Soma detection in Drosophila brain using Machine Learning

Abstract

Event Abstract Back to Event Soma detection in Drosophila brain using Machine Learning Guan-Wei He1, Nan-Yow Chen2, Ting-Yuan Wang3, ANN-SHYN CHIANG3, 4 and Yu-Tai Ching1, 5* 1 National Chiao Tung University, Department of Computer Science, Taiwan 2 National Center for High-Performance Computing, Taiwan 3 National Tsing Hua University, Institute of Biotechnology, Taiwan 4 National Tsing Hua University, Brain Research Center, Taiwan 5 National Chiao Tung University, Institute of Biomedical Engineering, Taiwan To compute the neuronal structure in the Drosophila Melanogaster’s brain is important to study the behaviors and gene function. Images of neurons are obtained by three dimensional confocal microscope. Two tasks in computing the neuronal structure are finding the cell-body and construct the neuronal structure. We present an automatic method to determine the cell-body by applying machine learning technique. Data set consists of raw data containing noises. The size of the image is 1024x1024x150. There could be more than one soma in the volume and the number of soma is not known. Data set consists of 100 volume images. These images are divided into 5-fold for cross validation. Proposed method starts with some candidate voxels that could be the center of the soma. These points are obtained by the distance transform of the image of the neuron. A regression model is then constructed based on the features that are voxel intensity, coordinate of the voxel, and density and balance factor around the candidate voxel. The result shows that sensitivity is more than 80% and true positive rate is near 90%.