6. Conclusion
This work describes mELMs. The proposed method uses neural networks with hidden layers of random weights, with nodes based on nonlinear morphological operators of dilation and erosion. The main motivation was to construct new methods to support diagnosis by imaging to detect and classify breast lesions in mammogram images. The main objective of the proposed method was to combine (1) the ability of morphological neural networks to build complex decision surfaces with (2) the simplicity of training algorithms of hidden layer neural networks with random weights, such as ELMs.
In the experiments performed with IRMA mammography images, 2796 images were employed, organized into four types of tissues: adipose, fibrous, heterogeneously ...