Integrating humanities in healthcare: a mixed-methods study for development and testing of a humanities curriculum for front-line health workers in Karachi, Pakistan.

Lady health workers (LHWs) provide lifesaving maternal and child health services to >60% of Pakistan's population but are poorly compensated and overburdened. Moreover, LHWs' training does not incorporate efforts to nurture attributes necessary for equitable and holistic healthcare delivery. We developed an interdisciplinary humanities curriculum, deriving its strengths from local art and literature, to enhance character virtues such as empathy and connection, interpersonal communication skills, compassion and purpose among LHWs. We tested the curriculum's feasibility and impact to enhance character strengths among LHWs.We conducted a multiphase mixed-methods pilot study in two towns of Karachi, Pakistan. We delivered the humanities curriculum to 48 LHWs via 12 weekly sessions, from 15 June to 2 September 2021. We developed a multiconstruct character strength survey that was administered preintervention and postintervention to assess the impact of the training. In-depth interviews were conducted with a subset of randomly selected participating LHWs.Of 48 participants, 47 (98%) completed the training, and 34 (71%) attended all 12 sessions. Scores for all outcomes increased between baseline and endline, with highest increase (10.0 points, 95% CI 2.91 to 17.02; p=0.006) observed for empathy/connection. LHWs provided positive feedback on the training and its impact in terms of improving their confidence, empathy/connection and ability to communicate with clients. Participants also rated the sessions highly in terms of the content's usefulness (mean: 9.7/10; SD: 0.16), the success of the sessions (mean: 9.7/10; SD: 0.17) and overall satisfaction (mean: 8.2/10; SD: 3.3).A humanities-based training for front-line health workers is a feasible intervention with demonstrated impact of nurturing key character strengths, notably empathy/connection and interpersonal communication. Evidence from this study highlights the value of a humanities-based training, grounded in local literature and cultural values, that can ultimately translate to improved well-being of LHWs thus contributing to better health outcomes among the populations they serve.

Digitalization of routine health information systems: Bangladesh, Indonesia, Pakistan.

Objective: To describe a systematic process of transforming paper registers into a digital system optimized to enhance service provision and fulfil reporting requirements. Methods: We designed a formative study around primary health workers providing reproductive, maternal, newborn and child health services in three countries in Bangladesh, Indonesia and Pakistan. The study ran from November 2014 to June 2018. We developed a prototype digital application after conducting a needs assessment of health workers' responsibilities, workflows, routine data requirements and service delivery needs. Methods included desk reviews, focus group discussions, in-depth interviews; data mapping of paper registers; observations of health workers; co-design workshops with health workers; and usability testing. Finally, we conducted an observational feasibility assessment to monitor uptake of the application. Findings: Researchers reviewed a total of 17 paper registers across the sites, which we transformed into seven modules within a digital application running on mobile devices. Modules corresponded to the services provided, including household enumeration, antenatal care, family planning, immunization, nutrition and child health. A total of 65 health workers used the modules during the feasibility assessment, and average weekly form submissions ranged from 8 to 234, depending on the health worker and their responsibilities. We also observed variability in the use of modules, requiring consistent monitoring support for health workers. Conclusion: Lessons learnt from this study shaped key global initiatives and resulted in a software global good. The deployment of digital systems requires well-designed applications, change management and strengthening human resources to realize and sustain health system gains.

Assessment of vaccination service delivery and quality: a cross-sectional survey of over 1300 health facilities from 29 districts in Sindh, Pakistan conducted between 2017-18.

