PAPNET analysis of reportedly negative smears preceding the diagnosis of a high-grade squamous intraepithelial lesion or carcinoma.

One hundred fourteen cervical smears obtained from 18 women developing biopsy-proven high-grade squamous intraepithelial lesions and two with invasive squamous carcinomas were analyzed by two pathologists using the PAPNET neural network-based automated screening system (PAPNET Analyses A and B). The smears were originally reported as negative and had been previously rescreened and reclassified according to The Bethesda System. Using the PAPNET video displays of 128 potentially abnormal cellular images per smear, each reviewer (PAPNET A and B) determined whether a smear required conventional rescreening. Results of the PAPNET triage were compared with the reclassification diagnoses of the smears by conventional microscopy. PAPNET Analysis A selected eight (14%) smears reclassified as negative, 25 (69%) as atypical squamous cells of undetermined significance, and 15 (71%) as squamous intraepithelial lesions (SIL) for rescreening. In PAPNET Analysis A, two (10%) SILs were not selected for rescreening, and four (19%) were considered unsatisfactory for analysis. PAPNET Analysis B selected 21 (37%) smears reclassified as negative, 25 (69%) as atypical squamous cells of undetermined significance, and 18 (86%) as SIL for review. In PAPNET Analysis B, two (10%) SILs were missed, and one (5%) smear was unsatisfactory for analysis. Each PAPNET analysis selected smears for rescreening in 19 (95%) of 20 patients and detected SILs in 10 patients that were missed in the original screening. Using PAPNET, SILs would have been detected a median of 56 months (PAPNET A) and 62 months (PAPNET B) before their actual discovery. These preliminary data suggest that PAPNET may help detect SILs missed in routine cytologic screening.(ABSTRACT TRUNCATED AT 250 WORDS)