IJMI: The screening accuracy rate is as high as 92.1%!

According to the World Health Organization, cervical cancer is the second most common cancer among women in the world.
In 2020, there will be 600,000 new cases of cervical cancer worldwide, and about 340,000 patients will lose their lives
.

Cervical cancer is caused by persistent human papillomavirus (HPV) infection .

Cervical cancer is the second most common cancer in women worldwide.

Pap cytology and HPV testing are common non-invasive primary screening methods

Recently, a research team from Suzhou Institute of Biomedical Engineering and Technology, Chinese Academy of Sciences proposed a non-invasive screening method for cervical lesions based on transmembrane fusion cytology, HPV detection and colposcopy image examination results.
The screening accuracy rate is as high as 92.
1%.
, significantly better than a single screening method
.

The findings, titled "Deep learning based cervical screening by the cross-modal integration of colposcopy, cytology, and HPV test", were published in the International Journal of Medical Informatics (IJMI), the official journal of the International Association of Medical Informatics

A research team from the Suzhou Institute of Biomedical Engineering and Technology, Chinese Academy of Sciences proposed a non-invasive screening method for cervical lesions based on transmembrane fusion cytology, HPV detection and colposcopy image examination results.

Research results (Image source: IJMI)

Research Results (Image Source: IJMI) Research Results (Image Source: IJMI)

In this study, the research team used the deep learning method to construct an automatic screening model of colposcopy images, output the probability of lesions in patients, and realized the objective and quantitative interpretation of colposcopy images
.

They also coded the cytology test results and the HPV test results by category, and used the logistic regression method to fuse the category coding with the lesion probability transmembrane state automatically output from the colposcopy image, and finally constructed a comprehensive cervical cancer screening model

The research team included examination data from a total of 2,160 women in the study, including 1,718 with normal or low-grade disease and 442 with high-grade disease or cervical cancer
.

The inclusion criteria are as follows:

1) Cervical cancer screening from 2016 to 2019;

2) Receive cytology, HPV test, colposcopy, colposcopy biopsy
.

The exclusion criteria are as follows:

1) Lack of one of the following cervical images: saline image, acetic acid image, iodine image;

2) The image is occluded or blurred
.

The researchers developed and tested a colposcopy-based deep learning model using three images of saline, acetic acid, and iodine images, combined with a single-image-based model, and used multivariate logistic regression to establish a multi-image-based deep learning model
.

The combined diagnosis model of cytology and HPV detection was established by applying MLR to the results of cytology and HPV detection

After testing, the AUC of the combined diagnostic model of cytology and HPV detection was 0.
837 (Note: the closer the AUC is to 1, the higher the authenticity of the detection method), which is significantly higher than that of single cytology and HPV detection
.

The AUCs of the deep learning models based on saline images, acetic acid images, and iodine images were 0.

Table 1 Diagnostic performance of single test and comprehensive model

Table 1 Diagnostic performance of single test and comprehensive modelTable 1 Diagnostic performance of single test and comprehensive model

Table source: IJMI

Source of table: IJMI Source of table: IJMI

Diagnostic performance of a single test and an integrated model (Image credit: IJMI)

Diagnostic performance of a single test and integrated model (Image credit: IJMI) Diagnostic performance of a single test and integrated model (Image source: IJMI)

Taken together, the findings suggest that the combined model has the best performance with a more balanced sensitivity and specificity than other single models, suggesting that cytology, HPV testing, and colposcopy have synergistic benefits in improving diagnostic performance
.

In addition, the integrated model is visualized as an easy-to-use nomogram that provides a visual representation of disease risk for each woman by summarizing the points corresponding to the variables in the model, which facilitates clinical understanding of the model rationale

In conclusion, combining all types of cervical images helps to improve the performance of colposcopy-based deep learning, and by combining HPV test results, cytology test results and colposcopy-based deep learning models, more accurate cervical screening can be achieved

Original source:

Original source:

Fu L, Xia W, Shi W, et al.

Deep learning based cervical screening by the cross-modal integration of colposcopy, cytology, and HPV test leave a comment here