Understanding how the brain encodes external stimuli and how these stimuli can be decoded from the measured brain activities are long-standing and challenging questions in neuroscience. In this paper, we focus on reconstructing the complex image stimuli from fMRI (functional magnetic resonance imaging) signals. Unlike previous works that reconstruct images with single objects or simple shapes, our work aims to reconstruct image stimuli that are rich in semantics, closer to everyday scenes, and can reveal more perspectives. However, data scarcity of fMRI datasets is the main obstacle to applying state-of-the-art deep learning models to this problem. We find that incorporating an additional text modality is beneficial for the reconstruction problem compared to directly translating brain signals to images. Therefore, the modalities involved in our method are: (i) voxel-level fMRI signals, (ii) observed images that trigger the brain signals, and (iii) textual description of the images.

Understanding how the brain encodes external stimuli and how these stimuli can be decoded from the measured brain activities are long-standing and challenging questions in neuroscience. In this paper, we focus on reconstructing the complex image stimuli from fMRI (functional magnetic resonance imaging) signals. Unlike previous works that reconstruct images with single objects or simple shapes, our work aims to reconstruct image stimuli that are rich in semantics, closer to everyday scenes, and can reveal more perspectives. However, data scarcity of fMRI datasets is the main obstacle to applying state-of-the-art deep learning models to this problem. We find that incorporating an additional text modality is beneficial for the reconstruction problem compared to directly translating brain signals to images. Therefore, the modalities involved in our method are: (i) voxel-level fMRI signals, (ii) observed images that trigger the brain signals, and (iii) textual description of the images.

Medical image reconstruction is often an ill-posed inverse problem. In order to address such ill-posed inverse problems, prior knowledge of the sought after object property is usually incorporated by means of regularization. For example, sparsity-promoting regularization in a suitable transform domain is widely used to reconstruct images with diagnostic quality from noisy and/or incomplete medical data. However, sparsity-promoting regularization may not be able to comprehensively describe the actual prior information of the objects being imaged. Deep generative models, such as generative adversarial networks (GANs) have shown great promise in learning the underlying distribution of images. Prior distributions for images estimated using GANs have been employed as a means of regularization with impressive results in several linear inverse problems in computer vision that are also relevant to medical imaging.

A new technique developed by neuroscientists at U of T Scarborough can, for the first time, reconstruct images of what people perceive based on their brain activity gathered by EEG. The technique developed by Dan Nemrodov, a postdoctoral fellow in Assistant Professor Adrian Nestor's lab at U of T Scarborough, is able to digitally reconstruct images seen by test subjects based on electroencephalography (EEG) data. "When we see something, our brain creates a mental percept, which is essentially a mental impression of that thing. We were able to capture this percept using EEG to get a direct illustration of what's happening in the brain during this process," says Nemrodov. For the study, test subjects hooked up to EEG equipment were shown images of faces.

In the past 12 months, interest in--and the development of -- using artificial neural networks for the generation of text, images and sound has exploded. In particular, methods for the generation of images have advanced remarkably in recent months. In November 2015, Radford et al. blew away the machine learning community with an approach of using a deep neural network to generate realistic images of bedrooms and faces using an adversarial training method in which a generator network generates random samples, and a discriminator network tries to determine which images are generated and which are real. Over time the generator becomes very good at producing realistic images that can fool the discriminator. The adversarial method was first proposed by Goodfellow et al. in 2013, but until Radford et al.'s paper, it hadn't been possible to generate coherent and realistic natural images using neural nets.