IMPACT & EVIDENCE · PUBLICATIONS

Research Papers & Policy Publications

Every research finding, evaluation report, and technical dataset CHRSD produces is published in full — open access under Creative Commons Attribution 4.0 (CC BY 4.0), with no registration, paywall, or summary-only version.

LATEST RELEASE

Featured Health & Welfare Oct 18, 2026

Unhygienic Street-Vended Foods in Bangladesh: Impacts on Public Health, Economy, and National Well-being

M.A. Ramim, Md. Motalib Hossain

A national mixed-methods study (2018-2023) assessing the public health, economic and regulatory impacts of unhygienic street-vended food in Bangladesh. Drawing on a vendor survey (n=1,248), a consumer survey (n=2,416), hygiene observations of 486 stalls, and a systematic review of 134 studies, the paper finds foodborne illness cases rose 27% to an estimated 19.5 million annually, with an associated economic burden of USD 8.1 billion (2.2% of GDP). It proposes a livelihood-sensitive, five-pillar National Action Framework for food safety reform.

Download PDF

PUBLICATION REPOSITORY

All papers

Health & Welfare Oct 6, 2026

Enhancing Cell Type Annotation with Imperfect Single-Cell RNA-Seq References

Mohammad Razib Mustafiz, Chandan Kumar Roy, Redwan Hossain, Akram Hossain, Khaled Mohsin

This paper presents a Siamese network-based pipeline for robust cell type annotation in single-cell RNA sequencing (scRNA-seq), designed to perform reliably even when reference datasets are noisy, incomplete, or poorly representative of the populations under study. The twin-subnetwork architecture, with shared weights, learns similarity representations across paired gene expression profiles, enabling accurate classification without depending on pristine, resource-intensive reference atlases. The methodology covers dataset selection, normalization and noise-reduction preprocessing, balanced batch training, and evaluation via accuracy, precision, recall, and F1 metrics benchmarked against established methods (scmap, scCATCH, CHETAH). By lowering dependence on expensive reference data and high-performance computing, the approach aims to democratize access to advanced genomic analysis for research institutions in low- and middle-income countries, strengthening local disease surveillance, oncology research, and public health decision-making. The paper concludes with strategic recommendations for funders, policymakers, and practitioners seeking to translate deep learning genomic tools into equitable health outcomes. Manuscript status: methodological and policy analysis; empirical benchmarking against standard datasets (PBMC, pancreas, tumour microenvironment atlases) is underway, with quantitative performance tables to be appended to the final release version.

Download PDF
Health & Welfare Oct 1, 2026

Occupational Ocular Morbidity and Productivity Interventions among Ready-Made Garment (RMG) Workers in Bangladesh

M.A. Ramim, Mohammad Razib Mustafiz, Md. Motalib Hossain

An empirical evaluation of a joint free eye care camp organised by CHRSD Foundation, Lunera Eye Care, and Nevronus Systems on 22 May 2026 at Omi Sweaters Limited, Deora, Tongi, Gazipur, which screened 202 garment-factory workers and community members using AI-assisted handheld colour fundus photography and a mobile diagnostic application. Bangladesh's Ready-Made Garment (RMG) sector employs an estimated 4 million workers, the majority women, in visually demanding, detail-oriented assembly work, yet occupational eye health remains largely absent from the sector's safety agenda. The patient base was nearly gender-balanced (49.5% male, 50.5% female) and predominantly working-age (18-44 years, approximately 62%). Refractive error was the leading presenting condition (25.2%, male-skewed), followed by dry eye syndrome (20.3%, female-skewed) and medication-related complaints (15.8%) - together over 61% of visits. Roughly half of patients were prescribed spectacles and 3.0-3.5% were referred for specialist care. The findings, benchmarked against WHO, IAPB, and ILO reference data, indicate that occupational eye health - including an under-recognised burden of colour vision deficiency with direct implications for quality control and workplace safety - constitutes a measurable, currently unaddressed gap in Bangladesh's RMG occupational health framework, with direct relevance to SDG 3, SDG 8, and SDG 5. The paper concludes with an institutional policy framework for the Ministry of Labour and Employment, BGMEA/BKMEA, international buyers, and civil society actors. This is a retrospective, single-camp empirical evaluation (N=202) benchmarked against WHO/IAPB/ILO reference data, not a representative sectoral survey.

Download PDF
Health & Welfare Aug 1, 2026

Assessing Automated Machine Learning Services for COVID-19 Detection from X-Ray and CT Images

Razib Mustafiz, Khaled Mohsin

A real-time smartphone application case study evaluating Microsoft Custom Vision, an automated machine learning (AutoML) platform, for detecting and classifying COVID-19-induced pneumonia against other viral/bacterial pneumonia and normal lung tissue from chest X-ray and CT images. The study tested whether a cloud-trained model could be ported to run real-time, offline inference on a smartphone. The optimized model achieved 96.8% average precision as a web application, with 100% recall and 99.4% average precision specifically for the COVID-19 class. However, the identical trained network performed markedly worse once ported to a smartphone for on-device inference, with average precision for some classes falling from the high-90s on the web to as low as 15% on mobile. This web/mobile performance gap, and the underlying COVID-19 data scarcity and class imbalance likely driving it, show that cloud-validated diagnostic AI cannot be assumed to perform equivalently on-device — a finding directly relevant to rural and resource-constrained health systems that depend on offline, smartphone-based screening where internet connectivity is unreliable. This is a retrospective, methodological case study built on publicly available chest-imaging datasets; prospective, demographically diverse clinical validation is identified by the authors as necessary future work.

Download PDF
Health & Welfare Jan 29, 2026

Association between Dental Caries and Body Mass Index (BMI) among Adolescents in Urban Area of Rangpur

M A Ramim, Mohammad Razib Mustafiz, CHRSD Department of Statistics and Data Science

A cross-sectional study of 300 adolescents (12–16) at a government school in Rangpur, using multivariate logistic and linear regression to test the relationship between BMI and dental caries (DMFT). Caries was statistically linked to BMI, fluoride toothpaste use, and socioeconomic status, while physical activity was the strongest independent predictor of BMI itself. Originally published in the Journal of Statistics, Probability, and Its Application (Vol. 4, No. 2, 2026), DOI: 10.24127/sciencestatistics.v4i1.10289.

Download PDF

Open access

All papers are published under Creative Commons Attribution 4.0 (CC BY 4.0). Reuse with attribution — no permission needed.

How to cite

Every paper's "Cite / BibTeX" button generates a ready-to-use citation you can paste straight into your reference manager.

Submit or collaborate

Researchers and partner organisations can propose joint studies or request underlying datasets — contact our Research & Evidence team.