Advancing precision single-cell analysis of red blood cells through semi-supervised deep learning using database of patients with post-COVID-19 syndrome

Andrey Kurenkov, Aigul Kussanova, Natasha S. Barteneva

2024

Data confidence 70 medium

How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

We developed a semi-supervised deep-learning-based system classifying different types of red blood cells (RBCs) images based on their shape, texture, and size. Specifically, pre-training a convolutional neural network was done on over 35,000 brightfield images of RBCs acquired with an imaging flow cytometer from a post-COVID-19 patient cohort. The system utilizes object localization powered by a YOLO-inspired block for cell identification and a de-blurring CNN block based on FocalNet. A series of convolutional and fully connected layers classifies images into side-view, biconcave, elongated, and additional categories for reticulocytes and erythrocytes. Fine-tuning was done using 7,000 manually labeled brightfield images. Consequent evaluation on a test dataset of 3,000 samples yielded an accuracy of 98.2%. This system can be used for other cell analysis tasks, not requiring large fine-tuning datasets while maintaining high efficiency.