Label-free prediction of fluorescence markers in bovine satellite cells using deep learning

TL;DR

A U-Net-based CNN model was employed to predict multiple fluorescence signals from a single bright-field microscopy image of cell culture, and two key biomarkers, DAPI and Pax7, were used to determine the abundance and quality of BSCs.

Abstract

Assessing the quality of bovine satellite cells (BSCs) is essential for the cultivated meat industry, which aims to address global food sustainability challenges. This study aims to develop a label-free method for predicting fluorescence markers in isolated BSCs using deep learning. We employed a U-Net-based CNN model to predict multiple fluorescence signals from a single bright-field microscopy image of cell culture. Two key biomarkers, DAPI and Pax7, were used to determine the abundance and quality of BSCs. The image pre-processing pipeline included fluorescence denoising to improve prediction performance and consistency. A total of 48 biological replicates were used, with statistical performance metrics such as Pearson correlation coefficient and SSIM employed for model evaluation. The model exhibited better performance with DAPI predictions due to uniform staining. Pax7 predictions were more variable, reflecting biological heterogeneity. Enhanced visualization techniques, including color mapping and image overlay, improved the interpretability of the predictions by providing better contextual and perceptual information. The findings highlight the importance of data pre-processing and demonstrate the potential of deep learning to advance non-invasive, label-free assessment techniques in the cultivated meat industry, paving the way for reliable and actionable AI-driven evaluations.

Figures (4)

• Figure 1: Schematic diagram for quality assessment of bovine cell culture: (A) Traditional immunofluorescence microscopy method; (B) Digital staining method using deep learning for predicting fluorescence markers. Brightness and contrast of the example images were adjusted for publication clarity.
• Figure 2: Image pre-processing pipeline for generating target fluorescence signals from ground truth fluorescence images. (A) Original ground truth data obtained by fluorescence microscopy were converted into (B) grayscale images, followed by (C) fluorescence denoising, and then (D) final normalization to enhance the target fluorescence signals for model training. Brightness and contrast of the example images were adjusted for publication clarity.
• Figure 3: Example images of model prediction of multiple fluorescence markers from a single bright-field (BF) image: (A) DAPI and (B) Pax7. Model predictions include direct output followed by post-processing using digital image enhancement techniques.
• Figure 4: Statistical performance evaluation metrics for DAPI and Pax7 predictions without (top row) and with (bottom row) fluorescence denoising. (A,D) Pearson correlation coefficient: higher values indicate better correlation. (B,E) SSIM: Structural Similarity Index, higher values indicate greater similarity. (C,F) MSE: Mean Squared Error, lower values indicate better accuracy.