Metrological and Algorithmic Traceability of Machine Learning in Laboratory Medicine

Area of Science:

• Clinical Laboratory Medicine
• Artificial Intelligence
• Machine Learning

Background:

• Machine learning (ML) integration in clinical laboratory medicine presents opportunities for quality improvement.
• ML models are being developed for test interpretation, outcome prediction, and operational support.
• These applications necessitate traceability for both measurement science and data computation.

Purpose of the Study:

• To examine the relationship between metrological traceability and algorithmic traceability.
• To analyze traceability requirements across three broad applications of ML in laboratory medicine.

Main Methods:

• Comparison of metrological traceability (linking results to reference systems via calibration and uncertainty) with algorithmic traceability (reconstructing computational paths).
• Evaluation of three ML applications in laboratory medicine and their specific traceability needs.

Main Results:

• Metrological traceability relies on calibration and uncertainty, while algorithmic traceability requires auditable computational paths.
• ML outputs need careful interpretation; traceable inputs do not guarantee traceable outputs.
• Persistent issues include data incomparability, undetected information leakage, underreported uncertainty, and post-deployment performance shifts.

Conclusions:

• A proposed framework aligns metrological principles with algorithmic documentation, distinguishing reference materials from training datasets.
• A tiered regulatory approach, traceability dossiers, and international collaboration are suggested.
• These measures aim to ensure AI/ML systems in laboratory medicine are reproducible, comparable, and clinically fit for purpose.