Pattern Recognition Strategies to Classify Traced Neurons
| Autor: | Juan V. Lorenzo-Ginori, José Daniel López-Cabrera, Leonardo Agustín Hernández-Pérez |
|---|---|
| Rok vydání: | 2021 |
| Předmět: | |
| Zdroj: | Progress in Artificial Intelligence and Pattern Recognition ISBN: 9783030896904 IWAIPR |
| DOI: | 10.1007/978-3-030-89691-1_15 |
| Popis: | This paper addresses two strategies for pattern recognition in high-dimension data sets, obtained from databases of digitally traced neurons. The first strategy has as distinctive characteristic that the features are obtained both from the whole neuron and from the axon and the dendrites as well. On the other hand, the second strategy is based in feature extraction from data sequences obtained from the decomposition of the traced neuron. Both strategies generate a wide variety of data which result in highly dimensional data sets. Two supervised pattern recognition alternatives were implemented for feature selection. When using the first strategy, the percentage of correctly classified cases raised up to 94.55% once the fusion of data extraction alternatives is performed and the feature selection methods are applied. These results are maintained for three different sets of traced neurons pertaining different regions of the cerebral cortex. By means of the second strategy, a group or recursive feature elimination alternatives was evaluated. Using this strategy, the mean percentage of correctly classified cases achieved was 78.5% and 76.79% which are comparable to those obtained using multiple iterations and the computational time was reduced three to four times. The results obtained encourage the use of the proposed strategies to reduce the dimensionality of the data sets obtained from the traced neurons. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|