Automatic Classification of Circulating Blood Cell Clusters based on Multi-channel Flow Cytometry Imaging
Suqiang Ma, Subhadeep Sengupta, Yao Lee, Beikang Gu, Xianyan Chen, Xianqiao Wang, Yang Liu, Mengjia Xu, Galit H. Frydman, He Li · 2025
Read Automatic Classification of Circulating Blood Cell Clusters based on Multi-channel Flow Cytometry Imaging by Suqiang Ma, Subhadeep Sengupta, Yao Lee, Beikang Gu, Xianyan Chen, Xianqiao Wang, Yang Liu, Mengjia Xu, Galit H. Frydman, He Li (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-17716.txt.
Introduction
Suqiang Ma and nine co-authors study clumps of blood cells that travel through the body. These clusters mix red cells, white cells, and platelets. Doctors link them to clotting, infection, and swelling in tissues. The team works with flow cytometry images, which pair bright views with glowing stains.
Their two-step method first sorts images into cluster and non-cluster groups. It uses a tuned YOLOv11 model, a fast object-spotting network. Next it matches cluster outlines to the stain channels to name each cell type. Tests passed ninety-five percent accuracy on both sorting tasks. The tool may later help study immune and tumor clusters too.
Catalog record
| CREATOR | Suqiang Ma, Subhadeep Sengupta, Yao Lee, Beikang Gu, Xianyan Chen, Xianqiao Wang, Yang Liu, Mengjia Xu, Galit H. Frydman, He Li |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 61 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-17716
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Automatic Classification of Circulating Blood Cell Clusters based on Multi-channel Flow Cytometry Imaging Suqiang Ma, Subhadeep Sengupta, Yao Lee, Beikang Gu, Xianyan Chen, Xianqiao Wang, Yang Liu, Mengjia Xu, Galit H. Frydman, He Li Abstract Circulating blood cell clusters (CCCs) containing red blood cells (RBCs), white blood cells (WBCs), and platelets are significant biomarkers linked to conditions like thrombosis, infection, and inflammation. Flow cytometry, paired with fluorescence staining, is commonly used to analyze these cell clusters, revealing cell morphology and protein profiles. While computational approaches based on machine learning have advanced the automatic analysis of single-cell flow cytometry images, there is a lack of effort to build tools to automatically analyze images containing CCCs. Unlike single cells, cell clusters often exhibit irregular shapes and sizes. In addition, these cell clusters often consist of heterogeneous cell types, which require multi-channel staining to identify the specific cell types within the clusters. This study introduces a new computational framework for analyzing CCC images and identifying cell types within clusters. Our framework uses a two-step analysis strategy. First, it categorizes images into cell cluster and non-cluster groups by fine-tuning the You Only Look Once(YOLOv11) model, which outperforms traditional convolutional neural networks (CNNs), Vision Transformers (ViT). Then, it identifies cell types by overlaying cluster contours with regions from multi-channel fluorescence stains, enhancing accuracy despite cell debris and staining artifacts. This approach achieved over 95% accuracy in both cluster classification and phenotype identification. In summary, our automated framework effectively analyzes CCC images from flow cytometry, leveraging both bright-field and fluorescence data. Initially tested on blood cells, it holds potential for broader applications, such as analyzing immune and tumor cell clusters, supporting cellular research across various diseases. Keywords: You Only Look Once, Convolutional Neural Networks, Vision Transformers, Residual Neural Networks, Blood Cell Cluster Classification, Flow Cytometry Imaging. 1 Introduction Background and significance. Coronavirus Disease 2019 (COVID-19), caused by SARS-CoV-2, is characterized by systemic hyperinflammation, cytokine storms, and an elevated risk of thrombotic complications [ 1 ] . The immune-thrombotic state induced by COVID-19 is a critical aspect of its pathogenesis, with current evidence indicating it is more severe than that caused by other viral pneumonia, such as Influenza H1N1 [ 2 , 3 , 4 ] . The recognized prothrombotic state associated with COVID-19 has spurred research into the pathology related to COVID-induced immunothrombosis. Potential mechanisms mediating COVID-related thrombosis include endothelial dysfunction, shifts in the procoagulant and antifibrinolytic plasma protein phenotype, platelet hyperactivity, and neutrophil extracellular traps (NETs) [ 5 , 6 , 7 , 8 ] . It is likely that a combination of these factors contributes to the prothrombotic state of COVID-19, with the exact contributors varying based on the patient’s disease state. For example, emerging clinical studies