Deep learning, a subset of artificial intelligence, has shown great potential in several recent applications to pathology. These have mainly involved the use of classifiers to diagnose disease, while generative modelling techniques have been less frequently used. Generative adversarial networks (GANs) are a type of deep learning model that has been used to synthesize realistic images in a range of domains, both general purpose and medical. In the GAN framework, a generator network is trained to synthesize fake images, while a dueling discriminator network aims to distinguish between the fake images and a set of real training images. As GAN training progresses, the generator network ideally learns the important features of a dataset, allowing it to create images that the discriminator cannot distinguish from the real ones. We report on our use of GANs to synthesize high resolution, realistic histopathology images of gliomas. The well- known Progressive GAN framework was trained on a set of image patches extracted from digital slides in the Cancer Genome Atlas repository, and was able to generate fake images that were visually indistinguishable from the real training images. Generative modelling in pathology has numerous potential applications, including dataset augmentation for training deep learning classifiers, image processing, and expanding educational material. LEARNING OBJECTIVES This presentation will enable the learner to: 1. Explain basic principles of generative modelling in deep learning. 2. Discuss applications of deep learning to neuropathology image synthesis.

The low-frequency asynchronous switch design (LF-ASD) was introduced as a direct brain computer interface (BCI) technology for asynchronous control applications. The LF-ASD has the advantage that it is operational at any time and not only at specific defined periods. It is activated only when a user intends control, otherwise it maintains an inactive state output. The intended control results from a certain movement attempt such as finger flexion. This paper presents the evaluation of a modified LF-ASD design with data collected from individuals with high-level spinal cord injuries and able-bodied subjects. The modifications are related to incorporating into the system more knowledge about the movement attempt. Specifically, the past values of the features extracted from the EEG signal related to the movement attempts are used. The error characteristics of this new asynchronous brain switch design are significantly better than the previous LF-ASD design, with true positive rate increases of approximately 8.5% for false positive rates in the range of 1-2%

Ensemble averaging of the electroencephalogram is known to be a good tool for characterizing various event related potentials. An important part of ensemble averaging is to know the time reference that the signals should be averaged. In able-bodied individuals the muscle activity or switch activation is used to time-lock the averages. In people with spinal cord injuries who lack the ability to produce muscle activity, the expected time of the attempted movement based on an external cue can be used. This time is not accurate and can result in poor ensemble averages. A method that automatically detects the onset of the movement related potentials and use this knowledge to time-lock the averages is introduced. This method is based on the estimation of the probability density distribution of the feature vectors related to spontaneous EEG. To estimate the probability density function Parzen's method is used which is known to be as the most accurate method when large population of data is available. Preliminary experiments demonstrate the feasibility of the proposed method and show that the proposed method could generate ensemble averages closer to the averages with muscle activity knowledge than the method based on an external cue.

The low-frequency asynchronous switch design (LF-ASD) was introduced as a direct brain-computer interface (BCI) technology for asynchronous control applications. The LF-ASD operates as an asynchronous brain switch (ABS) which is activated only when a user intends control and maintains an inactive state output when the user is not meaning to control the device (i.e., they may be idle, thinking about a problem, or performing some other action). Results from LF-ASD evaluations have shown promise, although the reported error rates are too high for most practical applications. This paper presents the evaluation of four new LF-ASD designs with data collected from individuals with high-level spinal cord injuries and able-bodied subjects. These new designs incorporated electroencephalographic energy normalization and feature space dimensionality reduction. The error characteristics of the new ABS designs were significantly better than the LF-ASD design with true positive rate increases of approximately 33% for false positive rates in the range of 1%-2%. The results demonstrate that the dimensionality of the LF-ASD feature space can be reduced without performance degradation. The results also confirm previous findings that spinal cord-injured subjects can operate ABS designs to the same ability as able-bodied subjects.