Table 1.
No | Year | Reference | Sample | Microscopy | Spines form representation (Figure 3) | Preprocessing and feature extraction approach | Number of features | Software (+ provided,− not provided) | Morphological features analysis approach | Software availability |
---|---|---|---|---|---|---|---|---|---|---|
1 | 2016 | Ghani et al. (2017) | 7–10 day old mouse brain slices neurons (region not specified) | Two-photon | 2D projection from image series | Histogram of oriented gradients (HOG), disjunctive normal shape models (DNSM), morphological features, intensity profile based features or their combination | Varying, from 12 for morphological to 346 for DNSM features | Custom (−) | X-means clustering, number of clusters selected automatically using the Bayesian information criterion (BIC; =4) | NA |
2 | 2016 | Bokota et al. (2016) | 19–21 days in vitro hippocampal neurons | Confocal | 2D projection from image series | The most often used morphological features according to literature data | 11 (reduced to 6) | Custom (−) | C-means clustering (=10), average-linkage hierarchical clustering (=10), data dimensionality reduction by 2D principal component analysis (PCA) | UR |
3 | 2018 | Luengo-Sanchez et al. (2018) | layer III pyramidal neurons of the human cingulate cortex | Confocal | 3D triangular surface mesh Multiresolution Reeb graph | Surface of a spine is modeled by 7 segments, which are presented as linked to each other ellipses. From 54 parameters used to describe dendritic spines, 36 reflect ellipses geometry and position and 18 describe more complex features, such as spine growth direction | 54 | Imaris for segmentation ($), custom software for feature extraction (+) | Clustering by probabilistic model with Gaussian finite mixtures, number of clusters selected automatically using the Bayesian information criterion (BIC; =6) | FA |
4 | 2019 | Kashiwagi et al. (2019) | 18–22 days in vitro hippocampal neurons | SIM | 3D triangular surface mesh | Segmentation of spines by multilevel thresholding based on Otsu’s method following geodesic active contour, combination of morphological features and high geometric features | 10 (reduced to 5) | Custom (+) | Division into mushroom and non-mushroom spines using SVM classifier, mapping the trajectories of individual spines shape transitions in the feature space, data dimensionality reduction by 3D principal component analysis (PCA) | FA |
5 | 2019 | Choi et al. (2019) | 18–22 days in vitro hippocampal neurons | SIM | 3D triangular surface mesh | Processing as in No4 with addition of 5 more features reflecting spines head and neck size | 10 | DXplorer (−) | K-means clustering, coordinate plot, radar plot and 2D scatter plot with t-Distributed Stochastic Neighbor Embedding | NA |
6 | 2019 | Driscoll et al. (2019) | CLARITY-cleared mouse brain neurons (region not specified) | LSM | Machine-learning based supervised spines detection | n/d | u-shape3D software (+) | Unsupervised hierarchical clustering (=9), data dimensionality deduction by 2D principal component analysis (PCA) | NA |
Software availability—NA, not available; UR, available upon request; FA, freely available.