Convolutional neural network for real-time localisation and classification of in situ microscope-images of animal cells cultured in bioreactor

The set of labelled images constituting the database was obtained in a context where there is a need to increase the quantity and quality of biopharmaceutical products. These molecules are produced by growing cells in bioreactors. During the production process, a significant number of cells may follow an undesirable death path, directly reducing process productivity and product quality. The state of the art shows that there is no technique for in-depth, real-time analysis of the cell population in culture within a bioreactor, despite the interest this could offer in terms of bioprocess control.

This work presents an automated real-time classification system based on the combination of two techniques, in situ microscopy (ISM) and artificial intelligence (AI). This system enables precise analysis of the cell population based on cell morphological characteristics. As the system provides early information on the state of cells, it is of great interest for the management of biopharmaceutical processes. For example, the evolution of different classes of animal cells, including viable, necrotic and apoptotic cells, can be monitored in real time. A new class of viable cells with bulges has also been identified in the course of this work. This class refers to cells that are sensitive to substrate concentration conditions. Considered as potential breakthrough in the catalog of Process Analytical Tools (PAT), AI-enhanced ISM is therefore also of potential interest for research.

This dataset was prepared for the training phase of the neural network used to localize and classify the cellular objects present in the images taken in real time by the ISM.