Artificial intelligence-based discrimination of hydrocephaly in brain CT scan

Hydrocephalus is characterized by excessive cerebrospinal fluid accumulation that enlarges the ventricles and compresses brain tissue, potentially leading to neurological dysfunction. Imaging-based assessment of the ventricular system is vital for clinical decision-making. This study developed an automated deep learning algorithm for hydrocephalus classification from brain CT slices. The dataset comprised CT scans from 55 hydrocephalic patients and 31 healthy controls collected at Loghman-Hakim Hospital, Tehran, Iran. Two strategies were compared: classical machine learning and deep learning (DL). In the first approach, LBP, distance LBP (d-LBP), and angle LBP (α-LBP) features were extracted and classified using kNN, random forest, and bagging. In the second, CT images were directly fed into a custom CNN for end-to-end classification without manual segmentation or preprocessing. The CNN achieved 99.53% accuracy, 0.9969 AUC, 99.66% precision, 99.21% recall, and 99.43% F1-score, outperforming all classical models. This AI-powered tool demonstrates high accuracy and real-time capability, showing promise for neurosurgical decision-making in emergency or resource-limited settings where CT is the primary imaging modality. Despite the limited dataset size, the results indicate strong potential, warranting further validation on larger cohorts.