Automatic Neuron Classification via Persistence Diagram of Neuron
- Slides: 9
Automatic Neuron Classification via Persistence Diagram of Neuron Trees Yanjie Li Advisor: Yusu Wang
Problems • Identifying neuron types • Neuron types have a tight relation with their geometric shapes
Persistence Diagram of Neuron Trees • Neuron cells can be represented by some tree structures.
Persistence Diagram of Graph Slides 73 from ‘Some applications of topological persistence’
Distance Between Two Neuron Trees 1 • Bottleneck distance & Wasserstein distance
Distance Between Two Neuron Trees 2 • Vectorized persistence diagram
Classification • Leave-one-out cross validation • Two datasets: 379 cells & 1268 cells • 89 categories in the first datasets; 2 categories in the second datasets Number of nearest neighbor used Accuracy 1 148 / 229 1 823 / 1130 3 195 / 229 3 1059 / 1130 5 205 / 229 5 1105 / 1130 10 213 / 229 10 1121 / 1130 Note: Some categories in the first datasets has only a few neuron cells. Thus I only consider those that contains more than 5 neuron cells. For the second dataset, since 138 neuron cells are unknown types, I only use the rest 1130 cells for testing.
Clustering • Test on the 1268 neuron cells datasets • Accuracy is around 65% Plot generated by Justin’s code Left is interneurons. Right is principal cells.
Dendrogram of two types of neuron cells from 1268 cells. Top one is interneurons. Bottom one is principal cells. The green dots and yellow dots are labels of cells from the same class.
