Automatic Classification Software for MRI brain scans: A Diagnostic tool

人类面临的一项重大挑战是全球人口老龄化以及阿尔茨海默病（AD）等退化性和衰弱性疾病的相关增加。早期诊断对于提高患者的生活质量和减少社会成本至关重要。磁共振成像（MRI）是一种成熟的研究大脑异常的工具。传统的MRI只能显示阿尔茨海默病的非特异性脑萎缩，但已经证明，“定量”MRI技术以及功能MRI （fMRI）可以提供痴呆症发病的特征（Bozzali et al., 2011）。静息状态功能磁共振成像是一种相对较新的方法来检测静息时的自发大脑活动（Greicius et al., 2004）。静息状态功能磁共振成像对痴呆的早期诊断可能非常有效（Zhou et al., 2010）。然而，由于数据量大，需要复杂的图像分析，以及结果解释困难，它被认为不适合临床使用。使用静息状态fMRI数据的独立成分分析（ICA）解构，已经确定了大脑中的许多功能网络。然而，原始图像甚至压缩的ICA数据的解释需要有经验的人眼。特别是，一旦确定了网络，它们通常使用单变量统计方法进行单独分析。提供快速、自动、客观的图像分类和诊断的多变量方法将在临床和研究领域产生巨大影响。我们的建议是说明的概念证明自动分类的MRI成像的大脑。我们将在静息条件下使用现有的ICA分解患者和对照组。我们的计划是使用机器学习技术，特别是使用高斯混合的贝叶斯分类器，该分类器已用于天文学研究。我们将编写一个基于静息状态fMRI的原型诊断工具，可以由CISC组进行测试。一旦验证，这种方法可以很容易地扩展到其他MRI模式的联合分析。该提案将苏塞克斯大学的天文学中心和临床成像科学中心（CISC）结合在一起。天文中心在统计分析和软件开发方面拥有广泛的专业知识，而CISC则拥有广泛的人类分类数据集和对临床问题的深刻理解，以及在静息状态功能磁共振成像和其他MRI技术方面的专业知识。

A major challenge facing humankind is an aging global population and the associated increase in degenerating and debilitating diseases, such as Alzheimer's disease (AD). Early diagnosis is essential to improve patient quality of life and minimize social costs. Magnetic Resonance Imaging (MRI) is a well-established tool for studying brain abnormalities. Conventional MRI can only reveal unspecific brain atrophy in AD, but it has been demonstrated that "quantitative" MRI techniques, together with functional MRI (fMRI) can provide signatures of the onset of dementia (Bozzali et al., 2011). Resting-state fMRI is a relatively novel approach to detect spontaneous brain activity at rest (Greicius et al., 2004). Resting-state fMRI is potentially very powerful for the early diagnosis of dementia (Zhou et al., 2010). However, it is regarded as unsuitable for clinical use owing to the volume of data, the complex image analysis required, and the difficulty in the interpretation of results.A number of functional networks in the brain have been identified using Independent Component Analysis (ICA) deconstructions of resting-state fMRI data. However, the interpretation of the raw images and even the compressed ICA data requires an experienced human eye. In particular, once the networks have been identified, they are typically analysed separately using univariate statistical approaches. A multivariate approach providing a quick, automatic, objective classification and diagnosis of images would have a huge impact in clinical and research arenas. Our proposal is to illustrate proof of concept for the automatic classification of MRI imaging of the brain. We will use existing ICA decomposition of patient and control groups in resting conditions. Our plan is to use a machine learning technique, specifically a Bayesian classifier using Gaussian mixtures, which has been used in Astronomical research. We will code a prototype diagnostic tool, based on resting state fMRI, which can be tested by the CISC group. Once validated, this approach can be easily extended to the joint analysis of other MRI modalities.This proposal brings together the Astronomy Centre and the Clinical Imaging Science Centre (CISC) at the University of Sussex. The Astronomy Centre brings extensive expertise in statistical analysis and software development while the CISC brings extensive human-classified data sets and deep understanding of the clinical problems and expertise in resting state fMRI and other MRI techniques.