Artificial intelligence enables precision diagnosis of cervical cytology grades and cervical cancer

Cervical cancer is a significant global health issue, its prevalence and prognosis highlighting the importance of early screening for effective prevention. This research aimed to create and validate an artificial intelligence cervical cancer screening (AICCS) system for grading cervical cytology. The AICCS system was trained and validated using various datasets, including retrospective, prospective, and randomized observational trial data, involving a total of 16,056 participants. It utilized two artificial intelligence (AI) models: one for detecting cells at the patch-level and another for classifying whole-slide image (WSIs). The AICCS consistently showed high accuracy in predicting cytology grades across different datasets. In the prospective assessment, it achieved an area under curve (AUC) of 0.947, a sensitivity of 0.946, a specificity of 0.890, and an accuracy of 0.892. Remarkably, the randomized observational trial revealed that the AICCS-assisted cytopathologists had a significantly higher AUC, specificity, and accuracy than cytopathologists alone, with a notable 13.3% enhancement in sensitivity. Thus, AICCS holds promise as an additional tool for accurate and efficient cervical cancer screening.

Moving to the validation phase, we initially validated the AICCS using the SYSMH internal validation dataset (n = 2,152), followed by the GWCMC (n = 600) and TAHGMU (n = 600) external validation datasets.We prospectively collected cervical cytology images from an additional 2,780 eligible participants at SYSMH to further assess the AICCS's generalizability and robustness in clinical practice.AICCS, Artificial Intelligence Cervical Cancer Screening System.SYSMH, Sun Yat-sen Memorial Hospital.GWCMC, Guangzhou Women and Children's Medical Center.TAHGMU, The Third Affiliated Hospital of Guangzhou Medical University.
Supplementary Figur e 5.The AICCS system assists the diagnosis of cer vical cytology gr ades.
For patients coming to a collaborating hospitals for a thin-prep cytologic test, the system is able to provide two key clinical applications as follows: 1) the precision diagnosis of cervical cytology grades can be obtained with the assistance of the AICCS, with the cervical cytology smears uploaded to the AICCS to increase the accuracy of diagnosis; and 2) the AICCS provides free access as a consulting service for patients and clinicians after their have uploaded their WSI on the AICCS.Experienced experts can discuss complex cases and reach consensus in diagnosis.AICCS, Artificial Intelligence Cervical Cancer Screening System; 5G, Fifth generation.
Supplementary Figur e 6.An example of quality contr ol.
Quality control measures were instituted by conducting a thorough assessment of participant eligibility and adhering to stringent criteria for specimen selection.Within the framework of the AICCS, an AI-assisted methodology was employed to identify and address potential issues related to scanning quality throughout the digitization process.To support this objective, an image classification model was developed.This model leverages thumbnail images of WSIs to detect occurrences of scanning quality impediments, including but not limited to blurriness (A) and instances of incomplete scanning.
Lastly, the WSIs pass quality control move on to AICCS system (B).

Supplementary Table 2. Performance of four deep learning algorithms in cervical cytopathological diagnosis
Figure legends Supplementary Figure 1.Study design of the training and validation datasets.

Supplementary Figure 7 .
Heat maps to visualize the outputs of the patch-level detection model.ASC-US, Atypical squamous cells of undetermined significance.LSIL, Low-grade squamous intraepithelial lesions.ASC-H, Atypical squamous cells -cannot exclude HSIL.HSIL, High-grade squamous intraepithelial lesions.SCC, Squamous cell carcinoma.AGC; Atypical glandular cells.

(
expected to represent less than 10 % of all ASC interpretations) in which the cytologic changes are suggestive of HSIL HSIL High-grade squamous intraepithelial lesion HSIL HSIL+ The cells are smaller and show less cytoplasmic maturity than cells of LSIL, nuclei are generally hyperchromatic and coarsely granular, and contour of the nuclear membrane is quite irregular SCC Squamous cell carcinoma SCC HSIL+ An invasive epithelial tumor composed of squamous cells of varying degrees of differentiation, some of which show the nuclear features of HSIL AGC-NOS Atypical glandular cells, not otherwise specified AGC AGC Atypical glandular cells (AGC) include atypical endocervical and endometrial cell lesions, which including AGC-NOS, AGC-FN, AIS,

Table 4 . Distribution of cervical cytology grades in the training and validation datasets
A total of 9,316 images from SYSMH were assigned into the training dataset, and 2,152 images from SYSMH were assigned into the internal validation dataset for algorithms' development and evaluation.WSI, Whole-slide Image.RCNN, Region Convolutional Neural Networks.DNN, Deep Neural Networks.NPV, Negative predictive values.PPV, Positive predictive values.AUC, Area under the curve.Lr, Learning rate.Cv, Cross validation.Sgd, Stochastic gradient descent.Source data are provided as a Source Data file.Supplementary Table 3. Abbreviations and definitions.SYSMH, Sun Yat-sen Memorial Hospital; GWCMC, Guangzhou Women and Children Medical Center.TAHGMU, The Third Affiliated Hospital of Guangzhou Medical University.NILM, Negative for intraepithelial lesion or malignancy.ASC-US, Atypical squamous cells of undetermined significance.LSIL, Low-grade squamous intraepithelial lesions.ASC-H, Atypical squamous cells -cannot exclude HSIL.HSIL, High-grade squamous intraepithelial lesions.SCC, Squamous cell carcinoma.AGC; Atypical glandular cells.Source data are provided as a Source Data file.Supplementary

Table 5 . Distribution of cervical cytology grades in the randomised observational trial
Atypical glandular cells.Source data are provided as a Source Data file.

Supplementary Table 10. Abbreviation list and morphology of each classification.
A designation reserved for the minority of atypical squamous cells (ASC) cases