Echo planar imaging (EPI), a fast magnetic resonance imaging technique, is a powerful tool in functional neuroimaging studies. However, susceptibility artifacts, which cause misinterpretations of brain functions, are unavoidable distortions in EPI. This paper proposes an end-to-end deep learning framework, named TS-Net, for susceptibility artifact correction (SAC) in a pair of 3D EPI images with reversed phase-encoding directions. The proposed TS-Net comprises a deep convolutional network to predict a displacement field in three dimensions to overcome the limitation of existing methods, which only estimate the displacement field along the dominant-distortion direction. In the training phase, anatomical T1-weighted images are leveraged to regularize the correction, but they are not required during the inference phase to make TS-Net more flexible for general use. The experimental results show that TS-Net achieves favorable accuracy and speed trade-off when compared with the state-of-the-art SAC methods, i.e., TOPUP, TISAC, and S-Net. The fast inference speed (less than a second) of TS-Net makes real-time SAC during EPI image acquisition feasible and accelerates the medical image-processing pipelines.

Accurate segmentation of different brain tissue types is an important step in the study of neuronal activities using functional magnetic resonance imaging (fMRI). Traditionally, due to the low spatial resolution of fMRI data and the absence of an automated segmentation approach, human experts often resort to superimposing fMRI data on high resolution structural MRI images for analysis. The recent advent of fMRI with higher spatial resolutions offers a new possibility of differentiating brain tissues by their spatio-temporal characteristics, without relying on the structural MRI images. In this paper, we propose a patch-wise deep learning method for segmenting human brain tissues into five types, which are gray matter, white matter, blood vessel, non-brain and cerebrospinal fluid. The proposed method achieves a classification rate of 84.04% and a Dice similarity coefficient of 76.99%, which exceed those by several other methods.

Accurate alignment of multi-session medical imaging is essential to the analysis of disease progression. By comparing the magnetic resonance imaging (MRI) data captured before and after a course of neoadjuvant chemoradiation (nCRT) treatment, physicians are able to evaluate the tumor response for further treatment of the disease. However, rectal MRI data captured in multi-session are often misaligned and not guaranteed to have one-to-one correspondence, making it challenging for physicians to observe the treatment response of tumor. To address this issue, we propose an unsupervised learning based volume registration framework, which enables accurate alignment even under a high degree of deformation between multi-session rectal data. Moreover, it works without the assumption of one-to-one correspondence between multi-session data, and hence is a general solution to rectal MRI volume registration. The experimental results show that the proposed registration framework accurately aligns rectal cancer images and outperforms other state-of-the-art methods in medical image registration. By providing accurate registration, it can potentially increase the efficiency and reduce the workload for physicians to evaluate the rectal tumor response to nCRT.