Available via license: CC BY 3.0
Content may be subject to copyright.
Bull World Health Organ 2023;101:381–390 | doi: http://dx.doi.org/10.2471/BLT.22.289061
Research
381
Introduction
Cervical cancer is a threat to women’s health. According to data
released by the International Agency for Research on Cancer,
the cancer research agency of the World Health Organization
(WHO), 0.6 million women worldwide were diagnosed with
cervical cancer and 0.3 million died from cervical cancer in
2020.1 ese numbers represent 3.1% of all newly diagnosed
cancers worldwide (making cervical cancer the ninth leading
cancer diagnosis), and 7.7% of global female cancer deaths
(making this cancer the fourth leading cause of death in
women).1 WHO released the Global strategy to accelerate
the elimination of cervical cancer in 2020, with the goal of
eliminating cervical cancer (a rate of fewer than 4 cases per
100 000 women) by 2030. One of the targets is that 70% of
women are screened with a high-performance test by 35 years
of age, and again by 45 years of age. us far, 194 countries,
including China, have made the commitment to eliminate
cervical cancer.2,3
In contrast to developed countries, where the incidence
of cervical cancer is low and prevention is the primary focus,
low- and middle-income countries, such as China, are still
far from achieving this goal,4 and many women are still not
screened or are under-screened.5–10 However, the costs of
deoxyribonucleic acid (DNA) detection of human papilloma-
virus in large populations and the expansion of the cervical
cancer vaccine, which are the main cervical cancer preven-
tion measures in developed countries,11 are high. erefore,
the use of relatively inexpensive cytology methods with good
sensitivity and specicity is needed to achieve universal and
reliable screening coverage in large populations in low- and
middle-income countries. WHO has also suggested that cyto-
logical techniques still play an important role in screening and
triage.12,13 However, the widespread use of cytology in China is
restricted by the lack of trained cytopathologists and dedicated
laboratories.14,15 Furthermore, dierences in prociency of
cytopathologists and the diculty in maintaining diagnostic
consistency among the limited number of cytopathologists
lead to inconsistencies in the quality of diagnosis.14 Resolving
all these problems is important to facilitate the use of cytology
in screening.
e need for improved digital health in remote and de-
prived areas of China has recently been highlighted.16–18 e
National Health Commission is encouraging the incorporation
of emerging articial intelligence and internet technologies in
screening to enhance cervical cancer prevention and treat-
ment.19 As machine learning develops, articial intelligence
has advantages in cervical cytology14,20,21 and colposcopy,22–24
and will gradually outperform human clinicians in performing
these procedures.25–27 WHO has also suggested that articial
intelligence can advance the screening process.13 Preparation
and interpretation of cervical cell slides using articial intel-
ligence can replace or assist cytopathologists, which addresses
the problem of a lack of experienced medical personnel in
screening.14,20,21,26–28 In addition, improved access to health-
care services in remote areas and lower health-care costs are
both made possible by the internet, cloud computing and
mobile devices.14,20
Based on the above-mentioned feasibility studies and
policies, we implemented an online cervical cancer screen-
ing programme aided by articial intelligence in Hubei
Province, China. is programme aimed to: increase access
Objective To implement and evaluate a large-scale online cervical cancer screening programme in Hubei Province, China, supported by
artificial intelligence and delivered by trained health workers.
Methods The screening programme, which started in 2017, used four types of health worker: sampling health workers, slide preparation
technicians, diagnostic health workers and cytopathologists. Sampling health workers took samples from the women on site; slide preparation
technicians prepared slides for liquid-based cytology; diagnostic health workers identified negative samples and classified positive samples
based on the Bethesda System after cytological assessment using online artificial intelligence; and cytopathologists reviewed positive
samples and signed reports of the results online. The programme used fully automated scanners, online artificial intelligence, an online
screening management platform, and mobile telephone devices to provide screening services. We evaluated the sustainability, performance
and cost of the programme.
Results From 2017 to 2021, 1 518 972 women in 16 cities in Hubei Province participated in the programme, of whom 1 474 788 (97.09%)
had valid samples for the screening. Of the 86 648 women whose samples were positive, 30 486 required a biopsy but only 19 495 had
one. The biopsy showed that 2785 women had precancerous lesions and 191 had invasive cancers. The cost of screening was 6.31 United
States dollars (US$) per woman for the public payer: US$ 1.03 administrative costs and US$ 5.28 online screening costs.
Conclusion Cervical cancer screening using artificial intelligence in Hubei Province provided a low-cost, accessible and effective service,
which will contribute to achieving universal cervical cancer screening coverage in China.
a School of Medicine and Health Management, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, 430030 China.
b School of Political Science and Administration, Wuhan University, Wuhan, China.
c Landing Artificial Intelligence Center for Pathological Diagnosis, Wuhan University, Wuhan, China.
d Landing Artificial Intelligence Industry Research Institute, Wuhan, China.
Correspondence to Zhiyong Liu (e-mail: zhiyongliu@ hust .edu .cn).
(Submitted: 17 August 2022 – Revised version received: 27 March 2023 – Accepted: 29 March 2023 – Published online: 1 May 2023 )
Cervical cancer screening aided by articial intelligence, China
Xingce Zhu,a Qiang Yao,b Wei Dai,a Lu Ji,a Yifan Yao,a Baochuan Pang,c Bojana Turic,d Lan Yaoa & Zhiyong Liua
382 Bull World Health Organ 2023;101:381–390| doi: http://dx.doi.org/10.2471/BLT.22.289061
Research
Cervical cancer screening, China Xingce Zhu et al.
to relevant health services for women
in underdeveloped areas who have lim-
ited screening coverage; alleviate some
of the health inequities these women
face; and contribute to the achievement
of universal cervical cancer screening
coverage and cervical cancer elimination
goals in China. In this paper, we describe
this technology-based service-delivery
model and evaluate its success in terms
of: programme sustainability (number
of women screened annually in dier-
ent cities); performance (number of
women with conrmed diagnoses and
some important indicators of screening
outcomes, such as number of inadequate
and positive slides); and costs for the
technology provider and public payer.
We also discuss the challenges encoun-
tered in the implementation of this
programme and the strategies developed
to overcome these issues.
Methods
Study setting and preparation
We rst implemented our screening
programme among rural women of low
socioeconomic status and gradually
expanded it to urban women in Hubei
Province, China, from 2017 to 2021.
Hubei Province has 36.32 million urban
residents and 21.43 million rural resi-
dents, and 12.67 million women aged 35
to 64 years are eligible for cervical cancer
screening.29 In the screening, all women
are initially evaluated using cytology,
and women with abnormal cytology
results are referred for colposcopy-di-
rected biopsies, if necessary.
is screening programme was
established through an open-tender
process to select a screening contractor
and technology provider. e Landing
Articial Intelligence Industry Research
Institute in Wuhan (hereaer called
Landing), which is certied to perform
cytological and histopathological di-
agnosis and has an online cytological
assessment system using articial intel-
ligence, won the tender in 2017. Screen-
ing can be performed in areas where
mobile telephones can receive fourth-
generation (4G) mobile communication
technology signals, which allow screen-
ing data to be sent through the network.
By the end of 2016, all administrative
villages and urban households in Hubei
Province had full 4G network coverage.
