Classification of factor deficiencies from coagulation assays using neural networks

Area of Science:

• Biomedical Engineering
• Clinical Pathology
• Artificial Intelligence in Medicine

Background:

• Activated partial thromboplastin time (APTT) and prothrombin time (PT) are standard coagulation assays.
• Current analysis focuses on clot times and signal magnitude.
• Enhanced diagnostic power may be achieved by interpreting additional assay parameters.

Purpose of the Study:

• To investigate if neural networks can increase the diagnostic utility of APTT and PT assays.
• To explore the use of multiple data parameters from coagulation assays for improved interpretation.
• To assess the capability of neural networks in identifying specific coagulation factor deficiencies and estimating concentrations.

Main Methods:

• Trained error back-propagation neural networks using multiple variables from APTT and PT optical data.
• Utilized data from 200 normal and abnormal patient specimens.
• Network functions included classifying factor deficiencies and estimating factor concentrations.

Main Results:

• Neural networks successfully identified specific factor deficiencies (<30% normal levels) with good specificity.
• Sensitivity for deficiency detection was variable.
• The estimation of precise coagulation factor concentrations by the networks was generally poor.

Conclusions:

• Neural networks show promise for identifying specific coagulation abnormalities from non-specific APTT and PT assays.
• Expanding the data parameter sets used in these assays can improve diagnostic capabilities.
• Further research may refine neural network applications for coagulation diagnostics.