Year: 2026
Type: article
Abstract: Hepatitis C Virus (HCV) is a common blood-borne pathogen and a leading contributor to chronic liver disease worldwide, including cirrhosis and hepatocellular carcinoma. Early and accurate diagnosis is essential to halt progression and improve outcomes, yet conventional approaches can be invasive and costly, highlighting the need for accessible, non-invasive decision support. This study examines whether representing routine clinical information as relational structure can enhance classification using the publicly available UCI Hepatitis C Virus (HCV) dataset, which contains standard biochemical and demographic variables. Rather than treating each patient independently, patients with similar laboratory profiles were connected through a patient–patient similarity graph, and a graph-based learning approach (GraphSAGE) was applied to leverage these relationships. For context and fairness, widely used non-graph machine learning methods were also evaluated on the same splits. Across repeated runs, the graph-based model achieves very high and stable performance (99.51% for both accuracy and macro-F1) and remains consistently competitive with, and in many cases slightly superior to, traditional models. Because the relational structure is derived directly from routine laboratory values, the learned representations reflect clinically meaningful similarity patterns without requiring additional inputs. Taken together, the findings suggest that reframing standard HCV data as a patient–patient network is a practical and effective strategy for non-invasive computer-aided diagnosis, complementing established baselines while offering a flexible path toward robust, clinically plausible deployment.
Language: Turkish
Subfield: Health Information Management
Field: Health Professions
Domain: Health Sciences
Open Access status: diamond
OpenAlex ID: W7162947503