Two prerequisites are necessary for the development of a web service for the simulation of airflow in the upper airways. On the one hand, a 3D model of the nasal cavities or parts of them must be created in a short time and on the other hand, a procedure for flow simulation must be used that delivers meaningful results with a time and economic justifiable effort. Manual methods for the creation of a high-quality 3D model of all nasal cavities can take several hours up to one working day. Within the Rhinodiagnost project, a method has been developed that uses DICOM images (axial sections) to determine the area of the human skull that corresponds to the air space enclosed by the nasal cavities. An artificial neural network (CNN: Convolutional Neural Network) has been developed for this purpose. For the deep learning process, it was sufficient to generate training data from the images of three patients. Including the post-processing (e.g. removal of the Eustachian tube) and the subsequent creation of a surface network for the 3D model, it now takes about 5 minutes for processing on a conventional PC. This model is the starting point for CFD (Computational Fluid Dynamics) and can also be used for other processes such as 3D printing of nasal cavities. Besides the two-class segmentation (air - non-air), the project started to develop an artificial neural network which aims at a multi-class segmentation in order to detect anatomical structures (e.g. a maxillary sinus) automatically. This should enable automatic volume measurements to determine the degree of swelling in a paranasal sinus (with the exception of ethmoidal cells). Further neural networks for detecting the access to the sinuses (ostia) or the position of the nasal septum (septum) are in preparation.