Label-free cell counting and viability prediction with brightfield imaging and deep learning
Authors: Amir Reza Vazifeh, Christian Zeigler, Sornanathan Meyyappan, Richard Jeske, Jason W. Fleischer
Organizations: Department of Electrical and Computer Engineering, Princeton University, Princeton, NJ 08544, USA · Waters Corporation, Immerse Cambridge, 301 Binney Street, Suite 102, Cambridge, MA 02142, USA · Waters Corporation, 34 Maple St, Milford, MA 01757, USA · Waters Corporation, Immerse Delaware, 590 Avenue 1743, Newark, DE 19130, USA
Cell viability assessment is a core requirement in cell culture systems, with critical applications in biopharmaceutical manufacturing and drug development. Conventionally, it is measured by adding membrane-impermeable dyes to a sample (a process called staining), which allows compromised cell membranes to be distinguished from intact ones. However, staining has several limitations: (a) chemical agents can perturb normal cellular processes of the cells being measured, (b) it is often ambiguous to assign viability to individual cells whose membrane integrity is only partially compromised. (c) photobleaching can undermine measurement accuracy over time when using fluorescent stains, and (d) staining cannot be performed in situ or in real time. Here, we show that (1) stained cells captured under brightfield imaging contain sufficient information to distinguish live and dead cells, and (2) cells captured under unstained brightfield imaging exhibit similar image features to their stained counterparts, enabling models trained on stained cells to generalize to unstained ones. We then report the development and validation of ViabiLens, an AI-assisted software for label-free cell viability analysis. The ViabiLens combines a cell detection model for localizing individual cells with a convolutional neural network (CNN) classifier for live/dead prediction, paired with an interactive UMAP-based viewer for visualizing and exploring individual cells across the sample. Evaluated on Chinese Hamster Ovary (CHO) cells spanning a wide range of viability conditions, ViabiLens achieves a mean absolute error of 2.68% on unstained samples against fluorescence-based reference measurements. We also release a benchmark dataset for label-free cell viability analysis to facilitate future research, available at https://amirrezavazifeh.github.io/ViabiLens-Project-Page/.
Figures & tables
Figure 1: Comparison of traditional and proposed methods for cell viability assessment. Left: A bioreactor used for cell culture. Middle: Traditionally, cells are stained and fluorescently imaged to distinguish live from dead cells. Right: The proposed deep-learning-based method detects and classifies individual cells as live or dead directly from brightfield images, without staining or fluorescence imaging. In either case, sample-level viability is computed as the ratio of live cells to the total cell count.
Figure 2: Overview of the cell detection and viability analysis pipeline. The brightfield image is divided into 100 overlapping crops (25% overlap in each direction), with red-shaded regions indicating overlap with adjacent crops. Each crop is independently fed into a trained model, and only bounding boxes falling fully within the green frame are retained to discard incomplete cells near boundaries. Non-Maximum Suppression (NMS) with an Intersection over Union (IoU) threshold of 0.3 is then applied to remove redundant detections within each crop. Bounding box coordinates are then shifted to full image space, where a second NMS (at IoU=0.2 ) eliminates duplicate detections arising from cells appearing in red-shaded regions. Each detected cell is resized to 32 × 32 pixels and classified as live or dead by the CNN model.
Figure 3: Comparison of live and dead cells under brightfield imaging. Top row: Two-dimensional embeddings obtained via PCA, t-SNE, and UMAP, computed from 137,434 stained cells captured from 60 samples. Live cells are shown in green and dead cells in red. A k-NN classifier is applied to the 2D embeddings and original cells for live vs. dead classification, as shown in the rightmost panel. Bottom row: Average Fourier spectra of live and dead cell patches (left two panels) and their horizontal and vertical cross-sections through the frequency origin (right two panels).
Figure 4: Unsupervised phenotyping of live and dead cells via UMAP. Left: UMAP embedding colored by cluster, with clusters manually determined based on visual separation. Right: representative brightfield cell patches from each cluster. Green and yellow frames around each cell represent live and dead, respectively.
Figure 5: Comparison of stained and unstained cells under brightfield imaging. Top row: 2D embeddings obtained via PCA, t-SNE, and UMAP, computed from 18,710 cells captured from 15 stained and unstained image pairs. Unstained patches were randomly subsampled to match the stained count; the two sets are not paired one-to-one. Middle row: 20 examples of stained and unstained cells. Bottom row: Average Fourier spectra of stained and unstained cellS (left two panels) and their horizontal and vertical cross-sections through the frequency origin (right two panels).
Samples
Total Cell Count
Precision / Recall / F1
mAP
PLX
Img. Proc.
Cellpose [ 42 ]
Ours
Img. Proc.
