Fig. 1: MoDL’s precise mitochondrial morphological analysis strategy.
From: Mitochondrial segmentation and function prediction in live-cell images with deep learning
a The first key pipeline of MoDL employs deep learning algorithms for precise and high-quality segmentation of mitochondrial fluorescence images. MoDL was trained on a set of over 20,000 independently manual annotations, derived from original fluorescence images obtained by SR microscopy. b The second key pipeline of MoDL centers on the regression analysis to accurately predict mitochondrial functions, including the mitochondrial membrane potential (MMP) polarization, respiration rate, reactive oxygen species (ROS) production, adenosine triphosphate (ATP) generation, and mitophagy level. This process was enhanced by a dataset for a training ensemble learning algorithm that comprising over 100,000 images and annotated functionalities. c The third key pipeline of MoDL, employing data fine-tuning and retraining alongside a substantial dataset, enables the precise prediction of mitochondrial functions in drug-resistant cancer cells, which exhibit abnormal morphology patterns, with small sample size training (created with BioRender.com).
