showed that CYNK-001-treated mice had an altered immune response to IAV with higher number of CD68+ macrophages and CD8+ T cells at 6 dpi. Conclusions: Our in vitro and in vivo data show the promising antiviral activities of CYNK-001 against IAV infection. In a severe IAV infection mouse model, CYNK-001 treatment demonstrates lower mortality rate, lower weight loss, lower lung viral load and reduced lung injury along with reduced inflammation. These results support our hypothesis that the adoptive transfer of CYNK-001 cou d reduce the burden of viral infection through the elimination of infected epithelial cells, coordinate a more effective immune response, and result in a clinical benefit in patients with severe viral infection. Disclosures: He: Celularity Inc.: Current Employment. Mahlakõiv: Celularity Inc.: Current Employment. Gleason: Celularity Inc.: Current Employment, Current equity holder in private company. Van Der Touw: Celularity Inc.: Current Employment. Kang: Celularity Inc.: Current Employment. Hariri: Celularity Inc.: Current Employment, Current equity holder in private company. Zhang: Celularity Inc.: Current Employment, Current equity holder in private company.

Ventilation and air cleaning to limit aerosol particle concentrations in a gym during the COVID-19 pandemic.

SARS-CoV-2 can spread by close contact through large droplet spray and indirect contact via contaminated objects. There is mounting evidence that it can also be transmitted by inhalation of infected saliva aerosol particles. These particles are generated when breathing, talking, laughing, coughing or sneezing. It can be assumed that aerosol particle concentrations should be kept low in order to minimize the potential risk of airborne virus transmission. This paper presents measurements of aerosol particle concentrations in a gym, where saliva aerosol production is pronounced. 35 test persons performed physical exercise and aerosol particle concentrations, CO2 concentrations, air temperature and relative humidity were obtained in the room of 886 m³. A separate test was used to discriminate between human endogenous and exogenous aerosol particles. Aerosol particle removal by mechanical ventilation and mobile air cleaning units was measured. The gym test showed that ventilation with air-change rate ACH = 2.2 h-1, i.e. 4.5 times the minimum of the Dutch Building Code, was insufficient to stop the significant aerosol concentration rise over 30 min. Air cleaning alone with ACH = 1.39 h-1 had a similar effect as ventilation alone. Simplified mathematical models were engaged to provide further insight into ventilation, air cleaning and deposition. It was shown that combining the above-mentioned ventilation and air cleaning can reduce aerosol particle concentrations with 80 to 90% , depending on aerosol size. This combination of existing ventilation supplemented with air cleaning is energy efficient and can also be applied for other indoor environments.

Visual Analysis Method for COVID-19 Epidemic Situation

Visualization analysis of the COVID-19 epidemic data can directly display epidemic dynamics, explore the epidemic spreading rule, and predict the trend of epidemic development. Based on the COVID-19 epidemic data obtained by public channels such as Ding Xiang Yuan and ministries and health committees, According to the multi-dimensional spatial and temporal characteristics of epidemic data, three data sets are designed, namely, case number data set, case source data set and case relationship data set. A COVID-19 epidemic visualization model was proposed based on data preprocessing and comprehensive analysis of time axis interaction and SEIR model. Progressive epidemic analysis was used to visualize the epidemic situation of typical infectious diseases. Take Henan Province as an example, the epidemic situation of COVID-19 is revealed, the source characteristics of COVID-19 are excavated, the epidemic pattern of COVID-19 is summarized, and the future trend of COVID-19 is predicted. © 2020, Beijing China Science Journal Publishing Co. Ltd. All right reserved.