Israeli Researchers' Brain-IT AI Model Reconstructs Images from Brain Activity

Researchers at Israel's Weizmann Institute of Science have developed an artificial intelligence model called Brain-IT that can reconstruct images from functional magnetic resonance imaging (fMRI) data recorded while a person views those images.

Unlike existing methods that typically require dozens of hours of brain scan data, Brain-IT needs only about one hour of training to decode a new subject's brain activity. The model is said to reproduce image content, composition and color more accurately.

The team built image encoders and decoders and used artificially generated predicted brain activity data to address the shortage of training samples. The encoder also identified 128 image-processing functional regions shared across individuals, some of which had not been previously discovered.

The study has been accepted by the International Conference on Learning Representations (ICLR). The researchers are currently exploring extending the technique to decoding sound information and studying the possibility of video decoding in the future.