Application of Self-Organizing Artificial Neural Networks on Simulated Diffusion Tensor Images

Figure 2

The pixel with the weight 𝑤_𝑗 at time 𝑡, nearest to the input pixel, is winning and is moving towards this input pixel according to (10) as shown with an arrow in the figure. Its neighbors, 𝑤_𝑗−1 and 𝑤_𝑗+1 at time 𝑡 on the string, are also moving to a lesser extent. The update in the orientation can be recognized through initial weight vector 𝑣[𝑡] and updated resulting vector 𝑣[𝑡+1]. Units close to the winner as well as the winner itself have their weights updated significantly. Weights associated with far away output pixels do not change significantly. The update continues until the sought topology is found and the feature map is consistent.