A study published in Nature (2008) demonstrates that it is possible to decode visual content from brain activity using a method based on quantitative receptive field models. The researchers used functional magnetic resonance imaging (fMRI) to record brain activity while subjects viewed natural images. They developed a model that characterizes how visual stimuli relate to fMRI activity in early visual areas, including spatial, orientation, and spatial frequency tuning. This model was estimated directly from responses to natural images and allowed them to identify specific images from a large set of completely novel natural images. The study shows that the receptive field models accurately characterize the selectivity of individual voxels to natural images, enabling high identification performance. For example, 92% of images were correctly identified for one subject, while chance performance was only 0.8%. The results suggest that it may soon be possible to reconstruct a person's visual experience from brain activity measurements alone. The study also addresses the challenge of identifying natural images, which are more complex than simple stimuli like gratings. The researchers tested the performance of their model under various conditions, including different set sizes and time delays between image viewing and data collection. They found that performance remained stable over time, even when images were viewed two months or more after the initial experiment. The study highlights the importance of spatial, orientation, and spatial frequency tuning in achieving accurate identification. A model that only included spatial tuning performed significantly worse than the Gabor wavelet pyramid model, which also included orientation and spatial frequency tuning. This indicates that these factors contribute to identification performance. The study also discusses the limitations of fMRI, including its modest spatial resolution and indirect coupling to neural activity. Despite these limitations, the researchers showed that fMRI signals can be used to achieve remarkable levels of identification performance, indicating that they contain a considerable amount of stimulus-related information. The study concludes that the developed method brings us close to achieving a general visual decoder. The final step would require devising a way to reconstruct the image seen by the observer, instead of selecting the image from a known set. The study suggests that model-based approaches, which use receptive field models to infer the most likely image given a measured activity pattern, may be successful in achieving this goal.A study published in Nature (2008) demonstrates that it is possible to decode visual content from brain activity using a method based on quantitative receptive field models. The researchers used functional magnetic resonance imaging (fMRI) to record brain activity while subjects viewed natural images. They developed a model that characterizes how visual stimuli relate to fMRI activity in early visual areas, including spatial, orientation, and spatial frequency tuning. This model was estimated directly from responses to natural images and allowed them to identify specific images from a large set of completely novel natural images. The study shows that the receptive field models accurately characterize the selectivity of individual voxels to natural images, enabling high identification performance. For example, 92% of images were correctly identified for one subject, while chance performance was only 0.8%. The results suggest that it may soon be possible to reconstruct a person's visual experience from brain activity measurements alone. The study also addresses the challenge of identifying natural images, which are more complex than simple stimuli like gratings. The researchers tested the performance of their model under various conditions, including different set sizes and time delays between image viewing and data collection. They found that performance remained stable over time, even when images were viewed two months or more after the initial experiment. The study highlights the importance of spatial, orientation, and spatial frequency tuning in achieving accurate identification. A model that only included spatial tuning performed significantly worse than the Gabor wavelet pyramid model, which also included orientation and spatial frequency tuning. This indicates that these factors contribute to identification performance. The study also discusses the limitations of fMRI, including its modest spatial resolution and indirect coupling to neural activity. Despite these limitations, the researchers showed that fMRI signals can be used to achieve remarkable levels of identification performance, indicating that they contain a considerable amount of stimulus-related information. The study concludes that the developed method brings us close to achieving a general visual decoder. The final step would require devising a way to reconstruct the image seen by the observer, instead of selecting the image from a known set. The study suggests that model-based approaches, which use receptive field models to infer the most likely image given a measured activity pattern, may be successful in achieving this goal.