Table 2.
Processing features of fully automated toolboxes
| ALI | Voxel‐based GNB classification (lesion_gnb) | LINDA | |
|---|---|---|---|
| Compatible operating systems | Windows, Linux, Mac | Windows, Linux, Mac | Windows 10+, Linux, Mac |
| Platform dependencies | MATLAB, SPM5+ | MATLAB 2014b+ (requires statistics and machine learning toolbox), SPM12+ | R v.3.0+, ANTsR package |
| Year developed | 2007 | 2015 | 2016 |
| Open source | No | Yes | Yes |
| Learning type | Unsupervised | Supervised | Supervised |
| Training dataset | Requires user to provide segmented healthy training dataset | Provided (trained on 30 LHS subjects) | Provided (trained on 60 LHS subjects) |
| Amenable to left or right hemisphere lesions | Yes | Yes, provided that the user indicates which hemisphere first | No, right hemisphere lesions must be flipped |
| Template brain space | ICBM152 | ICBM152 | Colin 27 template |
| User‐defined parameters | Sensitivity (tuning factor), fuzziness index in fuzzy means clustering algorithm, threshold probability and size for the extra class prior | Optional smoothing, smoothing kernel, minimum cluster size, implicit masking while smoothing | None |
| Optional postprocessing steps | None | Resegmentation with a tissue prior | None |
Abbreviations: ALI, automated lesion identification; GNB, Gaussian naïve Bayes; lesionGnb, Gaussian naïve Bayes lesion detection; LHS, left hemisphere stroke; LINDA, lesion identification with neighborhood data analysis.