Spinocerebellar ataxia type 12 (SCA12) is a neurogenetic disease, marked with prominent action tremors in the upper limbs. Neuroimaging techniques like magnetic resonance imaging (MRI) are used by doctors to find the affected areas of SCA12 disease. In literature, most of the research work have used 2D-feature extraction methods, which do not consider pixel information from adjacent slices of the MRI volume, which may be relevant to distinguish healthy from the patient suffering from a particular disease. To overcome the problem of 2D-feature extraction method, we investigate six well-recognized 3D-feature extraction techniques based on varied principles individually and in combination from whole brain gray matter volume. To obtain the optimal set of relevant features, we investigated eight well-known feature selection methods. The support vector machine (SVM) was used as the classifier. Experimental results demonstrate the superior performance of 3D-feature extraction methods in comparison to 2D-feature extraction methods. The features obtained from the combination of feature extraction methods (COFEMS) combined with SVM with Recursive Feature Elimination method achieved maximum classification accuracy of 90% and F1-score of 89.25%. The subset of features so obtained is found statistically relevant and non-redundant. Ranking analysis on both feature extraction and feature selection methods is also carried out.

脊髓小脑性共济失调 12 型 (SCA12) 是一种神经遗传性疾病，以上肢明显的动作性震颤为特征。医生使用磁共振成像 (MRI) 等神经影像技术来查找 SCA12 疾病的受影响区域。在文献中，大多数研究工作都使用 2D 特征提取方法，该方法不考虑 MRI 体积相邻切片的像素信息，这可能与区分健康人和患有特定疾病的患者有关。为了克服 2D 特征提取方法的问题，我们研究了六种公认的 3D 特征提取技术，它们基于不同的原理，单独或组合地从全脑灰质体积中提取。为了获得最佳的相关特征集，我们研究了八种著名的特征选择方法。使用支持向量机（SVM）作为分类器。实验结果表明，与 2D 特征提取方法相比，3D 特征提取方法具有更优越的性能。特征提取方法（COFEMS）与 SVM 和递归特征消除方法相结合所获得的特征实现了 90% 的最大分类精度和 89.25% 的 F1 分数。如此获得的特征子集被发现具有统计相关性且非冗余。还对特征提取和特征选择方法进行了排名分析。