Feature extraction and classification of precancerous cervix lesions
Hernes, Dana L.
MetadataShow full item record
Cervical cancer is the second- most common type of cancer in women worldwide. Although great strides have been made in the prevention of invasive cervical cancer, these measures are not available to all women. Because trained personnel are limited in developing countries, the death rate is significantly higher than in developed countries like the United States. Therefore it is necessary to develop a method to automatically determine the region of abnormal cervix tissue to biopsy and have the methodology available to worldwide areas. This paper discusses the techniques investigated to create a fully automated system to locate precancerous regions in an image of a cervix generated by a digital colposcope or cerviscope. Segmentation was used to first isolate the acetowhite region, the region of interest, from the remainder of the image for further processing. Automatic identification of some precancerous markers within the acetowhite region was developed by extracting significant features. Some of the precancerous markers take the form of mosaicism and punctation. Color and geometric features based classification was used to locate these regions. The k-means clustering technique was applied in classifying mosaic regions from normal regions within a digital cervix image.