Multi-Modal Deep Learning for Brain Tumor Prognosis Prediction Using MRI and Clinical Data

Abstract

Medical imaging segmentation serves as a critical preprocessing step that enables reliable diagnosis, quantitative analysis, and treatment planning. Its clinical application spans tissue differentiation, tumor detection, lesion sizing, and cellular analysis, among others. FCM (Fuzzy C-Means) clustering is a commonly adopted segmentation approach, yet its accuracy degrades in noisy environment. To address this shortcoming, Weighted Fuzzy C-Means (WFCM) incorporates intensity patterns from surrounding pixels, producing more stable and noise-resilient segmentation outcomes. This work further enhances segmentation performance by coupling WFCM with the EPO (Emperor Penguin Optimizer), a bio-inspired meta-heuristic tool that emulates the collective huddling behavior strategy of penguins (emperor) to achieve a most suited trade-off between exploration with exploitation in the search space. The suggested hybrid EPO-WFCM framework is applied to brain MRI data for tumor identification, with its performance validated against ground-truth images using recognized evaluation metrics - offering clinicians a faster, more dependable diagnostic tool for treatment planning.