This paper focuses on the development of mathematical support and software for recognizing thyroid neoplasms in ultrasound images to improve diagnostic efficiency. An analytical review of the subject area, research, and solutions in the field of pattern recognition on ultrasound images of the thyroid gland has been carried out. The paper presents mathematical support including binary classification and multiclass classification models. The developed algorithmic software includes algorithms for processing thyroid ultrasound images to provide convenient and accurate analysis based on YOLOv8 and ResNet50 architectures. Performance analysis results are presented, which showed high accuracy of the models. In the binary classification task, the maP50 was 0.56. Thus, the model correctly classified 56% of positive examples. The mAP50-95 was 0.3. Thus, the model works well with an interval between 50 and 95% classification probability. For the multiclass classification task, an accuracy metric value of 0.70 was obtained. This indicates that the model correctly classified about 70% of the samples out of the total number of samples. The analysis of the efficiency of the developed software product implementation showed an average 70% de-crease in the time of diagnostics of thyroid neoplasms.