image: First, multimodal imaging, demographic, and clinical data obtained from 90-d HDTBR experiment. Further statistical analyses were conducted for alteration of CBF. Next, the CBF prediction models were developed based on 8 ML methods and explanation analysis was utilized to identify the important input features. Finally, the prediction model was implemented as a web application.
Credit: Zhenchang Wang, School of Biological Science and Medical Engineering, Beihang University.
“Maintaining adequate CBF is crucial for astronauts’ cognitive function during long-duration microgravity, but real-time monitoring in space is constrained by MRI’s complexity and payload limits,” explained study corresponding author Lijun Yuan from Air Force Medical University. The core innovations include (a) using −6° head-down tilt bed rest (HDTBR) to simulate microgravity, (b) integrating carotid ultrasound and brain MRI data to establish ML-based CBF prediction models, and (c) developing an interpretable web application for in-orbit deployment. “This solution enables non-invasive, lightweight CBF assessment, supporting timely medical interventions for astronauts.”
The model leverages key technical advancements: 36 healthy male participants underwent 90-d HDTBR, with multimodal data collected including internal carotid artery (ICA) Doppler ultrasound, brain 3D-pCASL MRI, and clinical parameters (BMI, HR, blood pressure). Eight ML algorithms were tested, with CatBoost emerging as the optimal model due to its ability to capture complex nonlinear relationships between ICA hemodynamic features and CBF changes. “SHAP analysis was used to interpret the model, identifying BMI, ICA pulsatility index (PI), and blood flow volume (FV) as key predictive features,” said lead author Linkun Cai.
The study authors validated the model through comprehensive experiments: after 90-d HDTBR, significant regional CBF decreases were observed in the right Heschl’s gyrus (AUC=0.88, accuracy=0.84), right middle cingulate gyrus (AUC=0.92, accuracy=0.83), and right superior frontal gyrus (AUC=0.82, accuracy=0.72). The CatBoost model outperformed other algorithms (logistic regression, SVM, random forest, etc.) across all evaluation metrics, and the developed web application allows astronauts to input clinical data and upload ultrasound files for real-time CBF prediction and risk visualization.
“While the model shows promising results, it still faces challenges: limited inclusion of vertebral artery blood flow data, single post-labeling delay in MRI acquisition, and a male-only participant cohort,” said Cai. Future work will focus on integrating vertebral artery features, adopting multi-PLD ASL technology, and validating the model in diverse populations including female astronauts. Overall, this ML-driven prediction framework provides a practical solution for CBF monitoring in simulated microgravity, laying the foundation for safeguarding brain health during long-duration human spaceflight.
Authors of the paper include Linkun Cai, Yawen Liu, Kai Li, Changyang Xing, Zi Xu, Lianbi Zhao, Ke Lv, Zhili Li, Hao Wang, Linjie Wang, Dehong Luo, Lijun Yuan, Lina Qu, Yinghui Li, Zhenchang Wang, and Pengling Ren.
The study was funded by the Space Medical Experiment Project of China Manned Space Program (no. HYZHXMH01005), China Manned Space Advanced Research Project (ES-2-NO.0041), Beijing Hospitals Authority Innovation Studio of Young Staff Funding Support (no. 202302), and Beijing Scholar 2015 (Z.W.).
The paper, “Multimodal Imaging-Based Cerebral Blood Flow Prediction Model Development in Simulated Microgravity” was published in the journal Cyborg and Bionic Systems Nov. 24, 2025, at DOI: 10.34133/cbsystems.0448.