Multimodal Imaging-Based Cerebral Blood Flow Prediction Model Development in Simulated Microgravity

Background: Abnormal alterations in cerebral blood flow (CBF) have been implicated in cognitive decline and neurodegeneration. Maintaining adequate CBF in astronauts during long-duration microgravity is therefore crucial for the success of manned spaceflight. However, the quantitative assessment of CBF during space missions remains challenging. Methods: Thirty-six participants underwent a 90-d -6° head-down tilt bed rest (HDTBR) protocol, a well-established ground-based analog of microgravity. Multimodal imaging data, including internal carotid artery Doppler ultrasound and brain magnetic resonance imaging, were collected during HDTBR. Multiple machine learning (ML) algorithms were developed to investigate carotid-CBF mapping relationship and establish CBF change prediction models. Results: After 90-d HDTBR, significant regional CBF decreases were observed, primarily in the right Heschl's gyrus, right middle cingulate gyrus, and right superior frontal gyrus. The optimal ML model CatBoost showed robust predictive performance for CBF in these regions (right Heschl's gyrus: AUC = 0.88, accuracy = 0.84; right middle cingulate gyrus: AUC = 0.92, accuracy = 0.83; right superior frontal gyrus: AUC = 0.82, accuracy = 0.72). To enhance accessibility and practical utility, the prediction model was implemented as an interactive web application for in-orbit deployment. Conclusion: This study demonstrates the feasibility of constructing ML-driven CBF prediction models under microgravity based on multimodal imaging. The developed prediction models show promise as early warning tools for brain health of astronauts in spaceflight.