BACKGROUND: Routine childhood immunization coverage in Pakistan remains sub-par, in part, due to suboptimal utilization of existing vaccination services. Quality of vaccine delivery can affect both supply and demand for immunization, but data for immunization center quality in Pakistan is sparse and in Sindh province in Southern Pakistan, no comprehensive health facility assessment has ever been conducted at a provincial level. We assessed health facilities, specifically immunization centers, and their associated health workers throughout the province to summarize quality of immunization centers.  METHODS: An exhaustive list of health facilities obtained from Sindh's provincial government was included in our analysis, comprising a total of 1396 public, private, and public-private health facilities. We adapted a health facility and health worker assessment survey developed by BASICS and EPI-Sindh to record indicators pertaining to health facility infrastructure, processes and human resources. Using expert panel ranking, we developed critical criteria (the presence of a cold box/refrigerator, vaccinator and vaccination equipment at the immunization center) to indicate the bare minimum items required by immunization centers to vaccinate children. We also categorized other infrastructure, process, and human resource items to determine high, low and moderate function requirements to ascertain quality. We evaluated presence of critical criteria, calculated scores for high, moderate and low function requirements, and displayed frequencies of infrastructure, process and human resource indicators for all immunization centers across Sindh. We analyzed results at the division level and utilized a two-sample independent clustered t-test to test differences in average function requirement scores between facilities that met critical criteria and those that did not. RESULTS: Out of the 1396 health facilities assessed across Sindh province from October 2017 to January 2018, 1236 (88.5%) were operational while 1209 (86.6%) offered vaccination services (immunization centers). Only 793 (65.6%; 793/1209) immunization centers met the critical criteria of having all the following items: vaccinator, a cold box or refrigerator and vaccine supplies. Of the 416 (34.4%; 416/1209) immunization centers that did not meet the critical criteria, most of the centers did not have a cold box or refrigerator (28.3%; 342/1209), followed by lack of vaccines (19.9%; 240/1209), and a vaccinator (13.0%; 157/1209). Of the 2153 healthcare workers interviewed, 1875 (87.1%) were vaccinators, of which 1745 (81.0%; 1745/2153) were male, and had an average of 12.4 years of schooling. A total of 1805 (96.3%; 1805/1875), 1655 (88.3%; 1655/1875) and 1387 (74.0%; 1387/1875) of the vaccinators were trained in vaccination, cold chain and inventory management respectively. CONCLUSION: One out of three immunization centers in Sindh lack the critical components essential for quality vaccination services. While the majority of health workers (>80%) were trained on vaccination and cold chain management, the proportion trained on inventory management was comparatively low. Our findings therefore suggest that suboptimal immunization center quality is partly due to inadequate infrastructure and inefficient processes contributed to an extent, by low levels of inventory management training among vaccinators. Our study presents critical research findings with high-impact policy implications for identifying and addressing gaps to improve vaccination uptake within a low-middle income country setting.

Effect of vaccine reminder and tracker bracelets on routine childhood immunization coverage and timeliness in urban Pakistan (2017-18): a randomized controlled trial.

BACKGROUND: Inability to track children's vaccination history coupled with parents' lack of awareness of vaccination due dates compounds the problem of low immunization coverage and timeliness in developing countries. We evaluated the impact of two types of silicone immunization reminder bracelets for children in improving immunization coverage and timeliness of Pentavalent-3 and the Measles-1 vaccines. METHODS: Children < 3 months were enrolled in either of the 2 intervention groups (Alma Sana Bracelet Group and Star Bracelet Group) or the Control group. Children in the intervention groups were provided the two different bracelets at the time of recruitment. Each time the child visited the immunization center, a hole was perforated in the silicone bracelet to denote vaccine administration. Each child was followed up till administration of Measles-1 vaccine or till 12 months of age (if they did not come to the center for vaccination). Data was analyzed using the intention-to-treat population between groups. The unadjusted and adjusted Risk Ratios (RR) and 95% confidence interval (CI) for Pentavalent-3 and Measles-1 coverage at 12 months of age were estimated through bivariate and multivariate analysis. Time-to-Pentavalent-3 and Measles-1 immunization curves were calculated using the Kaplan-Meier method. RESULTS: A total of 1,445 children were enrolled in the study between July 19, 2017 and October 10, 2017. Baseline characteristics among the three groups were similar. Up-to-date coverage for the Pentavalent-3 /Measles-1 vaccine at 12 months of age was 84.6%/72.0%, 85.4%/70.5% and 83.0%/68.5% in Alma Sana Bracelet group, Star Bracelet group and Control group respectively but the differences were not statistically significant. In the multivariate analysis, neither the Alma Sana bracelet (adjusted RR = 1.01; 95% CI: 0.96-1.06), (adjusted RR: 1.05; 95% CI: 0.97-1.13) nor the Star bracelet (adjusted RR = 1.01; 95% CI: 0.96-1.06) (adjusted RR: 1.03; 95% CI: 0.95-1.11) was significantly associated with Pentavalent-3 vaccination or Measles-1 vaccination. CONCLUSION: Although we did not observe any significant impact of the bracelets on improved immunization coverage and timeliness, our findings add to the existing literature on innovative, low cost reminders for health and make several suggestions for enhancing practical implementation of these tools. TRIAL REGISTRATION: ClinicalTrials.gov NCT03310762 . Retrospectively Registered on October 16, 2017.

