Abstract
Imaging flow cytometry (IFC) produces vast quantities of data that pose significant challenges for storage, management, and analysis. These approaches are often time-consuming, inflexible, and limited in their ability to uncoverrare subpopulations. Eidocell, a novel open-source desktop application implemented in Python and C++, addresses these challenges using modern computational techniques and an intuitive design for single-cell image analysis. The software integrates clustering and computer vision models, like convolutional neural networks (CNNs) and transformers, for robust visual feature extraction and assisted classification. Unlike conventional workflows that require users to annotate and classify samples based primarily on quantitative mask features, Eidocell uses a scale-friendly approach that does not use pre-defined populations and keeps advantage of deep learning methods. This approach minimizes bias and is particularly advantageous for experiments with heterogeneous or poorly defined sample populations.
Eidocell has a modular design, allowing users to incorporate custom ONNX models for feature extraction and automatic fine-tuning of built-in models on user datasets and eliminating the need for scripting model training for common cases. The application also includes a suite of segmentation algorithms - both classical and deep learning-based - with adjustable parameters to refine object detection and morphological analysis of subcomponents within a single image. Measured parameters such as area, circularity, and other phenotypic descriptors are visualized through interactive charts, which provide visualization and gating capabilities. For validation, Eidocell was tested on datasets of microalgae, phytoplankton, blood cells, and spermatozoa from ImageStream X MkII (Amnis-Cytek) and FlowCam (Yokogawa Fluid Imaging Technologies) instruments. Future development aims to expand its modular capabilities and enhance support for existing IFC platforms for applications in cell biology, ecology, and medical diagnostics.
Eidocell has a modular design, allowing users to incorporate custom ONNX models for feature extraction and automatic fine-tuning of built-in models on user datasets and eliminating the need for scripting model training for common cases. The application also includes a suite of segmentation algorithms - both classical and deep learning-based - with adjustable parameters to refine object detection and morphological analysis of subcomponents within a single image. Measured parameters such as area, circularity, and other phenotypic descriptors are visualized through interactive charts, which provide visualization and gating capabilities. For validation, Eidocell was tested on datasets of microalgae, phytoplankton, blood cells, and spermatozoa from ImageStream X MkII (Amnis-Cytek) and FlowCam (Yokogawa Fluid Imaging Technologies) instruments. Future development aims to expand its modular capabilities and enhance support for existing IFC platforms for applications in cell biology, ecology, and medical diagnostics.
Original language | English |
---|---|
Pages | P282 |
Number of pages | 1 |
Publication status | Published - May 31 2025 |
Event | CYTO 2025 - May 31-June 4 - Denver, Colorado - Denver, United States Duration: May 31 2025 → Jun 4 2025 |
Conference
Conference | CYTO 2025 - May 31-June 4 - Denver, Colorado |
---|---|
Country/Territory | United States |
City | Denver |
Period | 5/31/25 → 6/4/25 |
Keywords
- machine learning
- imaging flow cytometry
- FlowCam
- Imagestream