Before implementation of the pro-
gramme, we conducted a situational
analysis of the routes to screening
services and the referral routes for treat-
ment for women with positive cytology
results. Landing, local health centres,
and maternal and child health hospitals
were the main screening service provid-
ers. Local health centres, which may
have a gynaecology department, are
distributed in villages and urban com-
munities and provide primary health
care. Local maternal and child hospitals,
generally one for each county and city,
are secondary or tertiary hospitals dedi-
cated to providing comprehensive health
care to all women, infants and children
in the region.
We also established partnerships
with local health commissions and
women’s federations. ese groups and
the local health centres in cities across
Hubei Province typically conduct
screening mobilization in rural and
urban communities each year between
January and March. rough on-site
meetings and online announcements,
women were informed about cervical
health, the importance of early preven-
tion and screening, the time and place of
screening, the factors to consider before
screening (for example, ideally not to
screen during menstruation) and the
tests oered to women in the screening.
Women also received short messages on
their mobile telephones reminding them
to participate in the screening. Around
September each year, women in villages
and communities with insucient par-
ticipation in the cytology-based screen-
ing, as determined by aggregated screen-
ing data, were mobilized once more.
Health workers
We assigned four types of trained health
workers to conduct the cytology-based
screening: sampling health workers;
slide preparation technicians; diagnostic
health workers; and cytopathologists
(Box 1). ese health workers mostly
came from local health centres, mater-
nal and child hospitals, and Landing,
and were their regular medical sta.
Landing undertook the training of the
health workers in the use of the screen-
ing system.
e sampling health workers from
rural and urban health centres were
assigned to collect and x cervical
exfoliated cells for the women being
screened. Local maternal and child
health hospitals in urban areas provided
urban women with access to sampling.
Given that some trained sampling
health workers in local health centres
may lack prociency and experience in
sampling, particularly in rural health
centres, the local maternal and child
health hospital can send someone to
help with the sampling process. Slide
preparation technicians used specic
devices for slide coding and slide prepa-
ration for liquid-based cytology. e
trained diagnostic health workers were
assigned to identify negative cases, and
classify the cases with abnormalities
based on the Bethesda System aer the
cytological assessment using online
articial intelligence had been done.
Trained cytopathologists then reviewed
the positive cases and their classication,
and checked a random 10% of cases clas-
sied as negative by diagnostic health
workers for quality control.
e sampling health workers were
employed full-time at local health cen-
tres or maternal and child health hospi-
tals, where they oversaw the provision of
daily health-care services to residents of
a specic area; sampling women in the
area was a regular part of their work.
Box 1. Institutions and sta involved in the online cervical cancer screening programme,
Hubei Province, China, January 2017–December 2021
Rural and urban health centres, and local maternal and child health hospitals
Sampling health workers
• Main tasks: collect identity information and cervical exfoliated cells from the women.
Landing Articial Intelligence Industry Research Institute
Slide preparation technicians
• Main tasks: prepare slides for diagnosis.
Diagnostic health workers
• Main tasks: Identify negative cases and carry out simple classification of positive cases based
on the Bethesda System after cytological assessments by the online artificial intelligence
system.
Cytopathologists
• Main tasks: Review all positive cases and 10% of negative cases (randomly selected) classified
by diagnostic health workers, and sign reports of the results.
383
Bull World Health Organ 2023;101:381–390| doi: http://dx.doi.org/10.2471/BLT.22.289061
Research
Cervical cancer screening, China
Xingce Zhu et al.
e salaries of these health workers were
paid by dierent medical institutions.
Landing employed the diagnostic health
workers, slide preparation technicians
and some of the trained cytopathologists
on a full-time basis. Other cytopatholo-
gists were from large hospitals, such as
Tongji Hospital and Union Hospital in
Wuhan. ese cytopathologists were
paid by Landing based on the number
of cases they diagnosed online in their
spare time; these cytopathologists un-
dertook this work to supplement their
income.
All of the health workers had previ-
ously received medical education train-
ing, but no formal training in online
cytological assessment using articial
intelligence. e sampling health work-
ers attended a 1-day screening meeting
and training on sampling standards.
Slide preparation technicians were
procient in liquid-based cytology slide
preparation before joining Landing,
and required no additional training
aer joining the screening programme.
Senior pathologists employed by Land-
ing provided all diagnostic health
workers with 2–3 weeks of specialized
training on digital cytological assess-
ment, including: grading criteria of the
Bethesda System and the Papanicolaou
test; procedures for dierentiating
between negative and positive cases by
reviewing digital cell images online;
and the online operational workow.
e cytopathologists did not require
medical skills training, but had 3–5 days
of training on the operation of online
cytological assessment and issuing of
reports. Courses for the diagnostic
health workers were held at the Landing
facilities; other training was held at the
workplace of the health workers. All the
health workers were supervised through
an online screening management plat-
form (Landing, Wuhan, China), as the
data generated at each step of their
operation was recorded.
Implementation
e implementation of the screening
programme had four phases: sampling,
prediagnostic preparation, cytology-
based assessment and treatment for
positive cases (Fig. 1).
Each city in Hubei Province de-
veloped its own screening workplan.
In accordance with this workplan, the
sampling health workers took samples
from women who signed consent forms
at local health centres or maternal and
child health hospitals over the screen-
ing period, usually 1 week or more.
First, they used a networked identity
card reader to capture women’s identity
information and archive it in the online
screening management platform. Each
woman was assigned a unique quick re-
sponse (QR) code that could be scanned
with a mobile device, such as a smart-
phone or tablet, to access her online
cytological assessment report. Next, the
sampling health workers collected and
xed the exfoliated cervical cells and
arranged for their transport to Landing’s
Wuhan laboratory. For women in vil-
lages or communities who had diculty
travelling to the centres, sampling health
workers were taken by bus to provide a
door-to-door sampling service.
Aer specimens from the sampling
facilities arrived at the laboratory, the
slide preparation technicians prepared
slides for liquid-based cytology. en,
the slides were loaded into fully au-
tomated scanners (Landing, Wuhan,
China) for scanning and digital cell
image uploading. e QR code on each
slide was used to match it with the
Fig. 1. Flowchart of processes in the online cervical cancer screening programme, by phase, Hubei Province, China, January 2017–
December 2021
Sampling
Local health centres
Local maternal & child health
hospitals
Pre-diagnosis preparation
Landing laboratory
Cytology-based diagnosis
Online screening management
platform
Treatment for positive cases
Cytological reports
Colposcopy or colposcopy-
directed biopsies
Surgery
Local maternal & child health
hospitals
Negativea
Normal
Positiveb
Abnormal
Women signed consent forms
Sampling health workers recorded
women´s information and
assigned them a unique identifier
(QR code)
Sampling health workers collected
and fixed exfoliated cervical cells
Slide preparation technicians
coded and stained slides, and
loaded slides into automated
scanners
Slides were scanned by
automated scanners
Digital cell images were
uploaded to the online
screening management
platform
Slide quality evaluations and
cytology assessment were done
using the using artificial
intelligence system
Diagnostic health workers
confirmed negative cases and
classified positive cases with
abnormal segments
Cytopathologists reviewed all
positive cases and 10% of
negative cases and signed reports
of the results
QR: quick response.
a These electronic reports could be accessed by the women online.
b These paper reports needed to be picked up by the women at a designated health centre.
