Screening of cervical cytological samples using coherent optical processing Part 1

Abstract

Coherent optical signal processing methods for screening Pap smears were evaluated and are presented in a three-part sequence. In Part 1, 2-D Fourier spectra of normal and abnormal cells generated from many high resolution cell photographs are presented. Each cell spectrum was measured with a coherent optical data processing system containing a special geometry detector and automated data collection capability. Several parameters, determined from weighted measures of the cell transform intensity variations, were tested as feature discriminators to separate normal from abnormal cells. An analysis of the experimental data demonstrates that several transform features are good discriminators of normal/abnormal cells (using standard Baysian decision algorithms with quadratic decision rules). In Part 2, mathematical model studies to guide and validate the experimental work show certain transform parameters to be functionally related to cell and nucleus diameter and optical density. Other parameters appear to be related to cell characteristics such as clumping of nuclear deoxyribonucleic acid (DNA). The model studies also show that the photographic variables play a key role in cell image preprocessing prior to Fourier analysis. Part 3 discusses an optical transducer that was used as a film replacement to modulate a laser beam spatially with a cell image. Several of the feature discriminators used in Part 1 with photographic film images served also to separate normal and malignant cell types when the cell Fourier spectrum was obtained from a transducer image. Based on an understanding of the procedure from the model studies and the demonstrated ability to separate normal and malignant cells using certain transform features, a coherent optical processing system to screen Pap smears from cell or slide photographs is feasible and appears practical in terms of the number of cells to be processed. A high-speed optical transducer would be required for processing large numbers of cells without photography in a reasonable time interval.