End-to-end deep learning approach to mouse behavior classification from cortex-wide calcium imaging

Researchers at Kobe University have made groundbreaking advancements in the field of neuroscience, particularly in the realm of brain-machine interfaces. Their work centres around the development of an artificial intelligence (AI) algorithm capable of predicting mouse movement with an impressive 95% accuracy rate. What makes this achievement even more remarkable is the utilization of whole-cortex functional imaging data, a method that holds immense potential for revolutionizing brain-machine interface technology.

Traditionally, neural decoding efforts have heavily relied on analyzing the electrical activity of brain cells through implanted electrodes. However, the Kobe University team took a different approach by leveraging functional imaging technologies, such as calcium imaging, to monitor brain activity. Unlike traditional methods, calcium imaging provides a faster and more detailed view of brain activity, making it an attractive option for neural decoding endeavours.

Led by medical student AJIOKA Takehiro and neuroscientist TAKUMI Toru, the team developed an innovative “end-to-end” deep learning method for predicting behavioural states based on whole-cortex functional imaging data. This method bypasses the need for data preprocessing, streamlining the analysis process and maximizing efficiency. By combining spatial and temporal pattern recognition algorithms, the AI model can distinguish between resting and running states in mice with remarkable accuracy.

One of the key advantages of this approach is its potential for personalized, near-real-time applications in brain-machine interfaces. The AI model can generate predictions from just 0.17 seconds of imaging data, demonstrating its rapid processing capabilities. Moreover, the model’s effectiveness across different mice showcases its ability to filter out individual characteristics, further enhancing its applicability.

In addition to achieving high prediction accuracy, the researchers also devised a technique to discern which parts of the data are pivotal for prediction. By identifying critical cortical regions responsible for behavioural classification, the team has enhanced the interpretability of deep learning in neuroscience. This not only sheds light on the AI’s decision-making process but also contributes to our understanding of brain function and behaviour.

The research process involved the integration of expertise from various fields, including medical imaging, neuroscience, and artificial intelligence. This interdisciplinary approach allowed the team to tackle complex challenges in neural decoding and brain-machine interface development. Through rigorous experimentation and analysis, they were able to validate the effectiveness of their AI algorithm in predicting mouse movement based on whole-cortex functional imaging data.

The implications of this research are vast and extend beyond the realm of neuroscience. The development of non-invasive brain-machine interfaces capable of decoding behaviour in near real-time holds promise for a wide range of applications. From neuroprosthetics aimed at restoring lost functions to enhancing human-computer interaction, the potential applications of this technology are far-reaching.

In summary, the work conducted by the Kobe University team represents a significant step forward in the field of brain-machine interfaces. By harnessing the power of AI and functional imaging data, they have laid the groundwork for the development of advanced neural decoding techniques with broad implications for both medical and technological advancements.

Original publication T. Ajioka et al.: End-to-end deep learning approach to mouse behavior classification from cortex-wide calcium imaging. PLOS Computational Biology (2024). DOI: 10.1371/journal.pcbi.1011074

Until then follow

plain logo

Contact Us:

Follow Us

Get the latest news, analysis and stories on Neurotechnology (Neurotech).

©️ 2024 Neureads | All Rights Reserved