Artificial Intelligence Enables Precision Diagnosis of Cervical Cytology Grades and Cervical Cancer

Cervical cancer, the fourth most common cancer globally and the fourth leading cause of cancer-related deaths, can be effectively prevented through early screening. Detecting precancerous cervical lesions and halting their progression in a timely manner is crucial. However, accurate screening platforms for early detection of cervical cancer are needed. Therefore, it is urgent to develop an Artificial Intelligence Cervical Cancer Screening (AICS) system for diagnosing cervical cytology grades and cancer.

Study Overview

• Status: Completed
• Conditions: Cervical Cancer; Artificial Intelligence; Diagnostic Platform; Diagnosing Cervical Cytology Grades and Cancer
• Intervention / Treatment: Diagnostic test: Cervical Cancer Artificial Intelligence Screening System

Detailed Description

A total of 16,056 eligible individuals were enrolled in this study, which consisted of a multicenter population-based study and a randomized controlled trial in China. In the initial exploring stage, a competition in cervical cancer screening between cytotechnicians and AI was held to preliminary exploration into the field in Guangzhou, China on August 04, 2019. 15 slides were randomly selected from 108 eligible individuals' WSIs from SYSMH for performance evaluation of the AI-based cervical cancer screening model and cytotechnicians. In the multicenter population-based study, 11,468 individuals' slides for the cervical cytology screening collected from the Sun Yat-sen Memorial Hospital (SYSMH, Guangzhou, China) between January, 2016 and January, 2020, were randomly assigned to the training dataset (n = 9,316) and the internal validation datasets (n = 2,152) in order to train the Cervical Cancer Artificial Intelligence Screening System (CAISS). To enhance the universality of the CAISS in clinical practice, slides from the other two participating hospitals were used as two independent external validation datasets: These included 600 slides from Guangzhou Women and Children Medical Center (GWCMC, Guangzhou, China) and 600 slides from The Third Affiliated Hospital of Guangzhou Medical University (TAHGMU, Guangzhou, China), which were obtained between January 2016 and January 2020. Further, a prospective validation dataset was conducted to distinguish the diagnostic performance of the cytotechnician, CAISS, and CAISS-assisted groups, in which 2,780 eligible slides from 2,780 individuals were obtained and prospectively labeled between August 28, 2020 and October 16, 2020 at SYSMH. In the third stage of this study, a prospective randomized controlled trial was conducted to compare the performance of the cytotechnician, CAISS, and CAISS-assisted groups in SYSMH. Here, 618 slides were collected between August 13, 2020, and December 14, 2020, to build the SYSMH randomized controlled trial. The remaining 608 slides after quality control were randomly assigned (1:1:1) to the CAISS group (n = 201), the cytotechnician group (n = 203), and the CAISS-assisted combination group (n = 204).

• Study Type: Observational
• Enrollment (Actual): 16164

Contacts and Locations

Study Locations

• China
  • Guangdong
    • Guangzhou, Guangdong, China, 510000
      • Guangzhou Women and Children's Medical Center
    • Guangzhou, Guangdong, China, 510000
      • The Third Affiliated Hospital of Guangzhou Medical University
    • Guangzhou, Guangdong, China, 510120
      • Sun Yat-sen Memorial Hospital of Sun Yat-sen University

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: 25 years to 65 years (Adult, Older Adult)
• Accepts Healthy Volunteers: Yes

• Sampling Method: Non-Probability Sample

Study Population

Female patients who were 18 years or older with clear diagnostic results of cervical liquid-based cytological examination were included. All cases were collected from Sun Yat-Sen Memorial Hospital of Sun Yat-Sen University.

Description

Inclusion Criteria:

1. Women Aged 25-65 years old.
2. Availability of confirmed diagnostic results of the cervical liquid-based cytological examination, and satisfactory digital images from the liquid-based cytology pap test: at least 5000 uncovered and observable squamous epithelial cells, samples with abnormal cells (atypical squamous cells or atypical glandular cells and above).

Exclusion Criteria:

1. Unsatisfactory samples of cervical liquid-based cytological examination: less than 5000 uncovered, observable squamous epithelial cells, and more than 75% of squamous epithelial cells affected because of blood, inflammatory cells, epithelial cells over-overlapping, poor fixation, excessive drying, or contamination of unknown components.
2. Women diagnosed with other malignant tumors other than cervical cancer.

Study Plan

How is the study designed?