384 Bull World Health Organ 2023;101:381–390| doi: http://dx.doi.org/10.2471/BLT.22.289061
Research
Cervical cancer screening, China Xingce Zhu et al.
woman recorded in the system. In some
cities, a sub-laboratory was established
with the help of Landing, which enabled
slide preparation and data uploading lo-
cally. Furthermore, a portable hand-held
automated scanner (Landing, Wuhan,
China) could be sent to remote villages
or communities so that data could be
uploaded to the online screening man-
agement platform immediately aer
sampling. is strategy was used to
overcome the diculty and high cost of
preserving and transporting specimens
in remote villages or communities with
few women.
Aer scanning the samples, the
online articial intelligence system
evaluated the adequacy of the slides and
made an initial cytological assessment.
Women whose samples were considered
inadequate needed to be resampled;
samples that were adequate were clas-
sied as negative (negative for intraepi-
thelial lesion or malignancy) or positive.
All samples read by the online articial
intelligence system were reviewed by
the diagnostic health workers. ese
health workers logged into the screening
management platform to identify nega-
tive cases and classify positive samples
based on the Bethesda System.
e main task of the cytopatholo-
gists was to conrm or adjust the clas-
sication of samples with abnormalities,
and to sign the reports of the results on
the screening management platform.
For quality control, the cytopathologists
also reviewed 10% of the negative sam-
ples (randomly selected) that had been
classied by the diagnostic health work-
ers. e report was available between 1
and 10 days aer the specimen arrived
at the Landing laboratories. Women
with a negative diagnosis could directly
view their report online using WeChat
(Tencent, Guangzhou, China) on their
mobile telephone devices and they were
placed on a list for the next cycle of the
screening programme.
Women with positive samples could
not view their report online; instead,
they were sent a message informing
them of a possible abnormal diagnosis
and the place where they could obtain
their paper report. Landing sent the pa-
per reports of cytological abnormalities
to the local health centre, or maternal
and child health hospital in the area
where the women with positive samples
lived. When the women picked up their
reports, they were told where they could
obtain a free re-examination – usually
at the local maternal and child hospital.
Women were classied as requiring a
colposcopy-directed biopsy if their phy-
sicians recommended a biopsy during
re-examination.
We calculated the cost of the screen-
ing programme from two perspectives:
the public payer (provincial government
nance department) and the technology
provider (Landing).
Technology
The technology and devices we used
for screening were able to meet our
screening demands. A fully automated
scanner in the laboratory can load
200 slides at a time, and automatically
change slides in sequence to scan and
upload data to the online screening
management platform. Each slide took
fewer than 2 minutes to be scanned
by the scanner and the data uploaded
to the online system for cytological
assessment using the online artificial
intelligence system, and a single
scanner could process 1000 slides a
day. With 50 networked automated
scanners, the Landing laboratory was
able to perform around 30 000 high-
throughput cytological assessments a
day with an average diagnostic time
of 1.5 minutes per sample, which is
equivalent to the maximum workload
of a pathologist for a year with quality
assurance and error avoidance.25,30 A
portable hand-held scanner can only
load one or two slides at a time, but
it scans and uploads with the same
performance as the laboratory scanner.
In addition, the diagnostic output
within a certain time could be increased
by adding more scanners or opening a
sub-laboratory elsewhere with these
networked devices.
e accuracy of the online
cytological assessment system aided by
articial intelligence has been improving
as cell images of screened women have
gradually accumulated in the digital cell
image library. Landing ran well-dened
cell data through the system at many
dierent times to improve the accuracy
of the identication of slides as negative,
positive or inadequate.26,27 On the basis
of the Bethesda System, negative meant
negative for intraepithelial lesion or
malignancy, and the rest of the grades
were positive. Inadequate slides had
fewer than 5000 epithelial cells in the
eld of view, a large accumulation
of cells or a blurred background,
typically because of irregular sampling
procedures. Adequate slides generated
high-denition microelectronic cell
images, and were given a comprehensive
score based on over 1000 indicators
analysed by the online articial
intelligence system. e 20 most likely
abnormal segments were selected
for diagnostic health workers and
cytopathologists to review.
e online screening management
platform, which was built on AliCloud
(Alibaba, Hangzhou, China), synchro-
nized screening results between the
network and mobile device soware.
China’s Ministry of Public Security has
certied this platform as a National
Information Security Classied Protec-
tion Level 3. is rating is the highest
level of national certication for the
information security management
capabilities of non-banking institu-
tions. e diagnostic health workers
and cytopathologists used the account
and password given by Landing to log
into the platform and review the slides.
Only the case number, age, current as-
sessment result and a panoramic view
of the cell images were displayed during
the review. Information that could be
used to identify an individual was en-
crypted. e diagnostic health workers
and cytopathologists completed their
online reviews on their work computers
or mobile devices (smartphones, tablets,
laptops or personal computers) without
compromising condentiality because
of the protection within the system and
professional ethical conduct.
Data extraction
With the assistance of Landing, we ex-
tracted data from the online platform
into an Excel (Microso, Redmond,
USA) spreadsheet for data integration
and analysis. e data we obtained and
analysed were big data and any informa-
tion on the women’s identities was re-
moved. Information about patient safety
and condentiality was the responsibil-
ity of the medical service providers. We
also used Excel to record and quantify
some information management indica-
tors for the screening programme.
Ethical considerations
is study of health information
management was considered exempt
from ethical review and all women
provided informed consent for the
use of their data. Before samples were
taken, women signed a written informed
consent form which noted that their
385
Bull World Health Organ 2023;101:381–390| doi: http://dx.doi.org/10.2471/BLT.22.289061
Research
Cervical cancer screening, China
Xingce Zhu et al.
samples could be used in big data
studies. e data we obtained for this
study did not include any sensitive
information such as women’s identity
or human tissue.
Results
A total of 1 518 972 eligible women who
signed consent forms participated in the
programme in 16 cities of Hubei Prov-
ince between 2017 and 2021 (Tab le 1).
Of these women, 44 184 (2.91%) had
inadequate samples and could not be
classied, 1 388 140 (91.39%) had no
abnormalities on cytological assessment
and 86 648 (5.70%) were classied as
positive (Table 2 ). Of the women with
positive diagnosis, 30 486 were recom-
mended to have a colposcopy-directed
biopsy but only 19 495 (63.95%) had
the procedure (Tab l e 2). Of the women
who had a biopsy, 2785 were diagnosed
with cervical precancerous lesions and
191 with inltrating cervical cancer
(Table 2 ).
e cost of the screening pro-
gramme for the public payer was 6.31
United States dollars (US$) per woman,
including administration and screening
costs (Tab le 3). e screening fee was
paid as a package by the public payer
to Landing. e other health service
providers involved in the screening also
shared the cost based on the actual work
done. Most of this fee was intended to
cover the various screening services pro-
vided by health workers to the women.
From the perspective of Landing, the
technology provider, the project imple-
mentation costs also included training
for three types of health worker which
is not covered by traditional screening
services, including US$ 29.34 per person
for the training of 3174 sampling health
workers, US$ 293.40 per person for the
training of 100 diagnostic health work-
ers and US$ 440.10 per person for the
training of 40 cytopathologists. e costs
to Landing also included US$ 0.07 per
specimen for transport (Tab l e 3).
