An AI implementation for forecasting fMRI time series

Introduction: Artificial Intelligence (AI) is having a revolutionary impact across various sectors, transforming industries, enhancing capabilities, driving innovation and efficiency, and creating new opportunities. In the health sector, AI algorithms have been used to analyse medical images and data more speedily and accurately than humans, aiding in the early detection of diseases such as cancer. It has also been used for the development of personalised treatment plans based on an individual’s genetic makeup and lifestyle. Furthermore, surgical robots powered by AI have been used to perform complex procedures with precision. Neuroimaging and AI are converging to offer transformative advancements in understanding the human brain and diagnosing neurological conditions.

Methods: We implement a deep learning neural network model that forecasts the time series of a functional magnetic resonance imaging (fMRI) dataset. However,, applying deep learning models to fMRI is not trivial due to the high-dimensional nature of fMRI images. We execute a long short-term memory (LSTM) recurrent neural network (RNN) model that forecasts the time series of fMRI datasets.

Results: The training of the LSTM-RNN model, which includes the root-mean-squared error (RMSE) and the loss, is shown. Also, a histogram depicting the RMSE and the errors is shown. The mean RMSE over all test observations was calculated, and the predictions were compared with the target values.

Conclusion: We have successfully implemented an LSTM-based RNN model for the forecasting of fMRI time series. The model will enable the prediction of future brain states and facilitate advancements in neuroscience research and clinical applications.