Coherent optical processing of cervical cytologic samples.

Abstract

Earlier research using a very limited data base gave encouraging results for the automated screening of exfoliated cytologic samples using coherent optical processing techniques to examine individual isolated cells. A more thorough investigation involving a larger data base has confirmed our initial results. This investigation was performed using a specially designed Fourier spectrum analyzer and a solid state optical detector array. An analysis was made to determine the performance of a screening system using such a cell-by-cell discriminating device. This analysis indicated that less than 20,000 cells would have to be examined to obtain a system performance level of 1% false negative and 10% false positive error rates with a 1% probability of occurrence of malignant cells in a malignant sample. This performance figure was inferred from measured statistical performance characteristics of a laboratory cell-by-cell screening device using optically generated Fourier transfrom techniques for cell discrimination. The performance of the system was shown to be much more sensitive to cell-by-cell false error rates than false negative error rates. It was also found that the majority of false positive errors were due to misclassifying parabasal cells as malignant. By eliminating parabasal cells, which comprised more than 25% of our normal cell data base, the number of cells needed to be screened dropped by an order of magnitude. It was also shown that there is an inverse quadratic relationship between the percentage of malignant cells in a malignant sample and the number of cells that must be screened to achieve any desired system performance.

The tradeoff of cell classifier error rates

“…The cell classifier examines some number (N) of cells and generates an estimate q = K / N of the abnormal cell proportion, where K is the number of cells classified as abnormal. The second classifier tests the hypothesis that q corresponds to an abnormal specimen (1,3).…”

Effect of random abnormal cell proportion on specimen classifier performance

“…For this analysis, one considers the model (4,(6)(7)(8)(9)11) shown in Figure 1. We assume that there are only two classes of cells, normal and abnormal, and that there are two classes of specimen: Normals, which contain only normal cells, and abnormals, which contain some fixed proportion p of abnormal cells.…”

“…Several multi-class models have been presented (2,5,13). However, considerable basic understanding of the behavior of such systems can be gained from the analysis of simplified models (4,(6)(7)(8)(9)11,12).The doubly two-class model applies to prescreening instruments that operate by first classifying some number of cells and then, from those results, the specimen. A series of papers studying that model has elucidated several important principles (4,6-9,111.…”

“…Several multi-class models have been presented (2,5,13). However, considerable basic understanding of the behavior of such systems can be gained from the analysis of simplified models (4,(6)(7)(8)(9)11,12).…”

“…The doubly two-class model applies to prescreening instruments that operate by first classifying some number of cells and then, from those results, the specimen. A series of papers studying that model has elucidated several important principles (4,(6)(7)(8)(9)111 In this paper we extend the basic model to account for the case where the proportion of abnormal cells in an abnormal specimen is not fixed, but is distributed according to some known probability distribution. We also show that a previously reported (11) fundamental lower limit on the specimen classifier false negative rate does not, in fact, exist.…”