Discussion
Our online cervical cancer screening
programme aided by articial intelli-
gence was successful in several ways. e
low cost, high quality and quick results
of the online cytology-based screening
were benecial to large numbers of
women. e programme increased the
eciency of data acquisition and track-
ing in the initial cytology-based screen-
ing, reduced reliance on a large number
of cytopathologists and had a high par-
ticipation rate. However, the programme
encountered several challenges, such as:
the diculty for some women to travel
to the sampling facilities; the inadequate
sampling prociency of sampling health
workers at local health centres; the high
workload of reviewing negative cases;
and the diculty in specimen conserva-
tion and transportation for remote areas
with few women. We developed some
mitigation strategies that were success-
fully applied during the programme and
have continued until now (Box 2).
Because of a lack of funding for
development, diagnostic facilities and
qualied medical sta, local health centres
are oen underused31,32 and not trusted33
Table 1. Number of women screened for cervical cancer through the online screening
programme, by city and year, Hubei Province, China
City No. of women screened
2017 2018 2019 2020 2021aTotal
Enshizhou 90 260 51 345 74 485 45 034 NA 261 124
Ezhou 3 913 2 377 7 254 2 849 NA 16 393
Huanggang 69 821 55 882 67 138 45 369 745 238 955
Huangshi 6 219 6 371 12 821 9 163 1 940 36 514
Jingmen 16 326 14 654 20 850 13 274 121 65 225
Jingzhou 24 489 21 943 37 977 21 678 NA 106 087
Qianjiang 2 739 9 179 7 502 5 894 NA 25 314
Shennongjia 79 1 538 1 166 704 902 4 389
Shiyan 39 707 50 310 69 229 33 147 13 328 205 721
Suizhou 11 886 16 576 18 819 13 576 28 60 885
Tianmen 4 109 3 192 9 302 5 414 6 466 28 483
Xiangyang 24 254 20 554 22 006 20 168 7 554 94 536
Xianning 8 746 32 396 41 543 19 266 12 840 114 791
Xiantao 4 689 2 530 8 809 5 368 NA 21 396
Xiaogan 34 030 17 188 36 534 19 974 2 630 110 356
Yichang 28 830 27 040 40 446 23 372 9 115 128 803
Total 370 097 333 075 475 881 284 250 55 669 1 518 972
NA: not applicable.
a Cities with NA had met their cytology-based screening coverage targets ahead of schedule.
Table 2. Outcomes of the online cervical cancer screening programme, Hubei Province,
China, January 2017–December 2021
Outcome Women screened
(n = 1 518 972)
Diagnosis on initial cytology screening, no. (%)
Negative for intraepithelial lesion or malignancy 1 388 140 (91.39)
Atypical squamous cells of undetermined significance 60 029 (3.95)
Low grade squamous intraepithelial lesion 21 045 (1.39)
Atypical squamous cells, cannot exclude high-grade squamous
intraepithelial lesion
3 401 (0.22)
High-grade squamous intraepithelial lesion 2 148 (0.14)
Atypical glandular cells 25 (< 0.00)
Total positive samples 86 648 (5.70)
Inadequate samples 44 184 (2.91)
Result of examination of women with positive samples, no.
Women who needed a colposcopy-directed biopsy 30 486
Confirmed diagnosis, no.
Women with a colposcopy-directed biopsy 19 495
Women with cervical precancerous lesions 2 785
Women with cervical infiltrating cancer 191
386 Bull World Health Organ 2023;101:381–390| doi: http://dx.doi.org/10.2471/BLT.22.289061
Research
Cervical cancer screening, China Xingce Zhu et al.
in China. However, because they provide
long-term routine health care to targeted
women in the community or village
and thus have a good foundation for
interaction with the population, sampling
at local health centres as the rst stage
of the screening process may increase
women’s participation in cytology-based
screening.34 We arranged for local maternal
and child health hospitals to assist the less
skilled and experienced health workers
at local health centres in completing
the sampling. is approach allowed
local health centres to improve their
service skills while also receiving some
nancial assistance through the public
payment channel; it also gave women the
opportunity to access screening services
near their homes.
e use of eective cytology as-
sessment algorithms with good per-
formance to replace cytopathologists
in mass cytological assessment was an
important factor in the success of our
programme (Fig. 2). Previous studies
have shown that the online cytological
assessment system aided by articial
intelligence that we used had a 5.8%
better sensitivity than humans in iden-
tifying cervical intraepithelial neoplasia
grade 2 and above.27 A multicentre clini-
cal trial reached a similar conclusion
and provided additional ndings, for
example, that the specicity of arti-
cial intelligence was higher than that
of experienced cytopathologists while
maintaining the same sensitivity, and
that both the sensitivity and specicity
of articial intelligence were higher than
with less experienced cytology physi-
cians.26 Articial intelligence-enabled
methods are also consistent in quality,35
and oer a signicant advantage over
cytopathologists in traditional screen-
ing who need to maintain the quality
of their diagnostic skills and avoid er-
rors.25,30 Of the women screened, only
5.70% had positive samples; for cyto-
pathologists to nd these positive cases
would take more time or would be more
labour-intensive. However, in our online
cytological assessments, the diagnostic
health workers reviewed a sample in
an average of 10–12 seconds, which
allowed them to evaluate 2500–3000
samples a day.
Noteworthy is that online cyto-
logical assessment using articial intel-
ligence can only be used as a reference
for diagnosis, and the nal diagnostic
report must still be reviewed by a cyto-
pathologist. If cytopathologists were still
assigned to review the large number of
negative cases and potential false posi-
tives,26,27,36 their workload would not be
reduced. What we hoped by implement-
ing this type of cytological assessment
was that experienced cytopathologists
would be able to focus on interpreting
both uncomplicated and dicult posi-
tive cases that arose from the screening,
and that large populations in resource-
constrained areas would be able to ac-
cess good-quality screening services. We
achieved this aim by positioning full-
time trained diagnostic health workers
between the cytological assessment and
cytopathologists’ reviews.
If local computing resources are
constrained, completing cytological
assessments of large populations might
be time-consuming. We therefore made
the running of the algorithms dependent
on a reliable online platform to obtain
the required computing power and
achieve simultaneous processing of case
data.37 In Hubei Province, where the 4G
network is widespread, pending cases
Table 3. Cost of online cervical cancer screening programme, Hubei Province, China
Service or goods Cost, US$
Total for the programme Per person
For the public payer
Administrationa1 564 541.16 1.03b
Online screenings by the technology providerc8 020 172.16 5.28b
For the technology provider
Training for sampling health workers 93 125.16 29.34d
Training for diagnostic health workers 29 340.00 293.40e
Training for cytopathologists 17 604.00 440.10f
Specimen transport 106 328.04 0.07b
US$: United States dollars.
a Includes organization, advocacy and management of digital screening archives.
b Total number of women screened was 1 518 972.
c Includes liquid-based c ytology slide preparation, ThinPrep cytological test reports, re-examination and
biopsies.
d Total number of trained sampling health workers was 3174.
e Total number of trained diagnostic health workers was 100.
f Total number of trained cytopathologists was 40.
