Developing and validating a highly sensitive platelet clump detection model for the Sysmex haematology analyser.

Area of Science:

• Hematology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Mainstream hematology analyzers (HAs) exhibit limited sensitivity in detecting platelet clumps.
• Platelet clumps can interfere with accurate blood test results.

Purpose of the Study:

• To develop and validate a deep learning (DL) algorithm, a convolutional neural network (CNN), for enhanced detection of platelet clumps.
• To compare the performance of the CNN algorithm against the internal algorithm of a standard hematology analyzer (Sysmex XN-10).

Main Methods:

• Adenosine diphosphate (ADP) was used to induce platelet aggregation, creating samples with platelet clumps.
• Leukocyte scattergrams from the Sysmex XN-10 were collected and used to train and validate multiple CNN models via cross-validation.
• The optimal CNN model was tested on practical routine work samples.

Main Results:

• The best-performing CNN, utilizing scattergrams from the white count and nucleated red blood cells (WNR) channel, achieved high accuracy (0.940 in cross-validation, 0.961 in practical tests) and sensitivity (0.942 in cross-validation, 0.965 in practical tests).
• The CNN demonstrated a strong area under the curve (AUC) of 0.968 (cross-validation) and 0.916 (practical test).
• Dispersed spots around leukocytes in the WNR channel were identified as potential indicators of platelet clumping.

Conclusions:

• CNN algorithms can effectively identify platelet clumps using optical data from leukocyte channels.
• The developed CNN approach demonstrates superior ability in detecting platelet clumps compared to the Sysmex XN-10's internal algorithm in real-world settings.