Artificial intelligence-based determination of periventricular edema in hydrocephalic brain CT scan

Abstract

Hydrocephalus is excessive accumulation of cerebrospinal fluid within the cerebral ventricles. It has a complex pathogenesis with various causes. Periventricular edema refers to the abnormal accumulation of fluid in the brain tissue surrounding the cerebral ventricles, an indicative of elevated intracranial pressure or disruption in cerebrospinal fluid flow. Periventricular edema can serve as one of the severity indicators of hydrocephalus, and can assist physicians in predicting the outcomes of treatments and determining appropriate therapeutic interventions. In this study, our goal is to identify periventricular edema in hydrocephalus disease. In this regard, the smoothing-sharpening image filter (SSIF) algorithm is applied to enhance hydrocephalic CT images due to the low quality of CT images and the ambiguity between the boundaries of periventricular edema, ventricles, and other brain regions. Some well-known deep learning models including UNet, PSPNet, LinkNet and FPN are suggested to segment periventricular edema. From the obtained results, the FPN model, compared to the other models, achieves the best evaluation criteria with AUC, dice score, F1-score, precision, and recall values of 95 %, 93 %, 91 %, 91 %, and 92 %, respectively.