MLPTR-CC: Multi-label Pathology Test Recommendation using Classifier Chains and SHAP

The system recommends multiple tests together rather than one at a time: Unlike existing clinical decision-support tools that predict single diseases, this framework recognizes that pathology tests are interdependent and recommends sets of tests that work together, potentially improving diagnostic efficiency.

Classifier chains outperformed independent models on real clinical data: When tested on an expert-curated dataset from a pathology laboratory, the classifier-chain approach improved F1-scores and reduced errors (Hamming loss) compared to models that treat each test independently, while maintaining high accuracy for both common and rare tests.

SHAP explainability helps clinicians understand AI recommendations: The framework integrates SHAP-based explanations that show which patient symptoms drive each test recommendation, and these explanations "align with established clinical reasoning in most cases," helping clinicians verify that the AI's logic makes medical sense.

Transparency was built in as a core feature, not an afterthought: Rather than treating explainability as optional, the authors integrated SHAP from the start to provide "symptom-level attributions," addressing a key concern in responsible AI—that clinicians need to understand why a system recommends specific tests before trusting it in practice.

The approach shows promise for early-stage diagnostic support: By recommending appropriate test combinations at an early stage based on symptoms, this system could help clinicians make more informed decisions about which diagnostic tests to order, potentially reducing unnecessary testing while improving diagnostic accuracy.