- 296 Downloads
The results with the baseline SVM approach show that distinguishing between MEP and GMP using the morphology of the cells is possible. We achieve a good accuracy even though the time series is averaged for each feature to only a few time steps. However, the features had to be normalized by their mean and standard deviation of each movie separately, as not doing so lowers the AUC by about 10%. This procedure makes the features more robust.
KeywordsCell Image Recurrent Neural Network Image Patch Unsupervised Approach Convolutional Layer
Unable to display preview. Download preview PDF.