A Simple Yet Effective Method for Motion Detection in Structural Magnetic Resonance Images

Motion is one of the common challenges in automated analysis of MR images. The structural MR images are prone to motion, and it is preferred to detect and discard poor quality images. Several automated image quality control methods have been reported in the literature and these methods either rely on computationally expensive pre-processing steps or large number of trainable parameters. In this work, we have developed an image gradient-based metric that serves as a discriminative feature between motion-free and motion-corrupt images. This feature is fed to a multi layer perceptron model (MLP) with two hidden layers to classify the input MR image as motion-free or motion-corrupted. The experimental results demonstrated that the proposed method exhibits better generalization on unseen images regardless of whether the originating site is used in the training dataset or not. Also, the proposed method significantly reduced the preprocessing requirements as well as the number of trainable parameters and labeled dataset.