Note: The technology provider was Landing Artificial Intelligence Industr y Research Institute, Wuhan, China.
Box 2. Measures used to mitigate challenges to the online cervical cancer screening
programme, Hubei Province, China, January 2017–December 2021
• Establishment of sub-laboratories outside of Wuhan. We reached agreements with some
local maternal and child health hospitals to set up laboratories with devices linked to the
online screening platform, which allows the hospitals to perform the same procedures as
the Landing laboratory in Wuhan.
• Use of portable hand-held automated scanners. The technology provider (Landing Artificial
Intelligence Industry Research Institute) developed a hand-held version of the automated
scanner. This scanner is less efficient than the laboratory version, but it was used in remote
areas where there were few women so that their data could be uploaded to the screening
management platform immediately after sampling. This measure allowed these women
to be included in the screening, and reduced the cost of sample transportation to Wuhan
for diagnosis.
• Deployment of the sampling bus. The Landing Artificial Intelligence Industry Research
Institute deployed this bus to provide safe, door-to-door sampling ser vices for women who
had difficulty travelling to sampling centres. The bus had the same standardized operating
environment as the sampling centres.
• Deployment of trained sampling health workers from local maternal and child hospitals to
local health centres. These health workers were sent to assist the sampling health workers
at the local health centres who were inexperienced in sampling.
• Deployment of trained diagnostic health workers. The work of the diagnostic health
workers was a full-time job that focused on detecting all negative samples using cytological
assessments aided by artificial intelligence, and classifying positive cases based on the
Bethesda System. The use of these diagnostic health workers led to a substantial reduction
in the work of the cytopathologists who would otherwise have had to review the large
number of negative samples.
387
Bull World Health Organ 2023;101:381–390| doi: http://dx.doi.org/10.2471/BLT.22.289061
Research
Cervical cancer screening, China
Xingce Zhu et al.
could be uploaded in time for automatic
processing as long as the data upload
terminals were in a location where they
could connect to the network; demands
for data downloads could also be met at
any time (Fig. 2).
e screening programme had
cost advantages for the public payer
and the cost per woman was low. Since
there are economies of scale associated
with this standardized, batched and
automated screening service using
articial intelligence, this low cost per
woman can be further reduced as the
number of women screened continues
to increase.38
Concerns have been raised that
the use of emerging information and
communication technologies may
not facilitate coverage of new health
services, but rather exacerbate health
inequalities for vulnerable groups
such as elderly people, rural dwellers
and those of lower socioeconomic
status.39–43 However, we designed our
programme to use such technologies
to both increase access to screening
services for vulnerable groups of women
(Fig. 2) and ensure the stability of the
programme during the coronavirus
disease 2019 (COVID-19) pandemic.
Specifically, several types of networked
automated scanners provided greater
flexibility and speed in collating data
on the women and the samples. Since
the online platform synchronized
data generated and uploaded by any
terminal device in the system, we
could easily access data on screening
progress, diagnostic operations,
quality control and screening archives
for administrative supervision and
evaluation. Most importantly, the use
of the online artificial intelligence
system, diagnostic health workers and
cytopathologists enabled successful,
rapid and quality-assured cytological
assessments. Even though COVID-19
affected some of the planned work on
site, we took steps to mitigate these
issues. We used online announcements
and short message notifications to
encourage women’s participation in the
screening, while on-site meetings were
mostly postponed. Appointments for
sampling and treatment were arranged
online to limit the number of women
who could attend on-site services each
da y.
Our study has some limitations.
No standard measures were in place for
women whose slides were considered
inadequate. Some women were called
immediately and successfully resampled,
while others, especially rural women,
were unable to complete the resampling
because they were oen moving to nd
work. Some women who were notied
were resampled several months later.
Inadequate samples are unavoidable
because of sampling errors, so it is
important that resampling for women
with such samples is ocially planned.
Women with cervical precancerous
lesions and cancer in low- and middle-
income countries are oen unsupported
without timely detection. In accordance
with WHO’s goal of eliminating cervical
cancer and China’s campaign for
universal coverage of cervical cancer
screening, we have shown that an online
service-delivery system using articial
intelligence can provide increased
access to screening for poor, remote and
migratory women. e ecient design of
such systems requires a comprehensive
approach that includes the use of
digital technologies, the training and
monitoring of health workers, the
support of local government agencies
and eective plans for implementation
of screening.
Future research should focus on:
evaluation of the programme’s key
technologies and workforce deployment;
addressing any problems with this new
screening model; and exploration of the
potential integration of other services
using computer technology. ■
Acknowledgements
Xingce Zhu and Qiang Yao contributed
equally to this work.
Competing interests: None declared.
Fig. 2. Enabling factors of the online cervical cancer screening programme, Hubei
Province, China, January 2017–December 2021
• QR code generated for each woman in the screening system as a unique identifier.
• Digital screening archives set up for women on the online integrated screening
management platform.
• Widespread 4G network.
Sampling
Pre-diagnosis
Cytology-based
diagnosis
Report generation
and follow-up
• Automatic capture and batch upload of women’s identifiers and digital high-definition
microscopic images of cervical cells.
• Deployment of portable networked automatic scanners for timely uploading of data
from women screened in remote areas.
• Widespread 4G network.
• Automatic high-throughput online cytological assessment using artificial intelligence.
• Online review of cases by diagnostic health workers and cytopathologists.
• Visualization and conservation of diagnostic records, processes, reports and archives on
the online integrated screening management platform.
• Widespread 4G network.
• Negative reports accessed directly by women on their mobile devices.
• Follow-up results of cytology-positive cases documented on the online integrated
screening management platform.
• Widespread 4G network.
Enabling factorsSteps
4G: fourth generation; QR: quick response.
388 Bull World Health Organ 2023;101:381–390| doi: http://dx.doi.org/10.2471/BLT.22.289061
Research
Cervical cancer screening, China Xingce Zhu et al.
摘要
中国使用人工智能技术辅助实施宫颈癌筛查
目的 在中国湖北省,接受过培训的卫生工作者在人工
智能技术的支持下实施并评估了一项大规模的在线宫
颈癌筛查项目。
方法 该筛查项目自 2017 年起开始启用,需四种卫生
工作者协同工作 :采样卫生工作者、玻片制备技术人
员、诊断卫生工作者和细胞病理学家。采样卫生工作
者从现场女性身上采集样本 ;玻片制备技术人员制
备液基细胞学载玻片 ;诊断卫生工作者在使用在线
人工智能软件进行细胞学评估后,根据贝塞斯达系统
(Bethesda System) 识别阴性样本并分拣出阳性样本 ;细
胞病理学家仔细研究阳性样本,并在线签署结果报告。
该项目使用全自动扫描仪、在线人工智能软件、在线
筛查管理平台和移动电话设备来提供筛查服务。我们
评估了该项目的可持续性、成效和成本。
结果 在 2017 年至 2021 年期间,湖北省 16 个城市的
1,518,972 名女性参加了该项目,且其中 1,474,788 名
女性(占 97.09%)提供了有效筛查样本。在样本呈
阳性的 86,648 名女性中,30,486 名需进行活检,但仅
19,495 名接受了此检查。活检结果显示,2,785 名女性
存在癌前病变现象,且 191 名女性已患有浸润性癌症。
公共付款机构承担的筛查费用为每位女性 6.31 美元
(US$ ):1.03 美元管理费用和 5.28 美元在线筛查费用。
结论 湖北省利用人工智能技术进行宫颈癌筛查打造了
一种既具有低成本又方便有效的服务,将有助于在中
国普遍推行宫颈癌症筛查。
Résumé
Dépistage du cancer du col de l'utérus assisté par intelligence articielle en Chine
Objectif Déployer et évaluer un programme en ligne de grande
envergure destiné au dépistage du cancer du col de l'utérus dans la
province du Hubei, en Chine, reposant sur l'intelligence artificielle et
mis en œuvre par des agents de santé formés à cet effet.
