Artificial neural network supported multimodal analysis of living cells using simultaneous bright field and off-axis digital holographic microscopy

Artificial neural networks (ANNs) enhance cell biology by automating cell segmentation, counting, identification, and classification with increased accuracy. Label-free multimodal microscopy, in particularly off-axis digital holographic microscopy (DHM)-based quantitative phase imaging (QPI) combined with white-light bright-field (BF) microscopy, allows for minimally invasive determination of biophysical properties such as dry mass, refractive index, volume, and morphology. This combination also enables the detection of small intracellular organelles and the colocalization of absorbance and phase changes using noise-reduced BF images. Our previous work introduced a high-speed (video acquisition rate) multiplexed imaging approach using a single monochrome camera and 8-bit grayscale dynamic range multiplexing to simultaneously acquire BF and laser-based DHM QPI information (Opt. Lett. 48, 876-879 (2023)). This accelerates speed and avoids synchronization issues associated with sequential acquisition or multiple devices. QPI and BF images are recovered from multiplexed images via Fourier filtering and weighted subtraction. In this contribution, we extend the approach by using ANN-based algorithms to the recovered multimodal BF-QPI images for automatic and advanced segmentation and counting of living adherent cells. The performance of this approach is demonstrated by results from living pancreatic tumor cells and mouse fibroblasts.