A CNN-LASSO ensemble classification model for incomplete antibody reactants screening in coombs test

Area of Science:

• Medical Diagnostics
• Immunohaematology
• Machine Learning in Healthcare

Background:

• Accurate classification of incomplete antibody reactants (IAR) is crucial for preventing incompatible blood transfusions.
• Current manual classification methods are prone to human error, necessitating automated solutions.

Purpose of the Study:

• To develop an ensemble learning algorithm for precise IAR intensity classification.
• To integrate convolutional neural networks and LASSO regression for an automated IAR classification model.

Main Methods:

• An ensemble model was trained on 1302 IAR samples and validated on 326.
• The model integrates five convolutional neural networks and LASSO regression.
• Human-machine interaction was visualized using chord diagrams.

Main Results:

• The ensemble model achieved high accuracies across all IAR categories (98.4%–99.7%).
• Manual classification by immunologists had an average accuracy of 75.6%.
• Model-assisted classification improved immunologist accuracy by an average of 8.4%.

Conclusions:

• The proposed algorithm effectively enhances IAR intensity classification accuracy and efficiency.
• This facilitates the automation of haemolytic disease screening equipment.
• The findings support the use of AI in improving blood transfusion safety.