Méthodes Le programme de dépistage, qui a débuté en 2017,
s'appuyait sur quatre types d'agents de santé: des agents chargés de
prélever les échantillons, des techniciens chargés de préparer les lames,
des agents chargés d'établir le diagnostic et des cytopathologistes. Les
premiers ont prélevé des échantillons auprès des femmes sur place, les
deuxièmes ont préparé les lames pour la cytologie en phase liquide, les
troisièmes ont identifié les échantillons négatifs et classé les échantillons
positifs selon le système de Bethesda après analyse cytologique à l'aide
d'une intelligence artificielle en ligne, tandis que les cytopathologistes
ont examiné les échantillons positifs et signé les rapports de résultats
en ligne. Le programme était constitué de scanners entièrement
automatisés, d'une intelligence artificielle en ligne, d'une plateforme
de gestion du dépistage en ligne, ainsi que d'appareils de téléphonie
mobile pour les services de dépistage. Nous avons évalué la viabilité,
les performances et le coût d'un tel programme.
Résultats Entre 2017 et 2021, 1 518 972 femmes résidant dans 16
villes de la province du Hubei ont participé au programme, et 1 474
788 (97,09%) d'entre elles ont présenté des échantillons valides pour le
dépistage. Sur les 86 648 femmes dont les échantillons se sont révélés
positifs, 30 486 ont eu besoin d'une biopsie, mais seulement 19 495 en
ont effectué une. La biopsie a montré que 2785 femmes présentaient
des lésions précancéreuses et que 191 étaient atteintes de cancers
invasifs. Le coût du dépistage s'élevait à 6,31 dollars américains (USD)
par femme pour le payeur public: 1,03 USD pour les frais administratifs
et 5,28 USD pour le dépistage en ligne.
Conclusion Le dépistage du cancer du col de l'utérus assisté par
intelligence artificielle dans la province du Hubei est un service
abordable, accessible et efficace, qui contribuera à garantir une
couverture universelle en la matière dans le pays.
2017
Bethesda
151897220212017
147478816
(%97.09)
3048686648
19495
1912785
6.31
1.03($)
5.28
389
Bull World Health Organ 2023;101:381–390| doi: http://dx.doi.org/10.2471/BLT.22.289061
Research
Cervical cancer screening, China
Xingce Zhu et al.
Резюме
Скрининговое обследование рака шейки матки с помощью искусственного интеллекта, Китай
Цель Реализовать и оценить широкомасштабную онлайн-
программу скринингового обследования рака шейки матки с
использованием искусственного интеллекта и привлечением
обученных сотрудников сферы здравоохранения в провинции
Хубэй, Китай.
Методы В начавшейся в 2017 году программе скринингового
обследования участвовали четыре типа сотрудников сферы
здравоохранения: медицинские работники, осуществляющие
отбор образцов, техники, занимающиеся приготовлением
микроскопических препаратов, диагносты и цитологи.
Медицинские работники, осуществляющие отбор образцов,
брали образцы у женщин в исследовательских центрах; техники,
занимающиеся приготовлением микроскопических препаратов,
готовили микроскопические препараты для проведения
жидкостной цитологии; диагносты определяли отрицательные
образцы и классифицировали положительные образцы на
основе классификации Бетесда после цитологической оценки с
использованием искусственного интеллекта в режиме онлайн;
цитологи изучали положительные образцы и подписывали отчеты
о результатах в режиме онлайн. В программе использовались:
полностью автоматизированные сканеры, онлайн-искусственный
интеллект, онлайн-платформа управления скрининговым
обследованием и мобильные телефонные устройства для
предоставления услуг скрининга. Проводилась оценка
устойчивости, эффективности и стоимости программы.
Результаты С 2017 по 2021 год в программе приняли участие
1 518 972 женщины в 16 городах провинции Хубэй, из
которых у 1 474 788 (97,09%) имелись достоверные образцы
для скринингового обследования. Из 86 648 женщин, чьи
образцы оказались положительными, 30 486 потребовалась
биопсия, но провели ее только 19 495. По результатам
биопсии у 2785 женщин были обнаружены предопухолевые
состояния, а у 191 – инвазивный рак. Стоимость скринингового
обследования составляла 6,31 долл. США (US$) на одну женщину
для юридического лица: 1,03 долл. США на административные
расходы и 5,28 долл. США на скрининговое обследование в
режиме онлайн.
Вывод Скрининговое обследование рака шейки матки с
использованием искусственного интеллекта в провинции Хубэй
предоставил недорогую, доступную и эффективную услугу,
которая будет способствовать достижению всеобщего охвата
населения услугами скрининга рака шейки матки в Китае.
Resumen
Detección del cáncer del cuello uterino con ayuda de la inteligencia articial en China
Objetivo Implementar y evaluar un programa de detección del cáncer
del cuello uterino en línea a gran escala en la provincia de Hubei (China),
con el apoyo de la inteligencia artificial y a cargo de trabajadores
sanitarios formados.
Métodos El programa de detección, que comenzó en 2017, utilizó
cuatro tipos de trabajadores sanitarios: trabajadores sanitarios de
muestreo, técnicos de preparación de portaobjetos, trabajadores
sanitarios de diagnóstico y citopatólogos. Los trabajadores sanitarios
de muestreo tomaban muestras de las mujeres in situ, los técnicos
de preparación de portaobjetos preparaban los portaobjetos para la
citología mediante líquidos, los trabajadores sanitarios de diagnóstico
identificaban las muestras negativas y clasificaban las muestras positivas
según el Sistema Bethesda tras la evaluación citológica mediante
inteligencia artificial en línea, y los citopatólogos revisaban las muestras
positivas y firmaban los informes de los resultados en línea. El programa
utilizó escáneres totalmente automatizados, inteligencia artificial en
línea, una plataforma de gestión de la detección en línea y dispositivos
de telefonía móvil para proporcionar servicios de detección. Se evaluaron
la sostenibilidad, el rendimiento y el coste del programa.
Resultados Entre 2017 y 2021, 1 518 972 mujeres de 16 ciudades
de la provincia de Hubei participaron en el programa, de las que
1 474 788 (97,09%) tenían muestras válidas para la detección. De las
86 648 mujeres cuyas muestras dieron positivo, 30 486 necesitaron una
biopsia, pero solo 19 495 se sometieron a ella. La biopsia mostró que
2785 mujeres tenían lesiones precancerosas y 191 cánceres invasivos.
El coste de la detección fue de 6,31 dólares estadounidenses (US$)
por mujer para el organismo público pagador: US$ 1,03 de costes
administrativos y US$ 5,28 de costes de detección en línea.
