You have requested a machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Neither SPIE nor the owners and publishers of the content make, and they explicitly disclaim, any express or implied representations or warranties of any kind, including, without limitation, representations and warranties as to the functionality of the translation feature or the accuracy or completeness of the translations.
Translations are not retained in our system. Your use of this feature and the translations is subject to all use restrictions contained in the Terms and Conditions of Use of the SPIE website.
9 May 2002Optimal neural network architecture selection: effects on computer-aided detection of mammographic microcalcifications
Metin Nafi Gurcan,1 Heang-Ping Chan,1 Berkman Sahiner,1 Lubomir M. Hadjiiski,1 Nicholas Petrickhttps://orcid.org/0000-0001-5167-8899,1 Mark A. Helvie M.D.1
We evaluated the effectiveness of an optimal convolution neural network (CNN) architecture selected by simulated annealing for improving the performance of a computer-aided diagnosis (CAD) system designed for the detection of microcalcification clusters on digitized mammograms. The performances of the CAD programs with manually and optimally selected CNNs were compared using an independent test set. This set included 472 mammograms and contained 253 biopsy-proven malignant clusters. Free-response receiver operating characteristic (FROC) analysis was used for evaluation of the detection accuracy. At a false positive (FP) rate of 0.7 per image, the film-based sensitivity was 84.6% with the optimized CNN, in comparison with 77.2% with the manually selected CNN. If clusters having images in both craniocaudal and mediolateral oblique views were analyzed together and a cluster was considered to be detected when it was detected in one or both views, at 0.7 FPs/image, the sensitivity was 93.3% with the optimized CNN and 87.0% with the manually selected CNN. This study indicates that classification of true positive and FP signals is an important step of the CAD program and that the detection accuracy of the program can be considerably improved by optimizing this step with an automated optimization algorithm.
The alert did not successfully save. Please try again later.
Metin Nafi Gurcan, Heang-Ping Chan, Berkman Sahiner, Lubomir M. Hadjiiski, Nicholas Petrick, Mark A. Helvie M.D., "Optimal neural network architecture selection: effects on computer-aided detection of mammographic microcalcifications," Proc. SPIE 4684, Medical Imaging 2002: Image Processing, (9 May 2002); https://doi.org/10.1117/12.467095