Enhancing Lupus Anticoagulant Testing with Machine Learning: Deep Neural Networks Match Expert Performance without...

Area of Science:

• Medical Diagnostics
• Artificial Intelligence in Medicine
• Computational Biology

Background:

• Lupus anticoagulant (LAC) is crucial for diagnosing antiphospholipid antibody syndrome.
• Current LAC testing involves complex interpretation of up to 13 tests by physicians.
• Automating LAC interpretation is a significant goal for clinical laboratories.

Purpose of the Study:

• To explore the feasibility of using deep neural network (DNN) architectures for multilabel classification of LAC profiles.
• To assess the potential of DNNs in automating the interpretation of LAC testing.
• To compare different DNN architectures for accuracy and efficiency.

Main Methods:

• A dataset of 7,202 retrospective cases was used, randomly split for training, validation, and testing.
• Two DNN architectures were evaluated: single-output DNNs with feature selection and a multioutput DNN using all 13 inputs.
• LAC positivity (DRVVT, APTT) and anticoagulant presence (warfarin, heparin) were adjudicated by an expert.

Main Results:

• The domain-knowledge-naïve multioutput DNN showed comparable or improved performance across all four prediction tasks.
• Achieved high F1 scores: 0.977 for LAC-DRVVT, 0.954 for LAC-APTT, 0.961 for HEP, and 0.995 for WAR.
• Demonstrated the DNN's ability to learn feature importance without explicit selection.

Conclusions:

• Multioutput DNNs offer a versatile and potentially simpler approach to LAC interpretation compared to traditional methods.
• The comparable performance suggests DNNs can effectively standardize LAC diagnosis in clinical settings.
• The multioutput DNN is recommended for implementation to enhance efficiency and consistency in LAC testing.