April 22, 2022
Okayama University
Keio University
Rikkyo University
A joint research group consisting of Associate Professor Teppei Matsui of the Graduate School of Natural Science and Technology (Faculty of Science, Biology), Okayama University; Associate Professor Masato Taki of the Graduate School of Artificial Intelligence and Science, Rikkyo University; Project Assistant Professor Trung Quang Pham of the National Institute for Physiological Sciences; principal investigator Junichi Chikazoe of Araya Inc.; and Associate Professor Koji Jimura of the Faculty of Science and Technology, Keio University, has developed a new method to intuitively explain the behavior of deep neural circuits that decode brain activity.
These research findings were published as a Research Article in the Swiss neuroscience journal *Frontiers in Neuroinformatics* on March 16.
Brain activity decoding, which estimates what a person was doing from brain activity data measured by MRI or EEG, is a technology widely studied for its potential application in brain-machine interfaces (BMIs). Recently, research using deep learning for brain activity decoding has become active. However, the data processing by deep neural circuits is extremely complex, making it difficult to intuitively explain "why the decoder gives a particular answer for a given data." In this study, the researchers proposed a new approach to this problem: a method that combines a deep generative model, another deep learning technique, with a method called counterfactual explanation.
These research findings are expected to serve as a foundational technology that allows doctors and patients to use AI while understanding its behavior when applying deep learning to brain imaging diagnostics for dementia and neuropsychiatric disorders.
Please see below for the full press release.