Pap smear-based cervical cancer detection using hybrid deep learning and performance evaluation

Cervical cancer causes the abnormal growth of the cells at the cervix region of the woman's uterus. It is the second most common type of cancer after the breast cancer seen in women. Detection of cervical cancer at the early stage is vital. Various screening methods like Pap smear, Colposcopy and HPV testing are carried out for the detection of cervical cancer. The possible screening techniques to diagnose cervical cancer include the visual inspection of the cervix (VIA), Pap smear examination (cytology), Colposcopy, Biopsy and HPV-DNA detection. All these techniques need the involvement of an expert doctor and/or pathologist. Also, a cancer diagnosis is a subjective process where the experience and training of pathologists are significant factors. Intelligent and automated screening systems will be helpful in such scenarios. This paper presents the methodology for cervical cancer prediction based on pap smear images. Pre-trained deep neural network models are used for feature extraction and different machine-learning (ML) models are trained on extracted features. In the proposed methodology, four pre-trained models such as Alexnet, Resnet-18, Resnet-50 and Googlenet are fine-tuned for feature extraction followed by the different ML algorithms. Among these simple logistic regression, the algorithms have performed best with the highest accuracy of 95.14% with the Alexnet pre-trained model.