Mitochondrial organelle movement classification (fission and fusion) via convolutional neural network approach

Abstract

Mitochondria are highly dynamic cellular organelles with the ability to change size, shape, and position over the course of a few seconds. Mitochondrial organelle movement refers to the problem of finding fission and fusion and generates energy for the cell. In this paper, we proposed a deep learning method [mitochondrial organelle movement classification (MOMC)] for mitochondrial movement classification using a convolutional neural network. We present a three-step feature description strategy, such as local descriptions, which is first extracted via the GoogLeNet, followed by the production of mid-level features by ResNet-50 and global descriptor features by Inception-V3 model and final classification of the position of mitochondrial organelle movement. Our method consists of a deep classification network, MOMC for gathering the organelle position, and a verification network for classification accuracy by removing false positives. Using machine learning methods, logistic regression (LR), support vector machine (SVM), and convolutional neural networks (CNNs), we found that the CNN better classified the shape of mitochondrial organelles (fission and fusion). Employing 24 types (position) of images, a convolutional neural network was trained to identify mitochondrial organelle movement with 96.32% accuracy. This enabled the discovery of position, further advancing the clinical utility of human mitochondrial organelles.