Fully automated platelet differential interference contrast image analysis via deep learning.

Area of Science:

• Hematology
• Biotechnology
• Medical Imaging

Background:

• Platelets are key mediators of arterial thrombosis, a major cause of myocardial infarction and stroke.
• Platelet activation and substrate interactions are studied using platelet spreading assays, typically analyzed manually via microscopy.
• Manual analysis of these assays is prone to significant inter-annotator variability and bias.

Purpose of the Study:

• To develop and validate a fully automated method for analyzing platelet spreading assays using convolutional neural networks (CNNs).
• To compare the performance of the CNN-based analysis against manual annotation by multiple experts.
• To assess the CNN's ability to accurately quantify platelet morphology across various substrates and experimental conditions.

Main Methods:

• A convolutional neural network (CNN) was trained on 120 generalized training images for automated analysis of platelet spreading assays.
• Differential interference contrast (DIC) microscopy images of platelet spreading were used for training and validation.
• The CNN's performance was benchmarked against manual annotations from six human annotators.

Main Results:

• The CNN achieved automated analysis of platelet spreading assays captured by DIC microscopy.
• Increasing training data improved the CNN's mean average precision.
• The CNN demonstrated consistent quantification of spread platelet area across diverse substrates and in the presence of inhibitors and agonists, comparable to published data.

Conclusions:

• A CNN enables the first fully automated analysis of platelet spreading assays using DIC microscopy.
• Automated analysis mitigates the bias and variability inherent in manual annotation.
• This AI-driven approach offers a robust and reproducible method for investigating platelet function in thrombosis research.