Using Serosurvey Data Triangulation for More Accurate Estimates of Vaccine Coverage: Measured and Modeled Coverage From Pakistan Household Surveys.

Household surveys remain an essential method for estimating vaccine coverage in developing countries. However, the resulting estimates have inevitable and currently unmeasurable information biases due to inaccuracies in recall, low retention of home-based records (HBRs; i.e., vaccination cards), and inaccurate recording of vaccination on HBRs. We developed an innovative method with which to overcome these biases, enhance the validity of survey results, and estimate true vaccine coverage using nested serological assessments of immune markers. We enrolled children aged 12-23 months in vaccine coverage surveys in Karachi, Pakistan, from January to December 2016. Vaccination history was collected through verbal recall by the caregiver and, when available, by HBR. One-third of survey participants were randomly enrolled for serological testing for anti-measles virus immunoglobulin G antibody. We applied Bayesian latent class models to evaluate the misalignment among measles vaccination histories derived by recall, HBRs, and measles serology and estimated true measles vaccine coverage. The model-based estimate of true measles vaccine coverage was 61.1% (95% credible interval: 53.5, 69.4) among all survey participants. The standard estimate of 73.2% (95% confidence interval: 71.3, 75.1) defined by positive recall or HBR documentation substantially overestimated the vaccine coverage. Researchers can correct for information biases using serological assessments in a subsample of survey participants and latent class analytical approaches.

Effect of vaccine reminder and tracker bracelets on routine childhood immunization coverage and timeliness in urban Pakistan: protocol for a randomized controlled trial.

BACKGROUND: Inability to track children's vaccination history coupled with parents' lack of awareness of vaccination due dates compounds the problem of low immunization coverage and timeliness in developing countries. Traditional Reminder/Recall (RR) interventions such as paper-based immunization cards or mHealth based platforms do not yield optimal results in resource-constrained settings. There is thus a need for a low-cost intervention that can simultaneously stimulate demand and track immunization history to help reduce drop-outs and improve immunization coverage and timeliness. The objective of this study is to evaluate the impact of low-cost vaccine reminder and tracker bracelets for improving routine childhood immunization coverage and timeliness in Pakistani children under 2 years of age. METHODS: The study is an individually randomized, three-arm parallel Randomized Controlled Trial with two intervention groups and one control group. Infants in the two intervention groups will be given two different types of silicone bracelets at the time of recruitment, while infants in the control group will not receive any intervention. The two types of bracelets consist of symbols and/or numbers to denote the EPI vaccination schedule and each time the child will come for vaccination, the study staff will perforate a hole in the appropriate symbol to denote vaccine administration. Therefore, by looking at the bracelet, caregivers will be able to see how many vaccines have been received. Our primary outcome measure is the increase in coverage and timeliness of Pentavalent-3/PCV-3/Polio-3 and Measles-1 vaccine in the intervention versus control groups. A total of 1446 participants will be recruited from 4 Expanded Program on Immunization (EPI) centers in Landhi Town, Karachi. Each enrolled child will be followed up till the Measles-1 vaccine is administered, or till eleven months have elapsed since enrolment. DISCUSSION: Participant recruitment commenced on July 19, 2017, and was completed on October 10, 2017. Proposed duration of the study is 18 months and expected end date is December 1, 2018. This study constitutes one of the first attempts to rigorously evaluate an innovative, low-cost vaccine reminder bracelet. TRIAL REGISTRATION: ClinicalTrials.gov NCT03310762 . Retrospectively Registered on October 16, 2017.

The effects of experiment duration and supertrial analysis on EEG classification methods.

Bharadwaj et al. [1] present a comments paper evaluating the classification accuracy of several state-of-the-art methods using EEG data averaged over random class samples. According to the results, some of the methods achieve above-chance accuracy, while the method proposed in [2], that is the target of their analysis, does not. In this rebuttal, we address these claims and explain why they are not grounded in the cognitive neuroscience literature, and why the evaluation procedure is ineffective and unfair.

