Spectral analysis of cervical cells using the discrete Fourier transform.

This paper contains results from a preliminary study of spectral analysis techniques applied to the classification of cervical cells from routinely prepared Papanicolaou cervical smears. Experiments were conducted using a subset of 110 normals and 110 dyskaryotic single cell images randomly selected from a larger cell image data base. An assessment was made of the contribution of different regions within a cell image to the frequency spectrum. Three image sets were used, the original image itself plus two derived from it. In the first derived set, only nuclear size and shape were used. In the second set nuclear morphology and texture were included. Nuclear masking was performed using an interactive segmentation procedure. The discrete Fourier transform was applied to each image in the three image sets and classification experiments were performed using 80 features derived from the frequency spectra. An optimum set of features was selected for each experiment by canonical analysis. Good classification results were obtained when features extracted solely from nuclear shape were used. The inclusion of information relating to nuclear texture improved the results. However, inclusion of the extra nuclear region degraded the classifier's ability to discriminate between cell groups.

Automation in cervical cytology: an overview.

Cancer of the uterine cervix is one of the more common female cancers and a major source of premature female mortality. UK deaths in 1987 exceeded 2000. To reduce these rates, national screening programmes have been introduced using the Papanicolaou method. In screening for cancer, the examination of specimens is an intensive and expensive task. Its high cost has led to a number of attempts to automate the process, either fully or partially. Over the last 30 years, various experimental prescreening systems have been developed for the diagnosis of cytological samples including the use of image processing techniques. This paper presents a historical overview of automation in cervical cytology and the status of current developments for automated cytological diagnosis.