Morphological Neuron Classification Using Machine Learning

Abstract

Classification and quantitative characterization of neuronal morphologies from histological neuronal reconstruction is challenging since it is still unclear how to delineate a neuronal cell class and which are the best features to define them by. The morphological neuron characterization represents a primary source to address anatomical comparisons, morphometric analysis of cells, or brain modeling. The objectives of this paper are (i) to develop and integrate a pipeline that goes from morphological feature extraction to classification and (ii) to assess and compare the accuracy of machine learning algorithms to classify neuron morphologies. The algorithms were trained on 430 digitally reconstructed neurons subjectively classified into layers and/or m-types using young and/or adult development state population of the somatosensory cortex in rats. For supervised algorithms, linear discriminant analysis provided better classification results in comparison with others. For unsupervised algorithms, the affinity propagation and the Ward algorithms provided slightly better results.

Keywords: classification; machine learning; morphologies; neurons; supervised learning; unsupervised learning.

FIGURE 1

Data Sample of the 430 neurons with Layer 2/3, 4, 5 and 6 as L23, L4, L5 and L6 and m-type Basket, Bipolar, Bitufted, Chandelier, Double Bouquet, Martinotti, Pyramidal, and Stellate cells as BC, BPC, BTC, ChC, DBC, MC, PC, and SC.

FIGURE 2

Glimpse of two Layer 4 Pyramidal Cell from NeuroMorpho.org provided by Wang et al. (Wang et al., 2002) and visualized through the animation tool provided by neurophormo.org with neuron C140600C-P3 (A) 366°, (B) 184°, and (C) standard image; and with neuron C200897C-P2 (D) 356°, (E) 194°, and (F) standard image.

FIGURE 3

The mean accuracy scores with their respective standard deviation of the supervised algorithms. The mean accuracy scores have been computed 10 times using a randomly chosen 30% data subset to classify morphologies according to layers and m-types, m-types, and layers only.

FIGURE 4

Tests varying the percentage of train to test the ratio of samples from 1 to 80% showing a relative stability of the linear discriminant analysis (LDA) algorithm. The figure shows the mean accuracy scores with their respective standard deviation for each of the category tested.

FIGURE 5

Miss-classification matrices for the LDA algorithm providing the best accuracy for each of the categories with the true value and predicted value and the associated percentage of accuracy for the following categories: (A) combined layers and m-types in young and adult population, (B) combined layers and m-types in young population, (C) m-types in young and adult population, (D) m-types in young population, and (E) layers and pyramidal cells in young population.

FIGURE 6

V-measures comparison of the unsupervised clustering algorithms classifying morphologies according to layer and m-type, m-type and layer in young an/or adult population. The figure shows also homogeneity scores, completeness scores, adjusted rand index, adjusted mutual information, and silhouette coefficient.