Isolation and reconstruction of cardiac mitochondria from SBEM images using a deep learning-based method

•The deep neural network CDeep3M segmented cardiac mitochondria with an F1-score 95%.•2315 mitochondria were segmented and reconstructed in 3D automatically.•Mitochondrial volumes between 0.04 and 0.06 µm3 were most common.•Approximately 10% of mitochondria had volumes greater than 2.0 µm3.•Adjacent inter-fibrillar mitochondria connected to extend over 5 sarcomere lengths.