Optical reservoir computing with tumor spheroids*

Photonics enables the implementation of many modern neural network architectures, like deep learning and random neural networks. When a multiple scattering medium is adopted for mixing optical signals and perform computations, one can realize optically different conventional applications of artificial intelligence, like image recognition or time-series prediction.
Optical reservoir computing may have other applications beyond computing.
We demonstrate a random optical network realized by tumor spheroids. The tumor spheroids are samples of human tumors, like glioblastoma. The characterization of tumor spheroids is commonly done by optical imaging, and it is time-consuming and resource-demanding.
We experimentally realize a random optical network that uses a tumor spheroid as a computing reservoir. We train the device by data from conventional imaging to infer tumor metabolism and the growth rate. After the training, the network can predict the time-dynamics of the metabolism and also the effect of hyperthermia and chemotherapy.[1]
Our bio-random optical network represents a new kind of smart device for cancer morphodynamics and biophysical applications.
[1] Deep Optical Neural Network by Living Tumour Brain Cells, arXiv:1812.09311