Quantitative detection of cervical cancer based on time series information from smear images

Existing cervical cancer detection methods usually screen the samples based on separated cells. Cell misclassification leads to poor robustness and accuracy, quantitative analysis is missed. Global smear information and cell relationship are also not fully utilized. Aiming at the mentioned limitations, a cervical quantitative detection framework which combines the fine-tuned Long Short-Term Memory Fully Convolutional Network and Fuzzy Nonlinear Regression is proposed. Time series-based screening improves the detection performance. Deoxyribonucleic Acid (DNA) value is better expressed by the rectified method using soft computing. A cervical dataset containing 657 samples is used for training and validation, accuracy, sensitivity, and specificity of 98.3%, 98.1% and 97.9% are achieved with the time series features, providing an automatic and effective way for the computer-assisted screening of cervical cancer. (C) 2021 Elsevier B.V. All rights reserved.

Automated summary

A cervical quantitative detection framework combining fine-tuned Long Short-Term Memory Fully Convolutional Network and Fuzzy Nonlinear Regression is proposed to improve the detection performance of cervical cancer, achieving high accuracy and specificity through time series-based screening.