Single Stage Adaptive Multi-Attention Network for Image Restoration.

Recently attention-based networks have been successful for image restoration tasks. However, existing methods are either computationally expensive or have limited receptive fields, adding constraints to the model. They are also less resilient in spatial and contextual aspects and lack pixel-to-pixel correspondence, which may degrade feature representations. In this paper, we propose a novel and computationally efficient architecture Single Stage Adaptive Multi-Attention Network (SSAMAN) for image restoration tasks, particularly for image denoising and image deblurring. SSAMAN efficiently addresses computational challenges and expands receptive fields, enhancing robustness in spatial and contextual feature representation. Its Adaptive Multi-Attention Module (AMAM), which consists of Adaptive Pixel Attention Branch (APAB) and an Adaptive Channel Attention Branch (ACAB), uniquely integrates channel and pixel-wise dimensions, significantly improving sensitivity to edges, shapes, and textures. We perform extensive experiments and ablation studies to validate the performance of SSAMAN. Our model shows state-of-the-art results on various benchmarks, for example, on image denoising tasks, SSAMAN achieves a notable 40.08 dB PSNR on SIDD dataset, outperforming Restormer by 0.06 dB PSNR, with 41.02% less computational cost, and achieves a 40.05 dB PSNR on the DND dataset. For image deblurring, SSAMAN achieves 33.53 dB PSNR on GoPro dataset. Code and models are available at Github.

Evaluating an immunization carpool service for women in rural areas for facilitating routine childhood immunizations in Pakistan -a feasibility study on acceptability, demand, and implementation.

Introduction: Poor accessibility of immunization services coupled with limited options for transportation and socio-cultural norms that hinder women's mobility are among the key factors contributing to poor immunization coverage in rural areas. We assessed the feasibility and acceptability of establishing a free-of-cost, women-only carpool service for immunization in a rural setting in Pakistan and evaluated its preliminary impact on immunization coverage and timeliness among children. Methods: We conducted a feasibility study in four selected immunization facilities in Shikarpur District, Sindh. A local transport vehicle was hired and branded as an immunization carpool service. Women having un- or under-immunized children aged ≤2 years were invited to visit immunization facilities using carpool vehicles. Information on demographic indicators and service experience was collected. Child immunization details were extracted using the government's provincial electronic immunization registry to estimate immunization coverage and timeliness. Results: Between January and October 2020, six immunization carpool vehicles provided uninterrupted service and transported 2422 women-child pairs, completing 4691 immunization visits. Majority of women reported that the carpool service improved accessibility (99.6%) by offering group travel (82.9%) and reducing their dependency on family members (93.4%). Preliminary estimates reported an increase in immunization coverage and timeliness across antigens among participating children compared to non-participating children, with significant increase in proportion for BCG coverage (38.1%; p < 0.001, CI: 32.8%, 43.4%) and measles-2 timeliness (18%; p < 0.001, CI: 13.3%, 22.4%). Conclusion: A women-only immunization carpool service implemented within a rural setting is feasible and highly acceptable. Key factors contributing to the model's success include increased mobility and independence of women, cost-savings, and a culturally and contextually appropriate mechanism of transport embedded within the local setting. Increased accessibility to health services also contributed to improved immunization coverage and timeliness among children.

DFT, GC-MS analysis and biological evaluation of Limbarda crithmoides L. Dumort essential oil; an important edible halophyte grown in Pakistan.

Antimicrobial resistance is a major health burden in Pakistan, and therefore new herbal medicine-based therapeutic regimens are being largely investigated. Limbarda crithmoides essential oil was extracted by using hydrodistillation method. Chemical profiling of essential was evaluated by using FTIR and GC-MS analysis. A total of 20 components were identified including, p-xylene, o-xylene, ß-linalool, acetophenole and 3-isopropylphenyl methylcarbamate. The HOMO and LUMO analysis in DFT investigations presented that 3-isopropylphenyl methylcarbamate, p-xylene and o-xylene posess a substantial capacity to transfer charge through molecules. The antimicrobial potential of essential oil showed moderate inhibition against E. coli (MIC = 6.25 mg/mL), whereras a significant inhibition Staphylococos aureus was recorded (MIC = 3.12 mg/mL). Further, significant antioxidant activities were recorded in DPPH radical scavenging (IC50 = 80.5 µg/mL), H2O2 (64 ± 1.2%) and FRAP (60.3 µg ferrous equivalents) assays. It was therefore concluded that Limbarda crithmoides essential oil has potential antioxidant and anti-antimicrobial properties and can be used for further investigations.

