Automatic Classification of Circulating Blood Cell Clusters based on Multi-channel Flow Cytometry Imaging

TL;DR

This work presents a two-stage framework for automatic analysis of circulating blood cell clusters (CCCs) in multi-channel flow cytometry imaging. The first stage uses a fine-tuned YOLO11m-cls to distinguish cluster-containing images from non-cluster images (achieving 95.03% accuracy), while the second stage employs YOLOv8-seg supplemented by Segment Anything Model (SAM) masks and HSV-based fluorescence analysis to phenotype CCCs via overlap with CD61 and CD45 signals. The approach demonstrates high performance: 95.03% cluster classification accuracy, 72.3% mask mAP0.5:0.95 for segmentation, and 91.2% phenotyping accuracy, with YOLO11m-cls and YOLOv8-seg outperforming several CNN and transformer baselines. The framework leverages both brightfield morphology and fluorescence cues to robustly identify and phenotype CCCs, offering potential applicability to immune and tumor cell clusters and enabling automated, high-throughput analysis suitable for clinical and research workflows.

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.

Figures (8)

• Figure 1: (A) Examples of brightfield images of cell clusters. (B) Examples of brightfield images containing no cell clusters. (B,I) no cells; (B,II) single cells; (B,III-IV) multiple separated cells. (C) Images containing cell clusters with fluorescent staining. The left column displays the brightfield images. The color of fluorescence-stained regions is used to identify specific cell cluster types, with green fluorescence-stained regions corresponding to platelets and yellow fluorescence-stained regions indicating WBCs. (C,I-III) Images are featured with staining artifacts. (C,IV-VI) Images with normal staining.
• Figure 2: The proposed framework automatically classifies CCCs using multi-channel flow cytometry imaging. The classification is performed through two sequential steps: I) We first filter out all the images that do not contain CCCs; II) We identify the cell types within the CCCs based on fluorescence-stained images. MetaSAM kirillov2023segment is applied for segmentation mask generation, significantly reducing manual labeling effort while maintaining mask quality over 372 cluster images.
• Figure 3: Identify the cell cluster type by evaluating the overlap of the area of the cell cluster and fluorescence-stained regions. (A) The example of multi-channel flow cytometry images of blood cell clusters from one patient with COVID-19. (B) Identifying the cell types within the clusters by overlapping the fluorescent-stained region on the cell cluster mask extracted from brightfield images. Left: Overlaps are detected between the cell cluster mask and fluorescent-stained for CD61, indicating the presence of platelets within the cell clusters. Right: No overlapping is detected between the cell cluster mask and fluorescent staining for CD45, indicating the absence of WBCs within the cell clusters.
• Figure 4: Comparison of YOLOv8 vs YOLOv11 classification models across five size variants (nano (n), small (s), medium (m), large (l), and extra-large (x)), showing model parameter count (x-axis, millions) versus 5-fold cross-validation mean accuracy (y-axis).YOLOv8 variants are shown as blue circular markers connected by solid lines; YOLOv11 variants as red square markers connected by solid lines. Each marker is labeled by its size abbreviation.
• Figure 5: Examples of misclassified images by (A) YOLO11, (B) ResNet18 and (C) ViT. (A, B, C, I) display the images containing no cell clusters are misclassified as cell clusters. (A, B, C, II) display the images containing cell clusters are misclassified as non-cluster images.