[ CancerWeb Home
| Comments
| Up One Level ]
Original Summaries of Selected CANCERLIT Abstracts.
Screening For Breast Cancer: Automated Analysis vs Human Observation
Last modified on:
Tuesday, April 20, 1999 10:52:30
Copyright © 1994-2008, Information Ventures, Inc.
Humans are very effective at tasks requiring observation, but are
less efficient than computers in making weighted decisions based on
multiple pieces of information. This latter situation is well
illustrated by the decision making based on fine needle aspirates,
in which a range of cytological variables must be considered in
order to arrive at a diagnosis. A study from the University of
Sheffield Medical School in the UK (Cross; ICDB/95612240), tested
whether multivariate analysis of human observations could improve
the diagnostic accuracy of fine needle aspirates of breast lesions.
The study used 209 malignant and 170 benign aspirates, whose final
diagnosis was confirmed by the histology, clinical examination,
mammography and follow-up. Ten features suggesting malignancy were
studied. Multivariate analysis used a process known as dichotomous
logistic regression (DCL). The human observer also gave a benign
or malignant diagnosis for each lesion with overall accuracy of
91.6%, specificity for malignancy of 100% and sensitivity of 84.7%.
Using DCL on all features gave an accuracy of 96.0%, a specificity
of 96.5%, and a sensitivity of 95.7%. These results indicate that
multivariate analysis of human observations can give better results
than human judgment based on the same observations.
November, 1995
Copyright (c) 1994-2008, Information Ventures, Inc.
Mail us at: Customer-Service@infoventures.com
http://infoventures.com