Back

Cervical Cancer Screening Using Residual Learning

Abdulsamad, M. I.; Alshareef, E. A.; Ebrahim, F.

2022-12-20 oncology
10.1101/2022.12.19.22283665 medRxiv
Show abstract

Cervical cancer is a leading cancer in the female population. This disease is considered dangerous as its slow and unpredicted growth. The prevention of such cancer can be mostly achieved by screening its transformation zones. The cervical pre-cancerous zones can be considered as three types: type 1, type 2, and type 3. Screening and analyzing these three stages can be crucial for preventing their transformation into cancer. Hence, it is essentially important to have an automated and intelligent system that can grade the cervical pre-cancerous colposcopy images into one of the three types. This can help in providing the right treatment and prevent cancer transformation. In this paper, we develop a residual learning-based model (ResNet-50) to be trained for classifying the type of a colposcopy cervical image into type 1, type 2, and type 3. Experimentally, the model was fine-tuned and evaluated on a public dataset of colposcopy cervical images and achieved promising results in cervical cancer screening of accuracy of 77% and F1-score of 79%.

Matching journals

The top 5 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.