A Review of Applications of Image Analysis and Machine Learning Techniques in Automated Diagnosis and Classification of Cervical Cancer from Pap-smear Images

Cervical cancer ranks as the fourth most prevalent form of cancer affecting women worldwide and its early detection provides the opportunity to help save life. Automated diagnosis and classification of cervical cancer has become a necessity as it enables accurate, reliable and timely analysis of the condition's progress. This survey paper presents an overview of the state of the art as articulated in a number of prominent recent publications focusing on automated diagnosis and classification of cervical cancer from pap-smear images. It reviews thirty journal papers obtained electronically through four scientific databases searched using three sets of keywords: (1) Segmentation, Classification, Cervical Cancer; (2) Medical Imaging, Machine Learning, pap-smear Images; (3) Automated, Segmentation, Pap-smear Images. The review found that some techniques are used more frequently than others are: for example, filtering, thresholding and KNN are the most used techniques for preprocessing, segmentation and classification of pap-smear images. It has also been observed that the superiority of the results of a classification algorithm over the other greatly depends on a number of factors which include: the set of features selected, the accuracy of the segmentation, the type of pre-processing techniques used and the type of datasets used. Most of the existing algorithms result in an accuracy of nearly 93.78% on open pap-smear data set segmented using commercial digital image segmentation softwares. K-Nearest-Neighbours has been reported to be an excellent classifier for cervical images giving an accuracy of over 99.27% for a 2-class classification problem. The reviewed papers indicate that there are still weaknessess in the available techniques that result in low accuracy of classification in some classes of cells. This accuracy can be improved by extracting more features, improvement in noise removal, and using hybrid segmentation and classification techniques. Moreover, most of the developed classifiers are developed and tested on accurately segmented images using commercially available softwares. There is thus a deficit of evidence that these algorithms will work in clinical settings found in developing countries where 85% of cervical cancer incidences occur.