Artificial intelligence-assisted cervical dysplasia detection using papanicolaou smear images

Cervical dysplasia is a cancerous condition, and it is essential to correctly identify them from Pap smear images using machine intelligence. Regular screening and early diagnosis is the most vital step for detecting dysplastic stage, so as to treat them effectively. However, the manual inspection of Papanicolaou test screened under microscope is laborious, subjective and time-consuming task. Therefore, the objective of this research was to develop an artificial intelligence-enabled assistive tool to detect the cervical dysplasia cancer. Here, the pixel-based segmentation to classification mapping approach is introduced which is the two-step classification, i.e. cell segmentation and cell classification. In cell segmentation stage, the novel filter to feature map approach is used. Total 112 filtered images were generated from each original cell images. The feature vector was then created for every original pixel using filtered images. In Dysplasia cancer classification stage, the 163 features consisting the edge detector, texture, noise, membrane detector and colour features are considered. Three classifiers, namely artificial neural network (ANN), support vector machine (SVM) and random forest (RF), are used to detect and diagnose the dysplasia stage cancer. These classifiers are evaluated for performance using seven different performance measures. For cell segmentation approach, the RF reported accuracy of 99.07% and it outperformed in terms of accuracy over ANN and SVM classifiers. Finally, the cervical dysplasia is accurately identified with 97.5% accuracy using ANN as compared to SVM and RF.