Conclusión La detección del cáncer del cuello uterino mediante
inteligencia artificial en la provincia de Hubei proporcionó un servicio
de bajo coste, accesible y eficaz, que contribuirá a lograr la cobertura
universal de detección del cáncer del cuello uterino en China.
References
1. Wild C, Weiderpass E, Stewart BW. World cancer report: cancer research
for cancer prevention. Lyon: IARC Press; 2020. Available from: https://
publications .iarc .fr/ 586 [cited 2022 Aug 15].
2. Global strategy towards eliminating cervical cancer as a public health
problem. Geneva: World Health Organization; 2020. Available from: https://
apps .who .int/ iris/ handle/ 10665/ 336583 [cited 2022 Aug 16].
3. Gultekin M, Ramirez PT, Broutet N, Hutubessy R. World Health Organization
call for action to eliminate cervical cancer globally. Int J Gynecol Cancer.
2020 Apr;30(4):426–7. doi: http:// dx .doi .org/ 10 .1136/ ijgc -2020 -001285
PMID: 32122950
4. Canfell K. Towards the global elimination of cervical cancer. Papillomavirus
Res. 2019 Dec;8:100170. doi: http:// dx .doi .org/ 10 .1016/ j .pvr .2019 .100170
PMID: 31176807
5. Zhao YX, Ma L, Ren WH, Song B, Wang LH, Di JL, et al. [Analysis of the
reported data of National Cervical Cancer Screening Program in Rural
Areas in China from 2009 to 2018]. Zhonghua Yi Xue Za Zhi. 2021 Jun
24;101:1863–8. Chinese. PMID: 34192844
6. Zhao F, Qiao Y. Cervical cancer prevention in China: a key to cancer control.
Lancet. 2019 Mar 9;393(10175):969–70. doi: http:// dx .doi .org/ 10 .1016/
S0140 -6736(18)32849 -6 PMID: 30860036
7. Wang SM, Qiao YL. Implementation of cervical cancer screening and
prevention in China–challenges and reality. Jpn J Clin Oncol. 2015
Jan;45(1):7–11. doi: http:// dx .doi .org/ 10 .1093/ jjco/ hyu188 PMID: 25398583
8. Zhao F, Wen Y, Li Y, Tao S, Ma L, Zhao Y, et al. Epidemiologic and health
economic evaluation of cervical cancer screening in rural China. Asian Pac
J Cancer Prev. 2020 May 1;21(5):1317–25. doi: http:// dx .doi .org/ 10 .31557/
APJCP .2020 .21 .5 .1317 PMID: 32458639
9. Qiao J, Wang Y, Li X, Jiang F, Zhang Y, Ma J, et al. A Lancet Commission on 70
years of women’s reproductive, maternal, newborn, child, and adolescent
health in China. Lancet. 2021 Jun 26;397(10293):2497–536. doi: http:// dx
.doi .org/ 10 .1016/ S0140 -6736(20)32708 -2 PMID: 34043953
390 Bull World Health Organ 2023;101:381–390| doi: http://dx.doi.org/10.2471/BLT.22.289061
Research
Cervical cancer screening, China Xingce Zhu et al.
10. Xia C, Hu S, Xu X, Zhao X, Qiao Y, Broutet N, et al. Projections up to 2100
and a budget optimisation strategy towards cervical cancer elimination in
China: a modelling study. Lancet Public Health. 2019 Sep;4(9):e462–72. doi:
http:// dx .doi .org/ 10 .1016/ S2468 -2667(19)30162 -8 PMID: 31493843
11. Xie Y, Tan X, Shao H, Liu Q, Tou J, Zhang Y, et al. VIA/VILI is more suitable
for cervical cancer prevention in Chinese poverty-stricken region: a health
economic evaluation. BMC Public Health. 2017 Jan 25;17(1):118. doi: http://
dx .doi .org/ 10 .1186/ s12889 -017 -4054 -9 PMID: 28122530
12. Bao H, Zhang L, Wang L, Zhang M, Zhao Z, Fang L, et al. Significant
variations in the cervical cancer screening rate in China by individual-level
and geographical measures of socioeconomic status: a multilevel model
analysis of a nationally representative survey dataset. Cancer Med. 2018
May;7(5):2089–100. doi: http:// dx .doi .org/ 10 .1002/ cam4 .1321 PMID:
29573569
13. WHO guideline for screening and treatment of cervical pre-cancer lesions
for cervical cancer prevention. Geneva: World Health Organization; 2021.
Available from: https:// www .who .int/ publications/ i/ item/ 9789240030824
[cited 2022 Aug 16].
14. Tang HP, Cai D, Kong YQ, Ye H, Ma ZX, Lv HS, et al. Cervical cytology
screening facilitated by an artificial intelligence microscope: A preliminary
study. Cancer Cytopathol. 2021 Sep;129(9):693–700. doi: http:// dx .doi .org/
10 .1002/ cncy .22425 PMID: 33826796
15. Li Y, Ma L, Yang C, Chen Z, Zhao Y, Dang L, et al. A study on service capacity
of primary medical and health institutions for cervical cancer screening in
urban and rural areas in China. Chin J Cancer Res. 2019 Oct;31(5):838–48.
doi: http:// dx .doi .org/ 10 .21147/ j .issn .1000 -9604 .2019 .05 .13 PMID:
31814687
16. Cao M, Li H, Sun D, He S, Yu Y, Li J, et al. Cancer screening in China: The
current status, challenges, and suggestions. Cancer Lett. 2021 May
28;506:120–7. doi: http:// dx .doi .org/ 10 .1016/ j .canlet .2021 .02 .017 PMID:
33684533
17. Cervical cancer [internet]. Geneva: World Health Organization; 2022.
Available from: https:// www .who .int/ health -topics/ cervical -cancer [cited
2022 Dec 15].
18. [Cervical cancer screening work programme.] Chinese. Beijing: Department
of Maternal and Child Health of National Health Commission; 2021.
Available at: http:// www .nhc .gov .cn/ fys/ s3581/ 202201/ cad 44d88acca4
ae49e12dab 9176ae21c/ files/ 7a fa7269d02c 4ba793b405 61e8088c27 .pdf
[cited 2022 Dec 15].
19. [Cervical cancer screening work plan.] Chinese. Beijing: National Health
Commission of the People’s Republic of China; 2022. Available from:
http:// www .nhc .gov .cn/ fys/ s3581/ 202201/ cad 44d88acca4 ae49e12dab
9176ae21c/ files/ 7a fa7269d02c 4ba793b405 61e8088c27 .pdf [cited 2022
Aug15].