Development and feasibility testing of an artificially intelligent chatbot to answer immunization-related queries of caregivers in Pakistan: A mixed-methods study.

BACKGROUND: Gaps in information access impede immunization uptake, especially in low-resource settings where cutting-edge and innovative digital interventions are limited given the digital inequity. Our objective was to develop an Artificially Intelligent (AI) chatbot to respond to caregiver's immunization-related queries in Pakistan and investigate its feasibility and acceptability in a low-resource, low-literacy setting. METHODS: We developed Bablibot (Babybot), a local language immunization chatbot, using Natural Language Processing (NLP) and Machine Learning (ML) technologies with Human in the Loop feature. We evaluated the bot through a sequential mixed-methods study. We enrolled caregivers visiting the 12 selected immunization centers for routine childhood vaccines. Additional caregivers were reached through targeted text message communication. We assessed Bablibot's feasibility and acceptability by tracking user engagement and technological metrics, and through thematic analysis of in-depth interviews with 20 caregivers. FINDINGS: Between March 9, 2020, and April 15, 2021, 2,202 caregivers were enrolled in the study, of which, 677 (30.7%) interacted with Bablibot (users). Bablibot responded to 1,877 messages through 874 conversations. Conversation topics included vaccination due dates (32.4%; 283/874), side-effect management (15.7%;137/874), or delaying vaccination due to child's illness or COVID-lockdown (16.8%;147/874). Over 90% (277/307) of responses to text-based exit surveys indicated satisfaction with Bablibot. Qualitative analysis showed caregivers appreciated Bablibot's usefulness and provided feedback for further improvement of the system. CONCLUSION: Our results demonstrate the feasibility and acceptability of local-language NLP chatbots in providing real-time immunization information in low-resource settings. Text-based chatbots canminimize the workload on helpline operators, in addition to instantaneously resolving caregiver queries that otherwise lead to delay or default.

Exploiting the Brain's Network Structure for Automatic Identification of ADHD Subjects.

Attention Deficit Hyperactive Disorder (ADHD) is a common behavioral problem affecting children. In this work, we investigate the automatic classification of ADHD subjects using the resting state Functional Magnetic Resonance Imaging (fMRI) sequences of the brain. We show that the brain can be modeled as a functional network, and certain properties of the networks differ in ADHD subjects from control subjects. We compute the pairwise correlation of brain voxels' activity over the time frame of the experimental protocol which helps to model the function of a brain as a network. Different network features are computed for each of the voxels constructing the network. The concatenation of the network features of all the voxels in a brain serves as the feature vector. Feature vectors from a set of subjects are then used to train a PCA-LDA (principal component analysis-linear discriminant analysis) based classifier. We hypothesized that ADHD-related differences lie in some specific regions of the brain and using features only from those regions is sufficient to discriminate ADHD and control subjects. We propose a method to create a brain mask that includes the useful regions only and demonstrate that using the feature from the masked regions improves classification accuracy on the test data set. We train our classifier with 776 subjects and test on 171 subjects provided by The Neuro Bureau for the ADHD-200 challenge. We demonstrate the utility of graph-motif features, specifically the maps that represent the frequency of participation of voxels in network cycles of length 3. The best classification performance (69.59%) is achieved using 3-cycle map features with masking. Our proposed approach holds promise in being able to diagnose and understand the disorder.

Mesh Convolution With Continuous Filters for 3-D Surface Parsing.

Geometric feature learning for 3-D surfaces is critical for many applications in computer graphics and 3-D vision. However, deep learning currently lags in hierarchical modeling of 3-D surfaces due to the lack of required operations and/or their efficient implementations. In this article, we propose a series of modular operations for effective geometric feature learning from 3-D triangle meshes. These operations include novel mesh convolutions, efficient mesh decimation, and associated mesh (un)poolings. Our mesh convolutions exploit spherical harmonics as orthonormal bases to create continuous convolutional filters. The mesh decimation module is graphics processing unit (GPU)-accelerated and able to process batched meshes on-the-fly, while the (un)pooling operations compute features for upsampled/downsampled meshes. We provide an open-source implementation of these operations, collectively termed Picasso. Picasso supports heterogeneous mesh batching and processing. Leveraging its modular operations, we further contribute a novel hierarchical neural network for perceptual parsing of 3-D surfaces, named PicassoNet ++ . It achieves highly competitive performance for shape analysis and scene segmentation on prominent 3-D benchmarks. The code, data, and trained models are available at https://github.com/EnyaHermite/Picasso.

