Automatic identification of glomerulus in the antenna lobe of Drosophila Melanogaster

Antenna lobe (AL) in the Drosophila brain is composed of ~50 glomeruli. Each of the glomeruli has a stereotyped shape, position, and neuronal composition. However, manually identification of the glomeruli usually is very time consuming and demands high familiarity with the structure. In this project, we applied Principle Component Analysis method via Singular Value Decomposition to identify a single glomerulus in AL. 105 training examples were used to construct the eigen basis for the AL image. Based on the coefficient of constructing AL eigen basis, the eigen basis for the target glomerulus images in the training set was also calculated. To identify the target glomerulus in the test examples, its AL image information was used as the input. The test results from the about 40 examples showed reasonably good reliability and robustness of the algorithm. Although we only trained the algorithm to identify a certain glomerulus (V glomerulus) in this case, this approach should also be applicable to other glomeruli so that many glomeruli in the AL can be recognized automatically.