〔Abstract〕 To explore the role of neuroimaging features monitored by functional near-infrared spectroscopy (fNIRS) in postoperative delirium (POD) in elderly patients undergoing joint replacement during perioperative period, and to provide a basis for early clinical prediction. Methods A total of 105 elderly patients who underwent joint replacement under general anesthesia were included. The Mini-Mental State Examination (MMSE) scale was used to evaluate the patient's cognition one day before the operation. Before the start of the surgery, fNIRS was used to monitor the changes of cerebral blood oxygen saturation when the patient performed the task state. The 3-minute delirium diagnostic scale (3D-CAM scale) was used to evaluate the occurrence of POD at 24, 48 and 72 h after operation. Brain network analysis was performed and Logistic regression analysis was used to explore the relationship between fNIRS monitoring data and POD in elderly patients undergoing joint replacement surgery during preoperative task state. The receiver operating characteristic curve (ROC curve) was constructed to evaluate the diagnostic efficacy, and the Hosmer-Lemeshow goodness-of-fit test was used to test the goodness of fit of the model. Results Among 105 patients, 100 cases were effectively analyzed, of which 20 cases (20%) had POD. Brain network analysis showed that the r value of functional connectivity correlation coefficient in POD group (0.069±0.118) was lower than that in non-POD group (0.073±0.084). The low channel connectivity of right primary somatosensory cortex-right primary motor cortex (RS1-RM1) and left anterior pole-right Broca's triangle (LFP-RBA44) was an important factor affecting the occurrence of POD (P < 0.05). Based on this result, the area under the ROC curve was 0.797 and 0.784, respectively. The results of Hosmer-Lemeshow goodness-of-fit test showed that the model fitted well (all P>0.5). Conclusion The neuroimaging features extracted from the cerebral oxygen saturation data monitored by fNIRS are significantly correlated with the risk of POD in elderly patients undergoing joint replacement during perioperative period. Among them, the low connectivity of preoperative RS1-RM1 and LFP-RBA44 brain network channels is an important influencing factor of POD occurrence. Predicting the occurrence of POD based on fNIRS is conducive to the early intervention and risk reduction of perioperative complications, improving medical quality and promoting precision medical practice.