The use of computer vision to determine reference points on the human back is a relatively new approach in medicine. The relevance of this study is due not only to the need to develop new methods and approaches in determining postural disorders, but also to the growing interest in the development and application of artificial intelligence in medicine. This study is devoted to the creation and training of a neural network to find reference points of the back when diagnosing postural disorders. In this study, we use a set of ready-made neural network tools that are freely available. Training was carried out on models of photographed volunteers to determine the reference points of the back. The results of the study indicate that, as part of the study, a search was made for alternative (reference points) that determine the geometry of the back using developed neural network algorithms. It was shown that the accuracy of point detection by a neural network trained on more than 2400 grouped photographs reaches 85%, which indicates a good determination of object boundaries and their classification. The use of computer vision to identify reference points for assessing postural abnormalities can bring significant benefits to medical practice. This tool may be more accurate and efficient than traditional diagnostic methods, as well as more accessible and convenient for the patient. In addition, the use of neural networks can speed up the diagnostic process and reduce research costs.