PhD fellowship available on
Analysis and interpretation of 4D microscopy data
Recent developments in high throughput optical microscopy have demonstrated the possibility, in some animal models, of functional imaging of the whole brain with cellular resolution. In particular, the zebra fish larva allows optical access throughout the brain thanks to its natural transparency. LENS has recently developed a light-sheet microscope with extended depth of field, which allows high spatial resolution 3D imaging, with a volumetric frame rate of more than 10 Hz. The ability to track brain activity with single-cell resolution opens new perspectives for the understanding of the mechanisms underlying the functioning of the brain. However, in order to truly advance knowledge in this area, developments in optical imaging need to be complemented by innovations in the field of data analysis.
The project has several converging goals. First, it is necessary to develop a software platform capable of extracting time series of activities from the single neurons in the 4D datasets produced by the apparatus operating at LENS. Previous collaborations between the LENS microscopy group and the AI laboratory of DINFO have already produced efficient and accurate systems for the localization of cell nuclei in 3D images. On the other hand, the 4D data extension presents significant challenges because the stability of the 3D coordinates of the nuclei is not guaranteed and, at the same time, it is necessary to filter localization errors and to correlate 3D predictions over time. To this end, the study of specialized machine learning techniques is envisaged, which for example exploit hybrid architectures between convolutional and recurrent neural networks. Longer-term goals include the inference of the zebra fish functional connectivity graph (connectome) from the extracted time series and the characterization of the connectivity variability among different subjects, in order to determine the possible existence of topological invariants between different brains. This information will be used later to design targeted neuronal stimulation experiments in zebra fish larvae, with the aim of validating the obtained functional connectivity and developing new neuronal control strategies.
The ideal candidate has an engineering or scientific background, and good programming skills. Previous knowledge of machine learning is not necessary, whereas the ability and willingness to work in an interdisciplinary environment, covering areas from biology to optics and computer science, are required.
The position will be supervised in Florence by Prof. Paolo Frasconi (DINFO, email@example.com) and Dr. Ludovico Silvestri (LENS, firstname.lastname@example.org). Students interested in the fellowship are invited to get in touch with one of the supervisors as soon as possible. Application deadline will be mid-July 2018.