Deep-learning instance segmentation and explainable rule-based classification to resolve platelets, red blood cells and aggregates in imaging flow cytometry

Abstract Platelets link blood clotting with inflammation and are studied in conditions such as Long COVID. With imaging flow cytometry (IFC) the small, lowcontrast and touching objects of the platelet fraction remain difficult to segment and classify, which motivates a deep-learning approach. This study presents a pipeline that pairs a fine-tuned Cellpose3 instance segmentation model with an explainable rule-based decision tree. On held-out images, the segmentation model achieved a mean average precision (mAP) of 0.665, compared to 0.143 for a classical threshold method. Each object is classified as a platelet, red blood cell (RBC) or aggregate. On 3000 hand-labeled images, the classification agreed with human annotation on 83.4% of objects. Four blood sample preparation methodswere compared, including a high-yield platelet optimization (HPO) method. A pilot cohort of eleven control and nine Long COVID participants was analyzed. Platelet, RBC and aggregate concentrations in the HPO fraction were higher in the Long COVID cohort. These differences are not clearly due to underlying biology and the results are hypothesis-generating.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-16
DOI
https://doi.org/10.21203/rs.3.rs-11287429/v1
Primary Topic
Cell Image Analysis Techniques
Type
preprint

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Deep-learning instance segmentation and explainable rule-based classification to resolve platelets, red blood cells and aggregates in imaging flow cytometry

Etheresia Pretorius, Chantelle Venter, Willem Jakobus Perold, Thomas Niesler et al.
3 links
Research Square
Cell Image Analysis Techniques
preprint

Deep-learning instance segmentation and explainable rule-based classification to resolve platelets, red blood cells and aggregates in imaging flow cytometry

Etheresia Pretorius, Chantelle Venter, Willem Jakobus Perold, Thomas Niesler, Andreas Gerhard Wild
preprint en

Abstract

Abstract Platelets link blood clotting with inflammation and are studied in conditions such as Long COVID. With imaging flow cytometry (IFC) the small, lowcontrast and touching objects of the platelet fraction remain difficult to segment and classify, which motivates a deep-learning approach. This study presents a pipeline that pairs a fine-tuned Cellpose3 instance segmentation model with an explainable rule-based decision tree. On held-out images, the segmentation model achieved a mean average precision (mAP) of 0.665, compared to 0.143 for a classical threshold method. Each object is classified as a platelet, red blood cell (RBC) or aggregate. On 3000 hand-labeled images, the classification agreed with human annotation on 83.4% of objects. Four blood sample preparation methodswere compared, including a high-yield platelet optimization (HPO) method. A pilot cohort of eleven control and nine Long COVID participants was analyzed. Platelet, RBC and aggregate concentrations in the HPO fraction were higher in the Long COVID cohort. These differences are not clearly due to underlying biology and the results are hypothesis-generating.

Research Square
Stellenbosch University (ZA)
Universiteit Stellenbosch
Good health and well-being
Cell Image Analysis Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon