Minimizing unnecessary radiological imaging in gallstone diagnosis: a danets-based approach using bioimpedance and clinical data

Creative Commons License

BOZKURT M. H.

Gumushane Universitesi Fen Bilimleri Dergisi, cilt.16, sa.2, ss.682-697, 2026 (Scopus, TRDizin)

Özet

While ultrasonography and computed tomography (CT) remain the gold standards for gallstone diagnosis, their utility in primary screening is limited by high costs, ionizing radiation exposure, and acoustic attenuation in high-BMI patients. Bioelectrical impedance analysis (BIA) provides a cost-effective, non-invasive alternative; yet, traditional machine learning models often struggle to capture the high-dimensional, nonlinear metabolic relationships inherent in BIA data. This study evaluates a Deep Abstract Networks (DANets) framework, optimized for tabular data, to integrate BIA metrics with routine clinical parameters for gallstone classification. Unlike conventional tree-based or shallow architectures, the employed DANets model utilizes an abstraction layer to autonomously extract hierarchical features, mapping nonlinear correlations between hepatic biomarkers and body composition. The framework was validated on a clinical dataset of 319 patients. Rigorous methodological protocols, including 10-fold cross-validation and Bayesian hyperparameter optimization, were implemented to ensure robust evaluation. DANets achieved an accuracy of 88.08% (±3.91%) and an F1-score of 87.11%, demonstrating a statistically significant performance improvement over baseline gradient boosting (XGBoost: 83.07%) and multi-layer perceptron (MLP: 81.83%) models. Notably, the model yielded a precision of 93.27%. This rate suggests the model could help reduce the number of unnecessary radiological examinations by minimizing false positive alarms. In summary, the evaluated DANets approach serves as a promising proof-of-concept for a cost-effective clinical screening tool, warranting further external validation.