Cellpose [ 42 ]
Ours
Ours
Train
45 stained samples
112,947
105,190
136,334
132,680
0.56 / 0.52 / 0.53
0.76 / 0.89 / 0.81
0.84 / 0.95 / 0.89
0.90
Test
30 stained samples
32,543
29,000
33,868
33,083
0.80 / 0.73 / 0.76
0.80 / 0.83 / 0.81
0.92 / 0.95 / 0.94
0.93
15 unstained samples
NA
14,305
17,667
16,889
NA
NA
NA
NA
Table 1: Cell detection results on stained and unstained samples. PLX serves as the reference measurement; since it only provides bounding-box annotations for stained samples, some unstained metrics are marked NA.
Figure 6: Qualitative comparison of cell detection methods on stained and unstained brightfield images. Top two rows: PLX output used as reference annotation, shown alongside its two fluorescent channels: Ch1 (live cells, green bounding boxes) and Ch2 (dead cells, red bounding boxes). Detection results from the image-processing baseline, Cellpose, and our model are shown; confidence scores on our model’s detections reflect the predicted probability of each proposal being a cell. Bottom row: Bounding boxes from aforementioned methods on unstained samples. PLX relies on fluorescent channels and therefore does not support unstained samples.
Figure 7: Performance of our pipeline on cells under ten different focal planes. Cell appearance changes across focal planes, and while detection remains robust, the viability classifier can be more sensitive to focal planes outside its training distribution. Cells are also moving, so frames are not perfectly registrable. The video is available here .
Figure 8: Performance of the binary cell viability classifier. Left: training and validation confusion matrices. Middle: ROC curve and precision-recall curve. Right: loss and accuracy curves across 50 epochs.
Figure 9: Sample-level viability predictions and UMAP projections for stained and unstained samples. Scatter plots show predicted viability against PLX viability measurement, with the dashed line indicating perfect agreement. UMAP projections show live (green) and dead (yellow) cell populations, where stained samples use PLX-measured labels and unstained samples use CNN-predicted labels. Representative cell image samples from each cluster are shown below the corresponding UMAP projection.
Counting living cells is an important step in many biological research workflows. Our collaborators at the Wellcome Sanger Institute study vital genes in humans via large scale saturation genome editing screening, which requires repeatedly counting cells a great number of times. Computer Vision based automation is crucial for high throughput and resource efficiency. In this work, we develop a regression-based deep learning computer vision algorithm to detect and count cells in phase-contrast microscopy images. To reduce annotation effort, which in practice often becomes a bottleneck, we focus on counting cells only using sparse point annotations, which are fast and easy to acquire. By comparison to state-of-the-art 0-shot methods, we show that regression-based counting is a promising alternative in low data regimes. Through developing methods to automatically count living cells in microscopy images, we contribute to valuable research on the human genome. The code is available at https://github.com/beijn/cellnet.
Benjamin Eckhardt, Dmytro Fishman, Stuart Fawke +3
University of Göttingen · University of Tartu · Wellcome Sanger Institute
Label-free virtual staining offers a compelling, non-destructive alternative to standard histopathology; however, its clinical adoption is hindered by the computational bottlenecks inherent to processing gigapixel Whole Slide Images (WSIs). Current deep learning approaches require patch-based inference to avoid memory constraints, which disrupts global tissue continuity and introduces tiling artifacts--displaying visible seams and color shifts. To address this, we introduce the Consistency Memory Bank (COMB), a novel label-free virtual staining framework that enforces spatial and channel consistency across tiles without memory bottlenecks. COMB decouples context storage from computation, utilizing a dynamic retrieval mechanism to fetch feature representations from adjacent tiles. This enables a retrieval-based context integration strategy that adopts local padding to resolve spatial discontinuities and neighbor-aware channel attention to stabilize statistical drift. Further optimized with a sliding window schedule to ensure minimal memory overhead, our method demonstrates superior performance over state-of-the-art baselines, achieving significant improvements in both perceptual fidelity and tiling consistency, while suggesting its downstream utility in tumor segmentation. Code is available at https://github.com/dou0000/COMB.
Label-free single-cell imaging offers a scalable, non-invasive alternative to fluorescence-based cytometry, yet inferring molecular phenotypes directly from bright-field morphology remains challenging. We present a unified Deep Learning (DL) framework that jointly performs White Blood Cell (WBC) classification and continuous protein-expression regression from label-free Differential Phase Contrast (DPC) images. Our model employs a Hybrid architecture that fuses convolutional fine-grained texture features with transformer-based global representations through a learnable cross-branch gating module, enabling robust morpho-molecular inference from DPC images. To support downstream interpretability, we further incorporate a Large Language Model (LLM) that generates concise, biologically grounded summaries of the predicted cell states. Experiments on the Berkeley Single Cell Computational Microscopy (BSCCM) and Blood Cells Image benchmarks demonstrate strong performance, achieving a 91.3% WBC classification accuracy and a 0.72 Pearson correlation for CD16 expression regression on BSCCM. These results underscore the promise of label-free single-cell imaging for cost-effective hematological profiling, enabling simultaneous phenotype identification and quantitative biomarker estimation without fluorescent staining. The source code is available at https://github.com/saqibnaziir/Single-Cell-Phenotyping.
Saqib Nazir, Ardhendu Behera
Department of Computer Science, Edge Hill University, UK