MEASURING THE ACCURACY OF AUTOMATED CERVICAL CYTOLOGY PRESCREENING SYSTEMS BASED ON IMAGE ANALYSIS
Trials for automated cervical cytology prescreening systems must be designed primarily to determine the performance (both false negative and false positive rates) for the system. Because of the varying importance of errors at different stages of disease, error rates must be quoted for specimens with varying cytological grades. Also, the statistical reliability levels of all results should be stated. Trials must be carried out in phases during the design and final evaluation of a system. In the design of a trial phase, a wide number of factors must be considered. Those discussed include the trial strategy; the required statistical reliability; the size, make-up and representativeness of the patient sample; the ethics of the trial; proper experimental techniques; documentation; and cost-effectiveness. Two possible strategies are discussed, one suitable for early trial phases during the development of a system, and one suitable for final field-trial evaluation of a system. Levels of statistical reliability required for various trial phases are presented, and statistical methods are developed to determine the patient sample size required to achieve these levels of statistical reliability.
Bibliographic Reference: EUR 11451 EN (1988) FS, PP 52, ECU 5.00, AVAILABILITY: EUROFFICE LUXEMBOURG, BP 1003, GDL
Availability: Can be ordered online
Record Number: 1989126091600 / Last updated on: 1989-05-01
Available languages: en