Deep learning model for screening causes of activated partial thromboplastin time prolongation using clot waveform...

Area of Science:

• Hematology
• Medical Diagnostics
• Artificial Intelligence in Medicine

Background:

• Activated partial thromboplastin time (APTT) prolongation requires differentiation of causes like factor deficiencies, lupus anticoagulants (LA), and anticoagulant use.
• Conventional clot waveform analysis (CWA) has moderate accuracy due to reliance on visual interpretation and limited parameters.

Purpose of the Study:

• To develop a highly accurate classification model for APTT prolongation causes using deep learning (DL) on CWA data.
• To leverage multi-wavelength detection and DL to extract hidden features from clot waveforms for improved diagnostic discrimination.

Main Methods:

• Applied convolutional neural network-based DL models to numerical data from single- and multi-wavelength CWA curves.
• Trained and evaluated the DL model on 683 patient samples with various causes of APTT prolongation using 10-fold cross-validation.

Main Results:

• DL models using single-wavelength CWA curves achieved high diagnostic performance (AUC 0.943-0.988).
• Multi-wavelength CWA curves further enhanced performance (AUC 0.961-0.993), demonstrating high sensitivity (≥88.0%) and specificity (>92.0%).
• Conventional CWA parameters showed limited discrimination (AUC 0.532-0.858).

Conclusions:

• Deep learning models, particularly those utilizing multi-wavelength CWA curves, show significant promise as high-performance screening tools for classifying APTT prolongation.
• The development of these DL models may be best implemented within individual laboratories, considering potential variations in reagents and analyzers.