Artificial intelligence is not only changing several aspects of people's everyday life, but also the way of "doing science", inspiring new experiments and suggesting hitherto unexplored avenues of research.

Artificial intelligence systems are becoming increasingly futuristic, interdisciplinary and neuromorphic (i.e. similar to living systems), finding new application in a wide variety of fields, such as electronics, information technology, simulation and the different branches of medicine. The new models are developed to imitate the human brain, both in function, with very low energy consumption for learning, and in structure, using biological materials.

The team of researchers coordinated by Claudio Conti of the Department of Physics of Sapienza and Director of the Institute of Complex Systems of the CNR, in collaboration with Massimiliano Papi of the University Cattolica del Sacro Cuore of Rome, has created an optical neural network that incorporates living tumour cells that grow and multiply over time. It is a hybrid device, consisting of living tissue and physical parts, such as lenses, mirrors and traditional computers, which evolves over time and can be trained to provide information about cancer cells, their metabolism and the effect of chemotherapy and other treatments.

In the study, developed as part of the PRIN project "PELM: Photonic Extreme Learning Machine" and published on the journal Communications Physics, researchers used glioblastoma tumour cells, a very serious brain tumour, which were inserted into the optical device. Laser beams have been properly trained to pass through cancer cells, which behave like nodes in a neural network. At this point, the artificial intelligence system acts as a real biological neural network, stores and processes the data and then encodes the information contained in the light extracted from the cancer cells.

In addition, the living neural network can recognise external stimuli and react to changes: by adding a few doses of chemotherapy drugs, researchers have demonstrated the model's ability to calculate the effectiveness of therapy against glioblastoma.

The neural network, suitably trained, shows changes in the tumour that cannot be detected by traditional methods, such as microscopy or physicochemical techniques, and also provides new information on the dynamics of temporal evolution and the effects of temperature, previously only obtainable through cuts or invasive modifications to tumour samples. The potential of this technique lies in the important application effects in the field of new technologies used in the treatment of cancer and in particular in nanomedicine.

"This is an original and innovative application of the new concepts of Deep Learning to physics", says Claudio Conti. The idea is that we can use these mathematical models not only to do simple operations such as image recognition but also to do unconventional experiments that take advantage of physics and biophysics with an interdisciplinary approach."

References:

Living optical random neural network with three dimensional tumor spheroids for cancer morphodynamics − D.Pierangeli, V.Palmieri, G.Marcucci, C.Moriconi, G.Perini, M.DeSpirito, M.Papi, C.Conti − Communications Physics, 2020. DOI: https://doi.org/10.1038/s42005-020-00428-9

Further Information

Claudio Conti
Department of Physics
claudio.conti@uniroma1.it