Design Details

• Observational Models: Cohort
• Time Perspectives: Other

Number of groups / cohorts

6

Cohorts and Interventions

Group / Cohort
Training dataset; 11,468 eligible individuals' slides for the cervical cytology screening collected from the Sun Yat-sen Memorial Hospital (SYSMH, Guangzhou, China) between January 2016 and January 2020, were randomly assigned to the training dataset (n = 9,316) and the internal validation dataset (n = 2,152) in order to train and validate the Artificial Intelligence Cervical Cancer Screening (AICS).
SYSMH internal validation dataset; 11,468 eligible individuals' slides for the cervical cytology screening collected from the Sun Yat-sen Memorial Hospital (SYSMH, Guangzhou, China) between January 2016 and January 2020, were randomly assigned to the training dataset (n = 9,316) and the internal validation dataset (n = 2,152) in order to train and validate the Artificial Intelligence Cervical Cancer Screening (AICS).
TAHGMU external validation dataset; 600 slides from 600 eligible individuals were obtained in the Third Affiliated Hospital of Guangzhou Medical University (TAHGMU, Guangzhou, China) between January 2016 and January 2020, which was used to validate the Artificial Intelligence Cervical Cancer Screening (AICS).
GWCMC external validation dataset; 600 slides from 600 eligible individuals were obtained in Guangzhou Women and Children Medical Center (GWCMC, Guangzhou, China) between January 2016 and January 2020, which was used to validate the Artificial Intelligence Cervical Cancer Screening (AICS).
Prospective validation dataset; A prospective validation dataset was conducted to distinguish the diagnostic performance of the cytopathologists, AICS, and AICS-assisted cytopathologists, in which 2,780 eligible slides from 2,780 individuals were obtained and prospectively labeled between August 28, 2020 and October 16, 2020 at SYSMH.
Randomized controlled trial; A prospective randomized controlled trial was conducted to compare the performance of the cytopathologists, AICS, and AICS-assisted cytopathologists in SYSMH. Here, 618 slides were collected between August 13, 2020, and December 14, 2020, to build the SYSMH randomized controlled trial. The remaining 608 slides after quality control were randomly assigned (1:1:1) to the AICS group (n = 201), the cytopathologists group (n = 203), and the AICS-assisted cytopathologists group (n = 204).

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Area under ROC curve (AUC)	Area under the curve	Diagnostic evaluation will be performed within 1 week when the smear pictures are obtained

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Sensitivity	The true positive rate (TPR) of the diagnostic platform, which is the ratio between the number of positive individuals correctly categorized by platform and the total number of actual positive individuals (%).	Diagnostic evaluation will be performed within 1 week when the smear pictures are obtained
Specificity	The true negative rate (TNR) of the diagnostic platform, which is the ratio between the number of negative individuals correctly categorized by platform and the total number of actual negative individuals (%).	Diagnostic evaluation will be performed within 1 week when the smear pictures are obtained
Accuracy	The quantity of true positive (TP) plus true negative (TN) over the quantity of (TP) plus true negative (TN) plus false positive (FP) plus false negative (FN).	Diagnostic evaluation will be performed within 1 week when the smear pictures are obtained

Collaborators and Investigators

• Sponsor: Sun Yat-Sen Memorial Hospital of Sun Yat-Sen University
• Collaborators: The Third Affiliated Hospital of Guangzhou Medical University; Guangzhou Women and Children's Medical Center
• Investigators: Principal Investigator: Herui Yao, PHD, Sun Yat-sen Memorial Hospital of Sun Yat-sen University

Study record dates

Study Major Dates

• Study Start (Actual): July 1, 2019
• Primary Completion (Actual): December 14, 2020
• Study Completion (Actual): December 14, 2020

Study Registration Dates

• First Submitted: September 6, 2020
• First Submitted That Met QC Criteria: September 10, 2020
• First Posted (Actual): September 16, 2020

Study Record Updates

• Last Update Posted (Actual): August 8, 2023
• Last Update Submitted That Met QC Criteria: August 6, 2023
• Last Verified: August 1, 2023

More Information

Terms related to this study

• Keywords: Diagnostic Platform; Cervical Cytology Grade Diagnosis; Thinprep Cytologic Test; Artificial Intelligence Deep Learning Algorithm
• Additional Relevant MeSH Terms: Neoplasms; Urogenital Neoplasms; Neoplasms by Site; Uterine Neoplasms; Genital Neoplasms, Female; Uterine Cervical Diseases; Uterine Diseases; Female Urogenital Diseases; Female Urogenital Diseases and Pregnancy Complications; Urogenital Diseases; Genital Diseases; Genital Diseases, Female; Uterine Cervical Neoplasms
• Other Study ID Numbers: 2020-KY-114

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: NO
• IPD Plan Description: Requests for the data collected and analyzed in this study will be considered if the application is in line with public benefits and the applicant is willing to sign a data access agreement. Contact can be through the corresponding author.

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No