Automatic scanning for cervical smears.

In approaching the question of automatic recognition of malignant cells, it is necessary first to make one point about neoplasia in general. It would be convenient to have a detectable or measurable character which is specific for malignant cells, for instance a staining reaction. The fact that none has ever been found is itself an important datum which tells us something about cancer. There are plenty of specific characters for various normal cellsthings like secretion of insulin or keratin, or the possession of basophil granules or cilia: all these differentiated characters are specific because the cells are working to a program. In malignant transformation the various programs are lost or debased, and probably no common feature is to be expected. Instead, we find the utmost diversity. The above statements do not by any means imply that an experienced observer cannot ever identify malignant cells. In some situations, and this includes the cervix uteri, the cells shed from a carcinoma are completely different in appearance from the normal epithelial cells, though they differ greatly among themselves. In discussions about the possibility of automatic screening an engineer is apt to be baffled because he cannot get a straight answer to his question, 'What do you notice when you identify a cell as malignant?' There are therefore two sides to the problem of automatic screening for malignant cells. The first is the matter of deciding on one or more marker characters or parameters, and how to use or interpret them so as to screen specimens. The second is the technical one, how to serve up the cells to the sensing device and how to measure the parameters chosen at a speed high enough to satisfy population screening requirements. Both these aspects involve peculiar difficulties and these will be discussed below. But at first it is worth examining the standard against which our hypothetical machine will be measured.