have shown that circulating cell clusters (CCCs), such as platelet-white blood cell (WBC) clusters, platelet-red blood cell (RBC) clusters, and WBC clusters, in the blood samples from patients with COVID-19 may serve as a nidus for immuno-thrombosis [ 9 , 10 , 11 , 12 ] . These CCCs could also have an adverse impact on the physiological functions of microvasculature by directly blocking small blood vessels in the microvasculature. In our recent study [ 13 ] , we have systematically studied the correlation between the various types of CCCs and the clinical outcomes of patients with COVID-19. In this study, 46 blood samples from 37 COVID-19 patients and 12 samples from healthy controls were analyzed with imaging flow cytometry. The results of imaging flow cytometry were used to detect, differentiate, and quantify CCC phenotypes for healthy controls and patients with COVID-19. It is noted that in this study, conventional imaging flow cytometry data analysis was performed using “manual gating”, as illustrated in Fig. 2 (A, I-II-IV) [ 14 , 13 ] , which pertains to several drawbacks, e.g., hard to reproduce, subjective and biased, and time-consuming particularly for large datasets. Related work. On the other hand, with the significant advances in graphics processing unit (GPU) computational power over the past decade, artificial intelligence techniques and computer-aided analysis have become essential tools for automatically extracting critical features from biomedical images across various modalities, aiding in disease diagnosis and prognosis [ 15 , 16 , 17 ] . Specifically, deep convolutional neural networks (DCNNs) have been extensively utilized for the efficient identification and classification of blood cells [ 18 , 19 , 20 , 21 , 22 ] . Unlike traditional methods that rely on hand-crafted features, DCNNs offer a more powerful approach. The combination of convolutional layers and subsampling layers within CNNs serve as feature extractors, capable of learning spatial hierarchical features directly from input images, thereby enhancing performance and accuracy. So far, artificial intelligence (AI) models based on DCNNs have achieved good performance in many applications involving blood cell classification [ 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 , 38 ] . Knowledge Gap. Despite the fact that AI models have been widely employed to perform blood cell classification, existing models were mostly designed to classify single blood cells and bio-particles without considering cell clusters containing heterogeneous cell types, which requires information from not only the bright-field images, but also multiple staining channels. In addition, in contrast to single cells, cell clusters often exhibit irregular shapes and varied sizes, depending on the number of cells and how they interact. Herein, we propose to develop a computational framework to perform automatic analysis and classification of different types of CCCs based on the outputs from brightfield images and multi-channel of protein staining. This framework is built on the You Only Look Once-11 (YOLOv8) [ 39 ] , the latest iteration of the renowned real-time object detection and image segmentation model, which has gained increasing popularity in various domains, including image classification [ 40 , 41 , 42 , 43 ] . Specifically, Chen et al. introduced NBCDC-YOLOv8, which enhanced YOLOv8 with SPD-Conv, MultiSEAM, and BiFPN modules for blood cell detection and classification, achieving \sim 94.7% mAP on the BCCD dataset [ 44 ] . A white blood cell detection study applied YOLOv8 to identify leukemia subtypes from \sim 3,629 images, reporting 95.1% accuracy [ 45 ] . An arXiv preprint demonstrates the use of YOLOv8 (and YOLOv11) in early detection of acute lymphoblastic leukemia, achieving \sim 98.8% accuracy [ 46 ] . In this study, we will demonstrate the effectiveness of this framework by classifying various CCC phenotypes, including RBC clusters, platelet clusters, WBC clusters, and WBC+platelet clusters, from flow cytometry images of blood samples from patients with COVID-19. As shown in Fig. 2 (B), In this study, we present a two-stage classification and phenotyping framework based on multi-channel imaging flow cytometry for the efficient and accurate detection and analysis of CCCs. Our principal contributions are summarized as follows: 1) Two-stage pipeline with semi-automated annotation. In the first stage, a lightweight YOLO11m-cl network performs binary classification of cluster versus non-cluster, achieving an accuracy of 95.03%. In the second stage, YOLOv8-seg generates cluster masks, supplemented by a semi-automated annotation workflow using Meta’s Segment Anything Model to produce 372 high-quality masks. These masks are combined with HSV-based thresholding