Diffusion Models in Vision: A Survey.

Denoising diffusion models represent a recent emerging topic in computer vision, demonstrating remarkable results in the area of generative modeling. A diffusion model is a deep generative model that is based on two stages, a forward diffusion stage and a reverse diffusion stage. In the forward diffusion stage, the input data is gradually perturbed over several steps by adding Gaussian noise. In the reverse stage, a model is tasked at recovering the original input data by learning to gradually reverse the diffusion process, step by step. Diffusion models are widely appreciated for the quality and diversity of the generated samples, despite their known computational burdens, i.e., low speeds due to the high number of steps involved during sampling. In this survey, we provide a comprehensive review of articles on denoising diffusion models applied in vision, comprising both theoretical and practical contributions in the field. First, we identify and present three generic diffusion modeling frameworks, which are based on denoising diffusion probabilistic models, noise conditioned score networks, and stochastic differential equations. We further discuss the relations between diffusion models and other deep generative models, including variational auto-encoders, generative adversarial networks, energy-based models, autoregressive models and normalizing flows. Then, we introduce a multi-perspective categorization of diffusion models applied in computer vision. Finally, we illustrate the current limitations of diffusion models and envision some interesting directions for future research.

Evaluation of a Mobile-Based Immunization Decision Support System for Scheduling Age-Appropriate Vaccine Schedules for Children Younger Than 2 Years in Pakistan and Bangladesh: Lessons From a Multisite, Mixed Methods Study.

BACKGROUND: Missed opportunities for vaccination (MOVs), that is, when children interact with the health system but fail to receive age-eligible vaccines, pose a crucial challenge for equitable and universal immunization coverage. Inaccurate interpretations of complex catch-up schedules by health workers contribute to MOVs. OBJECTIVE: We assessed the feasibility of a mobile-based immunization decision support system (iDSS) to automatically construct age-appropriate vaccination schedules for children and to prevent MOVs. METHODS: A sequential exploratory mixed methods study was conducted at 6 immunization centers in Pakistan and Bangladesh. An android-based iDSS that is packaged in the form of an application programming interface constructed age-appropriate immunization schedules for eligible children. The diagnostic accuracy of the iDSS was measured by comparing the schedules constructed by the iDSS with the gold standard of evaluation (World Health Organization-recommended Expanded Programme on Immunization schedule constructed by a vaccines expert). Preliminary estimates were collected on the number of MOVs among visiting children (caused by inaccurate vaccination scheduling by vaccinators) that could be reduced through iDSS by comparing the manual schedules constructed by vaccinators with the gold standard. Finally, the vaccinators' understanding, perceived usability, and acceptability of the iDSS were determined through interviews with key informants. RESULTS: From July 5, 2019, to April 11, 2020, a total of 6241 immunization visits were recorded from 4613 eligible children. Data were collected for 17,961 immunization doses for all antigens. The iDSS correctly scheduled 99.8% (17,932/17,961) of all age-appropriate immunization doses compared with the gold standard. In comparison, vaccinators correctly scheduled 96.8% (17,378/17,961) of all immunization doses. A total of 3.2% (583/17,961) of all due doses (across antigens) were missed in age-eligible children by the vaccinators across both countries. Vaccinators reported positively on the usefulness of iDSS, as well as the understanding and benefits of the technology. CONCLUSIONS: This study demonstrated the feasibility of a mobile-based iDSS to accurately construct age-appropriate vaccination schedules for children aged 0 to 23 months across multicountry and low- and middle-income country settings, and underscores its potential to increase immunization coverage and timeliness by eliminating MOVs.

MutualNet: Adaptive ConvNet via Mutual Learning From Different Model Configurations.

Most existing deep neural networks are static, which means they can only perform inference at a fixed complexity. But the resource budget can vary substantially across different devices. Even on a single device, the affordable budget can change with different scenarios, and repeatedly training networks for each required budget would be incredibly expensive. Therefore, in this work, we propose a general method called MutualNet to train a single network that can run at a diverse set of resource constraints. Our method trains a cohort of model configurations with various network widths and input resolutions. This mutual learning scheme not only allows the model to run at different width-resolution configurations but also transfers the unique knowledge among these configurations, helping the model to learn stronger representations overall. MutualNet is a general training methodology that can be applied to various network structures (e.g., 2D networks: MobileNets, ResNet, 3D networks: SlowFast, X3D) and various tasks (e.g., image classification, object detection, segmentation, and action recognition), and is demonstrated to achieve consistent improvements on a variety of datasets. Since we only train the model once, it also greatly reduces the training cost compared to independently training several models. Surprisingly, MutualNet can also be used to significantly boost the performance of a single network, if dynamic resource constraints are not a concern. In summary, MutualNet is a unified method for both static and adaptive, 2D and 3D networks. Code and pre-trained models are available at https://github.com/taoyang1122/MutualNet.

Immunization Gender Inequity in Pakistan: An Analysis of 6.2 Million Children Born from 2019 to 2022 and Enrolled in the Sindh Electronic Immunization Registry.

Gender-based inequities in immunization impede the universal coverage of childhood vaccines. Leveraging data from the Government of Sindh's Electronic Immunization Registry (SEIR), we estimated inequalities in immunization for males and females from the 2019-2022 birth cohorts in Pakistan. We computed male-to-female (M:F) and gender inequality ratios (GIR) Tfor enrollment, vaccine coverage, and timeliness. We also explored the inequities by maternal literacy, geographic location, mode of vaccination delivery, and gender of vaccinators. Between 1 January 2019, and 31 December 2022, 6,235,305 children were enrolled in the SEIR, 52.2% males and 47.8% females. We observed a median M:F ratio of 1.03 at enrollment and at Penta-1, Penta-3, and Measles-1 vaccinations, indicating more males were enrolled in the immunization system than females. Once enrolled, a median GIR of 1.00 indicated similar coverage for females and males over time; however, females experienced a delay in their vaccination timeliness. Low maternal education; residing in remote-rural, rural, and slum regions; and receiving vaccines at fixed sites, as compared to outreach, were associated with fewer females being vaccinated, as compared to males. Our findings suggeste the need to tailor and implement gender-sensitive policies and strategies for improving equity in immunization, especially in vulnerable geographies with persistently high inequalities.

Using geographic information system to track children and optimize immunization coverage and equity in Karachi, Pakistan.

BACKGROUND: Despite the potential of geospatial technologies to track and monitor coverage, they are underutilized for guiding immunization program strategy and implementation, especially in low-and-middle-income countries. We conducted geospatial analysis to explore the geographic and temporal trends of immunization coverage, and examined the pattern of immunization service access (outreach and facility based) by children. METHODOLOGY: We extracted data to analyze coverage rates across different dimensions (by enrolment year, birth year and vaccination year) from 2018 till 2020 in Karachi, Pakistan using the Sindh Electronic Immunization Registry (SEIR). We conducted geospatial analysis to assess variation in coverage rates of BCG, Pentavalent (Penta)-1, Penta-3, and Measles-1 vaccines using Government targets. We also analyzed the proportion of children receiving their routine vaccinations at fixed centers and outreach and examined whether children received vaccinations at the same or multiple immunization centers. RESULTS: A total of 1,298,555 children were born, enrolled or vaccinated from 2018 till 2020. At the district level, analysis by enrollment and birth year showed coverage increased between 2018 and 2019 and declined in 2020, while analysis by vaccination year showed consistent increase in coverage. However, micro-geographic analysis revealed pockets where coverage persistently declined. Notably 27/168, 39/168 and 3/156 Union councils showed consistently declining coverage when analyzing by enrollment, birth and vaccination year respectively. More than half (52.2%, 678,280/1,298,555) of the children received all their vaccinations exclusively through fixed centers and, 71.7% (499,391/696,701) received all vaccinations from the same centers. CONCLUSION: Despite overall improving vaccination coverage between 2018 and 2020, certain geographic areas have consistently declining coverage rates, which is detrimental for equity. Making immunization inequities visible through geospatial analysis is the first step to ensure resources are allocated optimally. Our study provides impetus for immunization programs to develop and invest in geospatial technologies, harnessing its potential for improved coverage and equity.

