Changes to mitochondrial architecture are associated with various adaptive and pathogenic processes. However, quantification of changes to mitochondrial structures are limited by the yet unmet challenge of defining the borders of each individual mitochondrion within an image. Here, we describe a novel method for segmenting Brown Adipose Tissue (BAT) images. We describe a granular approach to quantifying subcellular structures, particularly mitochondria in close proximity to lipid droplets, peri-droplet mitochondria. In addition, we lay out a novel machine-learning-based mitochondrial segmentation method that eliminates the bias of manual mitochondrial segmentation and improves object recognition compared to conventional thresholding analyses. By applying these methods, we discovered a significant difference between cytosolic and peridroplet BAT mitochondrial H2O2 production, and validated the machine learning algorithm in BAT via norepinephrine-induced mitochondrial fragmentation and comparing manual analyses to the automated analysis. This approach provides a higher-throughput analysis protocol to quantify ratiometric probes in subpopulations of mitochondria in adipocytes.

High-Throughput Image Analysis of Lipid-Droplet-Bound Mitochondria