The precise segmentation of Brainstem gliomas (BSGs) tissue is crucial for surgical planning and radiomics. We present a cascaded deep convolutional neural network (CNN) for segmentation and genotype prediction of brainstem gliomas simultaneously. Segmentation task contains two feature-fusion modules: Gaussian-pyramid multiscale input and region-enhancement. Prediction model combines CNN features and support-vector-machine classifier to automatically predict genotypes. Experiments demonstrate that our method achieves good tumor segmentation results and competitive genotype prediction results.

We present a novel Granger causality estimator based on recurrent neural networks (RNNs) to deal with the multivariate brain connectivity detection problem. The varying-length propagation delay between brain signals poses a challenge for classical approaches using fixed time lags. Our method takes time series signals with arbitrary length of transmission time lags and learns the information flow from the data with RNNs. The proposed method works well on varying-length time lags and even on very long transmission delays in electroencephalography (iEEG) signals, and is promising to serve as a robust brain connectivity analysis tool in clinical applications.

Automatic and accurate classification and recognition of fMRI-derived functional brain networks are very important for subsequent steps of functional brain analysis. We design a deep 3D CNN framework for automatic and accurate classification and recognition of numbers of functional brain networks. Our work provides a new deep learning approach for modeling functional connectomes based on fMRI data.

We propose a novel calibration strategy to facilitate the decoding of covert somatosensory attentional changes by exploring the oscillatory dynamics induced by actual tactile sensation. Offline analysis showed that the proposed calibration method led to higher accuracies than the traditional calibration method based only on somatosensory attentional orientation (SAO) data.

We investigated a new approach for improving iBCI performance which takes advantage of the closed loop nature of BCIs: users perceive the errors, and this shows up in their neural activity. Using offline analyses, we show the feasibility of automatically detecting and undoing errors from neural activity alone.

We present an approach for fusing partial color-channel signals from an array of cameras that enables physiology measurements to be made from moving subjects, even if they leave the frame of one or more cameras. PRV estimates were significantly improved using our proposed approach compared to an alternative not designed to handle missing values and multiple camera signals, the error was reduced by over 50%.