Liver Tumour Detection and Segmentation through Graph Neural Network Enabled Optimized Model (GNN-EOM)

Early detection and accurate classification of
tumours remain critical challenges in medical image analysis, particularly due to complex spatial relationships and heterogeneous tumour structures. The proposed approach
leverages the MICCAI dataset to develop an effective graph
neural network (GNN)–enabled optimized model for early
tumour detection and classification. In this framework, medical images are transformed into graph representations where nodes capture salient anatomical and pathological features and edges encode spatial and contextual relationships among regions of interest. The GNN architecture exploits these relational dependencies to enhance feature learning beyond conventional convolution-based methods. To further improve performance,
an optimization strategy is incorporated to refine network
parameters and improve convergence, leading to robust
generalization across varying tumour shapes, sizes, and
intensity distributions. Experimental evaluation on the MICCAI
dataset demonstrates that the proposed model achieves
improved detection accuracy, reliable tumour classification, and
enhanced sensitivity to early-stage abnormalities. The results
indicate that graph-based learning provides a promising
direction for next-generation intelligent diagnostic systems,
supporting clinicians with more precise and early tumour
assessment.