The diagnosis of pediatric hydrocephalus traditionally relies on radiological examinations, with Evans' Index as a key severity indicator. To address the time-consuming and subjective nature of manual measurements, we propose an automated diagnostic approach using deep learning. Our enhanced U-Net model performs lateral ventricle segmentation and Evans' Index measurement from brain imaging, incorporating dense connectivity and attention mechanisms. Results demonstrate high segmentation accuracy and strong correlation with manual measurements, validated through clinical cross-validation. This automated solution offers efficient and objective diagnostic support, showing promise for broader medical imaging applications.