Research activity in 1999
Models for neurosensory processing have been developed for vision and magnetoperception in arthropods. Nervous systems process different aspects of vision in arthropods as well as in humans in parallel channels. In mantis shrimps one channel processes localization, size and motion of a prey or enemy, a parallel one the figure-ground discrimination of the same object. All inputs derive from the same photoreceptors but are distributed to different sets of neurons. The processing in different channels can be described mathematically and simulated on the computer. The outputs from the first processing station are used as inputs to different types of ANN which learn to classify inputs according to the size and shape, the direction and speed of motion of the target. In parallel the figure-ground discrimination is processed. The output of this channel is correlated to the first channel. The result is an information which can be forwarded to an actuator network which distributes its outputs to the muscle sets which execute the adequate motion of the animal.
Signal transmission in a stomatopod eye was studied using virtual reality. Transduction and nervous processing are illustrated by illumination of receptors and nerve fibers when light is absorbed in stimulated photoreceptors. Signals are then transmitted through integrating fibers. Impulses are visualized by light flashes travelling along the fibers showing the spatio-temporal distribution of the signals.
As in most animals, where it has been looked for, honeybees contain magnetite. Bees contain both SD and SPM magnetite. While the SD magnetite presumably determines a compass sense, SPM magnetite determines a map sense. Granules similar to SPM magnetite have been seen in compact structures in hairs with dendrites on the anterodorsal abdomen where sensitivity to geomagnetic fields is located. In the underlying ganglion neurons respond with spike frequency modulations to modulations in intensity and frequency of an extremely low intensity, ELF ambient magnetic field. Calculating the effects of the external magnetic field modulations on the magnetite structures strong amplification is found, which though- decays at short distances from the magnetite. Due to the small magnetite-dendrite interfaces electromotive forces and currents are small. Due to the decay of the amplified magnetic fields with distance a 105 difference of magnetic and electric fields exists across the dendritic membrane. Further amplification of electric potentials can be obtained by using the receptor outputs as inputs to integrating neurons, whose outputs can be distributed to neurons of an ANN. The neural network classifies the ambient magnetic field modulations (building a map of the magnetic environment within the nervous system). After learning, the network becomes sensitive to extremely small intensity and low frequency changes in the ambient geomagnetic field.
Early diagnosis of Downs disease relied until now on statistical analysis which results in a consistent number of uncertain cases. Using different neural networks the number of uncertain cases can be diminished. Four neural networks have been tested: The Multi-layer Perceptron, the Bayes network, the Radial Basis Function network and that of Kohonen. Best results were obtained with the Radial Basis Function. Si propone di utilizzare diverse reti in parallelo oltre allanalisi statistica per diminuire ulteriormente i falsi positivi e negativi.
Three thesis have been completed during the year 98/99:
- Girlanda, Andrea, Realizzazione di un filmato di grafica 3D sulla struttura e funzione nervosa dellocchio della Squilla mantis.
- Aiello, Anna Teresa, e Marangoni, Patrizia, Le reti neurali in campo diagnostico: la sindrome di Down.