Online Monitoring of Microcarrier Based Fibroblast Cultivations with In-situ Microscopy

Taking offline samples is time-consuming and risks a contamination of the culture during sampling. Culture inline monitoring and online data analysis helps to reduce manual sampling to a minimum and thus reduces the risk of contaminations. In this context in-situ microscopy has proven to be a reliable analysis method for different process parameters in cultivations of yeasts and mammalian cell lines. This has already been described in literature. The in-situ microscope is a sterilizable and noninvasive system equipped with a CCD-camera for image acquisition. A connected PC contains control and analysis software. The microscope is mounted in a 25 mm bioreactor side-port. During the whole cultivation process the microscope s sampling zone is directly immersed in the culture broth. To achieve high density cultures of anchorage-dependent cells the use of microcarriers is a common procedure. Adherent cells need surfaces to anchor before cell growth and proliferation is possible. After settlement cell proliferation continues until the available microcarrier surface is completely covered by a cell monolayer. For fibroblasts process parameters like plating efficiency and the average degree of population on the available microcarrier surface are most important for managing the process and moreover for the determination of harvest i.e. the initiation of following steps. This work presents the microscopic observation of a 5 L microcarrier based fibroblast cultivation which was subsequently inoculated with viruses for the virus production. The parameters degree of population as well as the plating efficiency were analysed with an image processing algorithm that uses a neural network. The first step of the algorithm is the application of a Sobel filter for edge detection in order to be able to separate the microcarriers from the background. After separation the grey value histograms of the microcarrier surfaces are calculated. Characteristic values (mean, variance, maximum of the grey value distribution etc.) are calculated from these histograms and are used as input values to a neural network which has been trained to predict the degree of population from these values. Since the connection between histogram values and the degree of population is unknown, the implementation of a neural network is advantageous. The neural net was trained before with a dataset of several microcarrier images which had been evaluated manually. Besides the changes in cell morphology, especially after virus inoculation, were observed in order to develop a neural network based algorithm which can distinguish between infected and non-infected cells