Histologic Validation and Diagnostic Consistency of Neural Network-Assisted Cervical Screening: Comparison with the Conventional Procedure
Mathilde E. Boon, Lambrecht P. Kok, S. Beck, Myrthe R. Kok · Birkhäuser Boston eBooks · 1996
For the cytologist, neural network-assisted screening reduces the amount of redundant visual information; however, there remains the question of how this influences the diagnostic performance in large-scale screening programs. In a routine setting, 78010 cervical smears were screened in a period of 18 months. Of these cases 42134 were screened conventionally. The other 35 876 smears were screened using the PAPNET®-assisted method in which neural network technology was a key feature. We performed a histologic validation of the two screening processes. The PAPNET-assisted screening resulted in significantly higher positive histologic scores for carcinoma and invasive carcinoma. In addition, the accuracy rates for these lesions increased. This histologic validation indicates that, using neural network technology, the value of the cytologic detection of severe abnormalities can be improved significantly. In an additional study concerning 77222 smears, it was found that not only did the mean positive scores of the seven screeners decrease through PAPNET, but that the coefficients of variability also (CV) became smaller as well. Due to the improvement of the performance of all cytologists involved, diagnostic consistency was enhanced by neural network technology.