NEURAL-NETWORK PROCESSING CAN PROVIDE MEANS TO CATCH ERRORS THAT SLIPTHROUGH HUMAN SCREENING OF PAP SMEARS

The problem of the false-negative smear deserves the attention of the cytologic community. We found that by using the PAPNET system, equippe d with neural network programming, cancer cells can be detected in rep eatedly misdiagnosed Pap smears. The correct diagnosis of these false- negative smears depended on the skills of the diagnostician to recogni ze the cells on one or more of the 128 tiles of the video display as a bnormal, and to make the proper decision when to turn to light microsc opy. In 8 of the 10 tested false-negative smears, the cancer cells wer e found exclusively in epithelial fragments; in two cases there were n o more than five abnormal cells in the smear (which were detected by P APNET). This study gave us insight into the nature of the false-negati ve problem and showed us that the application of neural network proces sing can provide means to catch errors that slip through human screeni ng of Pap smears. (C) 1993 Wiley-Liss, Inc.