[IEEE 2019 Conference on Lasers and Electro-Optics Europe & European Quantum Electronics Conference (CLEO/Europe-EQEC) - Munich, Germany (2019.6.23-2019.6.27)] 2019 Conference on Lasers and Electro-Optics Europe & European Quantum Electronics Conference (CLEO/Europe-EQEC) - Optical Neural Network by Disordered Tumor Spheroids

摘要： Optical neuromorphic computing processes information at the speed of light, but requires a careful design and fabrication of the deep layers, which strongly hampers the development of large-scale photonic learning machines [1,2]. New paradigms, as reservoir computing [3], suggest that brain-inspired complex systems such as disordered and biological materials may realize artificial neural networks with thousands of computational nodes trained only at the input and at the readout. Here we use real-brain cells for realizing a bio-inspired optical neural network able to extract information about cancer morphodynamics and chemotherapy that is inaccessible to imaging methods [4]. Specifically, we consider glioblastoma tumor spheroids as three-dimensional deep computational reservoirs with thousands of cells acting as wave-mixing nodes for the input light beam. These tumor models are largely used in oncology and are a promising platform for studying complex cell-to-cell interactions and anti-cancer therapeutics. In our hybrid bio/photonic scheme, the tumor model cellular layers are the diffractive deep layers of the optical neural network [Fig. 1(a)]. By exploiting structured light propagation in the disordered assembly [5], we show that the random neural network is a universal optical interpolant able to perform programmed functions in the transmission plane. Through external stimuli on the tumour brain cells – either of thermal or chemical nature – we control the internal weights of the living reservoir and its functionality. Once trained, the response of the living optical neural network follows subcellular cancer morphodynamics, not detected by more invasive and destructive optical imaging. In Fig. 1(b) we demonstrate morphodynamics sensing by inducing hyperthermia with an infrared pump laser. Moreover, we track cellular processes in the tumour model beyond the simple unconstrained growth; in Fig. 1(c) the network output allows to quantify the effect of chemotherapy inhibiting tumour growth. In this case, we realize a non-invasive smart probe for cytotoxicity assay, which is at least one order of magnitude more sensitive with respect to conventional imaging. Our random and hybrid photonic/living system is a novel artificial machine for computing and for the real-time investigation of tumour dynamics.