Analysing Open-source Images to Assess Face Mask Usage for Epidemiological Studies

Background: Face masks are an available intervention for respiratory emerging infectious diseases. Research during the COVID-19 pandemic has sought to manually quantify mask use in mass gatherings and public settings. Open-source images of mass gatherings or other large events on news and social med...

Full description

Saved in:
Bibliographic Details
Main Authors: Dr Hang Ding, Mrs Ashley Quigley, Professor Raina MacIntyre
Format: Article
Language:English
Published: Elsevier 2025-03-01
Series:International Journal of Infectious Diseases
Online Access:http://www.sciencedirect.com/science/article/pii/S120197122400482X
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850043530032971776
author Dr Hang Ding
Mrs Ashley Quigley
Professor Raina MacIntyre
author_facet Dr Hang Ding
Mrs Ashley Quigley
Professor Raina MacIntyre
author_sort Dr Hang Ding
collection DOAJ
description Background: Face masks are an available intervention for respiratory emerging infectious diseases. Research during the COVID-19 pandemic has sought to manually quantify mask use in mass gatherings and public settings. Open-source images of mass gatherings or other large events on news and social media contain valuable information about face mask usage. This study aims to develop and validate an Artificial Intelligence (AI) solution to automatically analyse open-source crowd photos for estimation of mask use. Methods: Our AI solution includes four analysis stages: 1) detecting individual faces on a given image, 2) excluding unusable (blurry or occluded) face images, 3) classifying face images with or without a mask, and 4) identifying the indoor or outdoor environment. Multiple machine learning (ML) techniques were employed, such as a model named ResNet152-pretrained for the mask-related classifications at Stages 2 and 3 and Microsoft AI Computer Vision Service at Stage 4. To train the models, we collected open-source images using data-searching strategies similar to those developed for the AI-driven outbreak warning platform, EPIWATCH® (The Kirby Institute, Sydney, NSW, Australia) Results: We finetuned those mask-related classification models using 25800 individual face images from 1871 open-source crowd images. We evaluated the face mask classification using images of mass gatherings. The face images were correctly classified for mask use in 88% of cases. Discussion: We demonstrated the potential to classify open-source images to assess face mask use. Current results support our open-source analysis strategy with AI innovations for mask use estimation using open-source data. This is a novel epidemiological tool and further research is required to assess its validity in providing accurate assessments of mask wearing. Conclusion: Analysing open-source images presents new opportunities to estimate mask use or other visual characteristics pertinent to public health and provide valuable insights into mandated mask-related health policies. AI-powered image analysis systems can help identify high-risk areas, track disease progression, and inform targeted interventions, ultimately contributing to more effective public health strategies and improved population health outcomes.
format Article
id doaj-art-1b16a003a0b64f5a977cecf02cf9cea7
institution DOAJ
issn 1201-9712
language English
publishDate 2025-03-01
publisher Elsevier
record_format Article
series International Journal of Infectious Diseases
spelling doaj-art-1b16a003a0b64f5a977cecf02cf9cea72025-08-20T02:55:13ZengElsevierInternational Journal of Infectious Diseases1201-97122025-03-0115210740710.1016/j.ijid.2024.107407Analysing Open-source Images to Assess Face Mask Usage for Epidemiological StudiesDr Hang Ding0Mrs Ashley Quigley1Professor Raina MacIntyre2The Kirby Institute, The University Of New South Wales, Sydney, AustraliaThe Kirby Institute, The University Of New South Wales, Sydney, AustraliaThe Kirby Institute, The University Of New South Wales, Sydney, Australia; College of Public Service & Community Solutions, and College of Health Solutions, Arizona State University, Tempe, United StatesBackground: Face masks are an available intervention for respiratory emerging infectious diseases. Research during the COVID-19 pandemic has sought to manually quantify mask use in mass gatherings and public settings. Open-source images of mass gatherings or other large events on news and social media contain valuable information about face mask usage. This study aims to develop and validate an Artificial Intelligence (AI) solution to automatically analyse open-source crowd photos for estimation of mask use. Methods: Our AI solution includes four analysis stages: 1) detecting individual faces on a given image, 2) excluding unusable (blurry or occluded) face images, 3) classifying face images with or without a mask, and 4) identifying the indoor or outdoor environment. Multiple machine learning (ML) techniques were employed, such as a model named ResNet152-pretrained for the mask-related classifications at Stages 2 and 3 and Microsoft AI Computer Vision Service at Stage 4. To train the models, we collected open-source images using data-searching strategies similar to those developed for the AI-driven outbreak warning platform, EPIWATCH® (The Kirby Institute, Sydney, NSW, Australia) Results: We finetuned those mask-related classification models using 25800 individual face images from 1871 open-source crowd images. We evaluated the face mask classification using images of mass gatherings. The face images were correctly classified for mask use in 88% of cases. Discussion: We demonstrated the potential to classify open-source images to assess face mask use. Current results support our open-source analysis strategy with AI innovations for mask use estimation using open-source data. This is a novel epidemiological tool and further research is required to assess its validity in providing accurate assessments of mask wearing. Conclusion: Analysing open-source images presents new opportunities to estimate mask use or other visual characteristics pertinent to public health and provide valuable insights into mandated mask-related health policies. AI-powered image analysis systems can help identify high-risk areas, track disease progression, and inform targeted interventions, ultimately contributing to more effective public health strategies and improved population health outcomes.http://www.sciencedirect.com/science/article/pii/S120197122400482X
spellingShingle Dr Hang Ding
Mrs Ashley Quigley
Professor Raina MacIntyre
Analysing Open-source Images to Assess Face Mask Usage for Epidemiological Studies
International Journal of Infectious Diseases
title Analysing Open-source Images to Assess Face Mask Usage for Epidemiological Studies
title_full Analysing Open-source Images to Assess Face Mask Usage for Epidemiological Studies
title_fullStr Analysing Open-source Images to Assess Face Mask Usage for Epidemiological Studies
title_full_unstemmed Analysing Open-source Images to Assess Face Mask Usage for Epidemiological Studies
title_short Analysing Open-source Images to Assess Face Mask Usage for Epidemiological Studies
title_sort analysing open source images to assess face mask usage for epidemiological studies
url http://www.sciencedirect.com/science/article/pii/S120197122400482X
work_keys_str_mv AT drhangding analysingopensourceimagestoassessfacemaskusageforepidemiologicalstudies
AT mrsashleyquigley analysingopensourceimagestoassessfacemaskusageforepidemiologicalstudies
AT professorrainamacintyre analysingopensourceimagestoassessfacemaskusageforepidemiologicalstudies