20. Holmström O, Linder N, Kaingu H, Mbuuko N, Mbete J, Kinyua F, et al.
Point-of-care digital cytology with artificial intelligence for cervical cancer
screening in a resource-limited setting. JAMA Netw Open. 2021 Mar
1;4(3):e211740. doi: http:// dx .doi .org/ 10 .1001/ jamanetworkopen .2021 .1740
PMID: 33729503
21. Wentzensen N, Lahrmann B, Clarke MA, Kinney W, Tokugawa D, Poitras N,
et al. Accuracy and efficiency of deep-learning-based automation of dual
stain cytology in cervical cancer screening. J Natl Cancer Inst. 2021 Jan
4;113(1):72–9. doi: http:// dx .doi .org/ 10 .1093/ jnci/ djaa066 PMID: 32584382
22. Xue P, Ng MTA, Qiao Y. The challenges of colposcopy for cervical cancer
screening in LMICs and solutions by artificial intelligence. BMC Med. 2020
Jun 3;18(1):169. doi: http:// dx .doi .org/ 10 .1186/ s12916 -020 -01613 -x PMID:
32493320
23. Xue P, Tang C, Li Q, Li Y, Shen Y, Zhao Y, et al. Development and validation
of an artificial intelligence system for grading colposcopic impressions and
guiding biopsies. BMC Med. 2020 Dec 22;18(1):406. doi: http:// dx .doi .org/
10 .1186/ s12916 -020 -01860 -y PMID: 33349257
24. Kudva V, Prasad K, Guruvare S. Hybrid transfer learning for classification of
uterine cervix images for cervical cancer screening. J Digit Imaging. 2020
Jun;33(3):619–31. doi: http:// dx .doi .org/ 10 .1007/ s10278 -019 -00269 -1
PMID: 31848896
25. Thrall MJ. Automated screening of Papanicolaou tests: A review of the
literature. Diagn Cytopathol. 2019 Jan;47(1):20–7. doi: http:// dx .doi .org/ 10
.1002/ dc .23931 PMID: 29603675
26. Bao H, Bi H, Zhang X, Zhao Y, Dong Y, Luo X, et al. Artificial intelligence-
assisted cytology for detection of cervical intraepithelial neoplasia or
invasive cancer: A multicenter, clinical-based, observational study. Gynecol
Oncol. 2020 Oct;159(1):171–8. doi: http:// dx .doi .org/ 10 .1016/ j .ygyno .2020
.07 .099 PMID: 32814641
27. Bao H, Sun X, Zhang Y, Pang B, Li H, Zhou L, et al. The artificial intelligence-
assisted cytology diagnostic system in large-scale cervical cancer screening:
A population-based cohort study of 0.7 million women. Cancer Med. 2020
Sep;9(18):6896–906. doi: http:// dx .doi .org/ 10 .1002/ cam4 .3296 PMID:
32697872
28. LeCun Y, Bengio Y, Hinton G. Deep learning. Nature. 2015 May
28;521(7553):436–44. doi: http:// dx .doi .org/ 10 .1038/ nature14539 PMID:
26017442
29. [China population census yearbook 2020]. Beijing: National Bureau of
Statistics; 2022. Chinese. Available from: http:// www .stats .gov .cn/ zs/ tjwh/
tjkw/ tjzl/ 202302/ t20230215 _1908007 .html [cited 2023 Feb 16].
30. Rezende MT, Bianchi AGC, Carneiro CM. Cervical cancer: Automation of Pap
test screening. Diagn Cytopathol. 2021 Apr;49(4):559–74. doi: http:// dx .doi
.org/ 10 .1002/ dc .24708 PMID: 33548162
31. Chen Y, Sylvia S, Wu P, Yi H. Explaining the declining utilization of village
clinics in rural China over time: A decomposition approach. Soc Sci Med.
2022 May;301:114978. doi: http:// dx .doi .org/ 10 .1016/ j .socscimed .2022
.114978 PMID: 35461080
32. Pan X, Dib HH, Wang X, Zhang H. Service utilization in community health
centers in China: a comparison analysis with local hospitals. BMC Health
Serv Res. 2006 Aug 3;6(1):93. doi: http:// dx .doi .org/ 10 .1186/ 1472 -6963 -6 -93
PMID: 16887022
33. Duckett J, Hunt K, Munro N, Sutton M. Does distrust in providers affect
health-care utilization in China? Health Policy Plan. 2016 Oct;31(8):1001–9.
doi: http:// dx .doi .org/ 10 .1093/ heapol/ czw024 PMID: 27117483
34. Zhang W, Ung COL, Lin G, Liu J, Li W, Hu H, et al. Factors contributing
to patients’ preferences for primary health care institutions in China: a
qualitative study. Front Public Health. 2020 Aug 18;8:414. doi: http:// dx .doi
.org/ 10 .3389/ fpubh .2020 .00414 PMID: 33014959
35. Chen H, Liu J, Wen QM, Zuo ZQ, Liu JS, Feng J, et al. CytoBrain: cervical
cancer screening system based on deep learning technology. J Comput Sci
Technol. 2021 Mar;36(2):347–60. doi: http:// dx .doi .org/ 10 .1007/ s11390 -021
-0849 -3
36. Zhu X, Li X, Ong K, Zhang W, Li W, Li L, et al. Hybrid AI-assistive diagnostic
model permits rapid TBS classification of cervical liquid-based thin-layer
cell smears. Nat Commun. 2021 Jun 10;12(1):3541. doi: http:// dx .doi .org/ 10
.1038/ s41467 -021 -23913 -3 PMID: 34112790
37. Morgenstern JD, Rosella LC, Daley MJ, Goel V, Schünemann HJ, Piggott T.
“AI’s gonna have an impact on everything in society, so it has to have an
impact on public health”: a fundamental qualitative descriptive study of the
implications of artificial intelligence for public health. BMC Public Health.
2021 Jan 6;21(1):40. doi: http:// dx .doi .org/ 10 .1186/ s12889 -020 -10030 -x
PMID: 33407254
38. Shen M, Zou Z, Bao H, Fairley CK, Canfell K, Ong JJ, et al. Cost-effectiveness
of artificial intelligence-assisted liquid-based cytology testing for cervical
cancer screening in China. Lancet Reg Health - West Pac. 2023 Mar: 100726.
doi: http:// dx .doi .org/ 10 .1016/ j .lanwpc .2023 .100726
39. McAuley A. Digital health interventions: widening access or widening
inequalities? Public Health. 2014 Dec;128(12):1118–20. doi: http:// dx .doi
.org/ 10 .1016/ j .puhe .2014 .10 .008 PMID: 25458115
40. Neter E, Brainin E. eHealth literacy: extending the digital divide to the realm
of health information. J Med Internet Res. 2012 Jan 27;14(1):e19. doi: http://
dx .doi .org/ 10 .2196/ jmir .1619 PMID: 22357448
41. Veinot TC, Mitchell H, Ancker JS. Good intentions are not enough: how
informatics interventions can worsen inequality. J Am Med Inform Assoc.
2018 Aug 1;25(8):1080–8. doi: http:// dx .doi .org/ 10 .1093/ jamia/ ocy052
PMID: 29788380
42. Zhang D, Zhang G, Jiao Y, Wang Y, Wang P. “Digital dividend” or “digital
divide”: what role does the internet play in the health inequalities
among Chinese residents? Int J Environ Res Public Health. 2022 Nov
17;19(22):15162. doi: http:// dx .doi .org/ 10 .3390/ ijerph192215162 PMID:
36429878
43. Ibrahim H, Liu X, Zariffa N, Morris AD, Denniston AK. Health data poverty: an
assailable barrier to equitable digital health care. Lancet Digit Health. 2021
Apr;3(4):e260–5. doi: http:// dx .doi .org/ 10 .1016/ S2589 -7500(20)30317 -4
PMID: 33678589