Significant reduction in the rate of false-negative cervical smears with neural network-based technology (PAPNET testing system)

Leopold G. Koss, Mark E. Sherman, Michael B. Cohen, Allen R. Anes, Teresa M. Darragh, Luciano B. Lemos, Betty Jane Mcclellan, Dorothy L. Rosenthal, Sedigheh Keyhani-Rofagha, Klaus Schreiber, Philip T. Valente

Research output: Contribution to journalArticlepeer-review

49 Scopus citations


False-negative cervical Pap smears may lead to disability or death from carcinoma of the uterine cervix. New computer technology has led to the development of an interactive, neural network-based vision instrument to increase the accuracy of cervical smear screening. The instrument belongs to a new class of medical devices designed to provide computer-aided diagnosis (CADx). To test the instrument's performance, 487 archival negative smears (index smears) from 228 women with biopsy-documented high-grade precancerous lesions or invasive cervical carcinoma (index women) were retrieved from the files of 10 participating laboratories that were using federally mandated quality assurance procedures. Samples of sequential negative smears (total 9,666) were retrieved as controls. The instrument was used to identify evidence of missed cytological abnormalities, including atypical squamous or glandular cells of undetermined significance (ASCUS, AGUS), low-grade or high-grade squamous intraepithelial lesions (LSIL, HSIL) and carcinoma. Using the instrument, 98 false-negative index smears were identified in 72 of the 228 index women (31.6%, 95% confidence interval [CI]: 25% to 38%). Disregarding the debatable categories of ASCUS or AGUS, there were 44 women whose false-negative smears disclosed squamous intraepithelial lesions (SIL) or carcinoma (19.3%; 95% CI: 14.2% to 24.4%). Unexpectedly, SILs were also identified in 127 of 9,666 control negative smears (1.3%; 95% CI: 1.1% to 1.5%). Compared with historical performance data from several participating laboratories, the instrument increased the detection rate of SILs in control smears by 25% and increased the yield of quality control rescreening 5.1 times (P < 0.0001). These data provide evidence that conventional screening and quality control rescreening of cervical smears fail to identify a substantial number of abnormalities. A significant improvement in performance of screening of cervical smears could be achieved with the use of the instrument described in this report.

Original languageEnglish (US)
Pages (from-to)1196-1203
Number of pages8
JournalHuman Pathology
Issue number10
StatePublished - 1997


  • Cervical cytology
  • Cervical smear screening
  • Computer-aided diagnosis
  • False-negatives
  • Neural networks
  • Papanicolaou smears

ASJC Scopus subject areas

  • Pathology and Forensic Medicine


Dive into the research topics of 'Significant reduction in the rate of false-negative cervical smears with neural network-based technology (PAPNET testing system)'. Together they form a unique fingerprint.

Cite this