Deep learning-based Cervical Cancer Classification
Ichrak Khoulqi, Najlae Idrissi · 2022
Cervical Cancer (CC) is one of the most preoccupied cancers in the medical midst, particularly affecting the vulnerable category of women in undeveloped and developing countries due to lack of screening programs and sensitization campaigns. The cervical cancer does not reveal any symptoms during its growth until it reaches advanced stages where by a simple visual diagnosis it can be detected. However, it is too late to be healing; early stage detection cannot be picked up. Thereby, there is a need of precise and reliable diagnosis will help radiologists to give proper and effective treatment. This paper contributes toward an achievement of a cervical cancer detection system based on Deep Learning (DL) dealing with axial and sagittal T2-weighted MRI images to determine the pathological stage of the tumor. Deep learning has presented significant potential in medical image analysis that allows gaining more importance in computer-aided detection system. In this paper, eight deep-learning-based approaches applied to cervical cancer are studied and compared. The proposed pipeline based on two pre-trained Convolutional Neural Networks (CNN) for automatic cervical classification consists of two main steps: Data Augmentation and Transfer Learning. The Sensitivity, Specificity and Accuracy of the best classification model to differentiate malign cases from benign ones were 100%, 100% and 100% respectively, with an AUC of 1.