Actuality: Malignant tumors of organs of head and neck account for 20—25% of all oncological pathologies in the Russian Federation. The frequency of detection of malignant tumors of oral cavity was 27.2 per 100 thousand of population in 2017. The peak incidence rate in men and women is observed at the age of 59.7—63.9 years, which includes the labour-potential population. Over the past few years, there has been a steady increase in the number of patients with this pathology. Despite significant progress in the treatment of cancer patients as a whole, the 5-year relapse-free survival of patients in this group has not changed significantly over the past 20 years and amounts to about 45—50%. Aim of study. Development of algorithm and software package for automatic detection of structures possibly having the malignant neoplasms in the lower jaw region by the method of analysis and segmentation of images obtained by computed tomography using deep learning technologies. Material and methods. To classify image fragments, a classification convolutional neural network with U-net architecture is used. For the purposes of training and network monitoring, CT of 22 patients with tumors of the lower half of the facial region, consisting of 383 marked sections 512×512 pixels in size, was used. The marking was performed by qualified maxillofacial surgeons and a team of radiologists. The probability map generated as a result of the classification was processed in order to obtain a binary mask of the pathological site. The network was evaluated in a control sample using the IoU metric. The resulting masks had IoU coefficients ranging from 0.6 to 0.76. Results. In the control sample (50 images), the algorithm showed result of getting into the neoplasm zone in 98% of cases, with an average contour accuracy of 0.68 according to the IoU metric. The final result of algorithm testing is presented in the pictures. For comparison, the results of manual marking of CT images, which were used for the «control», and accordingly were not shown to CNN before, are also presented. Conclusions. As a result of the study, a software package based on a trained artificial neural network was developed which is capable of performing high-precision automatic analysis and segmentation of images of the lower jaw obtained by computed tomography, identifying malignant tumors in this area and the boundaries of their distribution. In clinical practice, the developed software can be used to automatically search for objects possibly having malignant tumor, followed by an in-depth study of this area by a specialist, i.e. use as a decision support system for a surgeon and/or doctor of radiation diagnostics in order to save working time for evaluating computed tomography data.