A Background-Agnostic Framework With Adversarial Training for Abnormal Event Detection in Video.

Abnormal event detection in video is a complex computer vision problem that has attracted significant attention in recent years. The complexity of the task arises from the commonly-adopted definition of an abnormal event, that is, a rarely occurring event that typically depends on the surrounding context. Following the standard formulation of abnormal event detection as outlier detection, we propose a background-agnostic framework that learns from training videos containing only normal events. Our framework is composed of an object detector, a set of appearance and motion auto-encoders, and a set of classifiers. Since our framework only looks at object detections, it can be applied to different scenes, provided that normal events are defined identically across scenes and that the single main factor of variation is the background. This makes our method background agnostic, as we rely strictly on objects that can cause anomalies, and not on the background. To overcome the lack of abnormal data during training, we propose an adversarial learning strategy for the auto-encoders. We create a scene-agnostic set of out-of-domain pseudo-abnormal examples, which are correctly reconstructed by the auto-encoders before applying gradient ascent on the pseudo-abnormal examples. We further utilize the pseudo-abnormal examples to serve as abnormal examples when training appearance-based and motion-based binary classifiers to discriminate between normal and abnormal latent features and reconstructions. Furthermore, to ensure that the auto-encoders focus only on the main object inside each bounding box image, we introduce a branch that learns to segment the main object. We compare our framework with the state-of-the-art methods on four benchmark data sets, using various evaluation metrics. Compared to existing methods, the empirical results indicate that our approach achieves favorable performance on all data sets. In addition, we provide region-based and track-based annotations for two large-scale abnormal event detection data sets from the literature, namely ShanghaiTech and Subway.

On Symbiosis of Attribute Prediction and Semantic Segmentation.

Attributes are semantically meaningful characteristics whose applicability widely crosses category boundaries. They are particularly important in describing and recognizing concepts for which no explicit training example is given, e.g., zero-shot learning. Additionally, since attributes are human describable, they can be used for efficient human-computer interaction. In this article, we propose to employ semantic segmentation to improve person-related attribute prediction. The core idea lies in the fact that many attributes describe local properties. In other words, the probability of an attribute to appear in an image is far from being uniform in the spatial domain. We build our attribute prediction model jointly with a deep semantic segmentation network. This harnesses the localization cues learned by the semantic segmentation to guide the attention of the attribute prediction to the regions where different attributes naturally show up. As a result of this approach, in addition to prediction, we are able to localize the attributes despite merely having access to image-level labels (weak supervision) during training. We first propose semantic segmentation-based pooling and gating, respectively denoted as SSP and SSG. In the former, the estimated segmentation masks are used to pool the final activations of the attribute prediction network, from multiple semantically homogeneous regions. This is in contrast to global average pooling which is agnostic with respect to where in the spatial domain activations occur. In SSG, the same idea is applied to the intermediate layers of the network. Specifically, we create multiple copies of the internal activations. In each copy, only values that fall within a certain semantic region are preserved while outside of that, activations are suppressed. This mechanism allows us to prevent pooling operation from blending activations that are associated with semantically different regions. SSP and SSG, while effective, impose heavy memory utilization since each channel of the activations is pooled/gated with all the semantic segmentation masks. To circumvent this, we propose Symbiotic Augmentation (SA), where we learn only one mask per activation channel. SA allows the model to either pick one, or combine (weighted superposition) multiple semantic maps, in order to generate the proper mask for each channel. SA simultaneously applies the same mechanism to the reverse problem by leveraging output logits of attribute prediction to guide the semantic segmentation task. We evaluate our proposed methods for facial attributes on CelebA and LFWA datasets, while benchmarking WIDER Attribute and Berkeley Attributes of People for whole body attributes. Our proposed methods achieve superior results compared to the previous works. Furthermore, we show that in the reverse problem, semantic face parsing significantly improves when its associated task is jointly learned, through our proposed Symbiotic Augmentation (SA), with facial attribute prediction. We confirm that when few training instances are available, indeed image-level facial attribute labels can serve as an effective source of weak supervision to improve semantic face parsing. That reaffirms the need to jointly model these two interconnected tasks.