Automated Detection of Elevated Intracranial Pressure via Fundus Image Analysis: A Deep Learning Approach

Elevated intracranial pressure (ICP) is a critical neurological condition that requires timely detection to prevent vision loss and severe brain complications. Papilledema, characterised by swelling of the optic disc visible in retinal fundus images, is a key indicator of increased ICP. However, distinguishing papilledema from pseudopapilledema, which presents similar visual features but occurs without ICP elevation, remains a significant clinical challenge. This study presents a deep learning-based automated classification system for detecting ICP-related abnormalities using retinal fundus images. The proposed approach classifies images into three categories: Papilledema (ICP raised), Pseudopapilledema (ICP normal), and Normal. The EfficientNet-B2 architecture is trained on 1369 labelled fundus images using image preprocessing, data augmentation, and transfer learning to improve generalisation and classification performance. Experimental results demonstrate that the proposed model effectively differentiates these conditions, achieving a test accuracy of 98.55% and a validation accuracy of 99.51%. Explainable AI techniques including Grad-CAM and LIME are integrated to provide visual interpretability of model decisions. The system provides a non-invasive, reliable, and efficient diagnostic support tool for early detection of elevated ICP, potentially assisting clinicians in timely decision-making.