Mitochondria are organelles that perform essential functions in cells. Their malfunctioning is linked to several diseases, notably neurodegenerative ones such as Parkinson's and Alzheimer's. Interestingly, in many cases, abnormal morphology (particularly, excessive fragmentation) and abnormal functioning are tightly linked in mitochondria. As such, their objective and accurate observation in live animals should deepen our understanding of their functioning in both health and disease. Advances in microscopy techniques and the introduction of animal lines with fluorescent mitochondria have made this possible. Lacking so far is an automated method of image processing to extract and quantify the degree of fragmentation of mitochondria from tissue images. Here, we applied statistical learning to automatically estimate the degree of fragmentation of mitochondria from tissue images, based on a training set and estimates of experts. We validated the method, and tested it on images of mitochondria in brains of living mice prior to and after cardiac arrest. The results indicate that our tool is sensitive to changes in mitochondrial morphology, and thus will be of use in detecting abnormalities in mitochondrial morphology. Further, it should be useful in quantifying the fragmentation in other cellular structures as well, such as neurites, endoplasmic reticula, and actin filaments.