Effect of Training Artificial Neural Networks on 2D Image: An Example Study on Mammography
Xuejun Zhang, Hiroshi Fujita, Jing Chen, Zuojun Zhang · 2009
Several structures of artificial neural networks (ANNs) with different training patterns were investigated so as to compare their performances on detecting the cluster of microcalcifications (CM) on mammography. 150 region-of-interests (ROIs) around mass containing both positive and negative microcalcifications were selected for training the network by a standard or modified error-back-propagation algorithm. A rule-based triple-ring filter (TRF) was used for evaluating the performances of these two different types of methods. The results showed that the shift-invariant artificial neural network (SIANN) was the best ANN model to detect CM, while SIANN and TRF had different ability of detecting microcalcifications. In a practical detection of 30 cases with 40 clusters in masses, the sensitivity of detecting CMs was improved from 90% by our previous method to 95% by using both SIANN and TRF.