In order to analyze the application of artificial intelligence neural network imaging technology in children with refractory epilepsy prognosis infection, the magnetic resonance imaging (MRI), positron emission computed tomography (PET), and MRI/PET fusion technology based on artificial intelligence neural network were adopted to detect the refractory epilepsy of 78 children in this study. The diagnostic sensitivity, specificity, and accuracy of the three imaging methods were calculated. All children were followed up for one year. Variance analysis was used to analyze the risk factors of postoperative intracranial infection in children. The results showed that MRI/PET diagnostic accuracy (80.8%), sensitivity (86.9%), and specificity (81.1%) were higher than MRI diagnosis (64.1%, 77.7%, and 71%) and PET diagnosis (70.5%, 86.4%, and 71%); the intracranial infection rate of the children without using antibiotics (14.1%) was higher than that of the children using antibiotics (0%); the intracranial infection rate of the children with operation time less than 2 h (0%) was lower than that of the children with operation time between 2-4 h (4%) and that of the children with operation time more than 4 h (9.4%); the intracranial infection rate in children with less than 3 brain puncture times (0%) was lower than that in children with more than 3 puncture times (42.3%); the intracranial infection rate in children with intracranial catheterization time less than 24 h (0%) was lower than that in children with intracranial catheterization time between 34 h and 48 h (6.9%) and that in children with intracranial catheterization time longer than 48 h (12.5%). In summary, MRI/PET fusion detection based on artificial intelligence neural network had higher accuracy, specificity, and sensitivity. Long operative time, long intracranial catheterization time, multiple lumbar subarachnoid puncture times, and non-prophylactic use of antibiotics before and during operation were the risk factors for intracranial infection.