An Image Recognition Algorithm for Automatic Counting of Brain Cells of Fruit Fly

Today’s computing power enables us to simulate systems consist of tens or hundreds of thousands of realistic model-neurons such as the HodgkinHuxley neuron. What is absorbing is the number (10, 000 ∼ 1, 000, 000) here is roughly same to the number of brains of little but higher animals. An excellent example is the fruit fly (Drosophila melanogaster), whose brain consists of no more than 200, 000 neurons while it has abilities of flight and the sense of sound, visual, and smell [1]. Furthermore, observations on behavioral disturbances have shown that it has even higher level of intelligence such as learning and memory. Therefore it is expected that the simulations on coupled tens of thousands of model neuron system will reveal some secrets of the mechanisms of such intelligence. However, this “brain reproduction” is actually impossible since we do not know how the neurons in the fly’s brain are connected. At present C. elegans is the unique example that the connectivity among all of its 300 neurons’ is known [2]. About the fruit fly, in fact, we do not have even the accurate number of the neurons. Though the circuit information of its brain is not available, Drosophila has the following advantages in comparison with other higher animals. First, its all genome is known [3]. Second, the molecular genetics methods to label specific cells of the brain is well established [4, 5]. Third, its brain is small enough to observe the whole system virtually at the same time. The second and third facts together permit us to obtain the three-dimensional image data of the labeled cells [4, 5]. Thus, what should be tried is to extract the topological information of the neural circuit from the obtained image data. In the following we introduce a new algorithm of counting cells automatically from source images. Counting cells is essential not only because its biological importance of just giving an accurate value of the total number of neurons against the current rough estimation (40, 000 ∼ 200, 000), but because getting the detailed positions of cells is necessary to start the extractions of brain circuit information.