Configurable Fuzzy Logic Ensemble Matches Standard Methods for Multi-Label Chest X-ray Classification

A fuzzy logic-based ensemble for chest X-ray classification achieved an AUROC of 0.855 and AUPRC of 0.299 in a sensitivity-priority setting, comparable to Soft Voting (0.851 AUROC), while enabling clinician-configurable diagnostic priorities.

• Five base architectures (ViT, SwinV2, DenseNet121, and hybrids) were trained on NIH ChestX-ray14 with 10 random seeds.
• The fuzzy analytic hierarchy process (FAHP) ensemble allowed disease-specific model weighting and utility-guided aggregation (e.g., sensitivity priority).
• Per-disease analysis showed heterogeneous results; FAHP favored some conditions like Effusion, Emphysema, Mass, Nodule, and Pleural Thickening.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.