正则化多任务学习在精神分裂症核磁共振成像图像分类中的应用

为实现对精神疾病的计算机辅助诊断与预后，利用机器学习与图像处理技术分析多地区精神疾病的核磁共振成像数据，已成为该领域的必然趋势。本文首先提出切片提取的核磁共振成像图像预处理方法，然后提取图像的纹理特征，最后提出一种lp范数正则化的多任务学习支持向量机精神分裂症分类方法，同时学习3个数据中心精神分裂症图像的共享特征和各自独有的特征，用于分类精神分裂症患者和正常人。实验结果表明，该方法取得了优秀的诊断精度，可为精神分裂症患者的临床诊断与治疗提供生物学依据。

Application of regularized multi-task learning in schizophrenia MRI data classification

Abstract: Machine learning techniques and magnetic resonance imaging (MRI) techniques have been used in the analysis of MRI data of patients with mental diseases in various regions to achieve computer-aided diagnosis and prognosis of mental diseases such as schizophrenia, etc. Herein slice extraction is firstly used for MRI image preprocessing. Then texture features of gray-level co-occurrence matrices are extracted from the above processed images. Finally, a lp-norm regularized multi-task learning method based on support vector machine for MRI data classification is proposed to simultaneously learn the site-specific and site-shared features of schizophrenia images from 3 data centers, which can be used to discriminate schizophrenia patients from normal controls. Experiments show that the proposed method achieves a high diagnosis accuracy, providing a biological basis for the clinical diagnosis and treatment of schizophrenia.