An improved classification method for cervical cancer based on ResNet

An improved model named HDA-ResNet, which combines the improved residual network with the dual attention mechanism, was proposed to perform the cervical cancer grade classification based on colposcopy images. Firstly, the maximum pooling layer of the residual network is replaced by a hybrid dilated convolution. Then, the dual attention mechanism is embedded into the improved residual network. Finally, the model is trained and verified on the clinical real dataset provided by the First Affiliated Hospital of Science and Technology of China. The accuracy and precision of experimental classification are up to 93.40% and 94.30%, and the sensitivity and specificity are 95.13% and 96.80%, respectively. The proposed model can be used to improve the prediction of cervical cancer and assist clinicians in making preliminary judgments.