Aim of study. Development of a promising approach for determining longitudinal flatfoot based on a neural network, which effectively reduces the time spent by a radiologist without loss of accuracy. Materials and methods. 3458 radiographs of feet of patients with longitudinal flatfoot and 1726 cases of subjects without longitudinal flatfoot at the age of 17—75 years were used. Each x-ray for training the neural network was tagged by one radiologist, and at the testing stage, each x-ray image was tagged independently by two radiologists selected blindly. The diagnostic algorithm was developed basing on the identification of three anatomical points forming the angle of foot arch. The proposed approach consists of three stages: 1) preliminary processing and preparation of data for segmentation using a neural network; 2) segmentation of three areas — the bounding box around the three desired points; 3) determining the location of each of the required points within the corresponding area and calculating the appropriate measure of angle and degree of flat feet. The segmentation network is a convolutional neural network (CNN) of the encoder-decoder type based on the U-Net architecture using the ResNet50 architecture as an encoder Results. An effective, reliable and fast method of artificial intellect has been created, the accuracy of which as a whole is not inferior to radiologists, but requires about 6000 times less time. Conclusions. The developed artificial intellect is an effective tool for determining longitudinal flatfoot by segmenting an x-ray image and calculating the arch angle of the foot. It can be seen as a quick assistant, as accurate as an experienced radiologist