Practical use of S-Detect Thyroid artificial intelligence-based program for automatic detection and characterization of thyroid nodules

In order to assess the practical efficacy of the S-Detect Thyroid artificial intelligence-based program for automatic detection and analysis of thyroid lesions, the prospective assessment of 84 focal thyroid lesions was carried out. he risk of malignancy was stratified according to the EU-TIRADS at the same time. A fine-needle aspiration biopsy was performed for all detected nodules of EU-TIRADS 3–5 category and a diameter ≥10 mm. Cytological examination was performed using the Bethesda classification. According to the cytology data, all patients were divided into two groups: 73 patients with benign thyroid nodules (Bethesda II) and 11 patients with malignant nodules (Bethesda V). Patients with “uncertain” Bethesda categories “I”, “III”, and “IV” were excluded from the study. The results of the study showed that the use of the S-Detect program based on artificial intelligence allows for differential diagnostics of benign (Bethesda II) and malignant (Bethesda V) thyroid nodules with a sensitivity of 90.9%, specificity of 94.5%, positive and negative predictive value of 71.4% and 98.6%, accuracy of 94%, and AUC 0.941. In our opinion, the best results of all program settings show the S-Detect “High Accuracy” mode, which we recommend for practical use. In some cases, there was disagreement between the S-Detect and the doctor's opinion in characterizing the nodule structure and echogenicity, as well as in determining the presence of macro- and microcalcifications. In our opinion, the use of the S-Detect benign/malignant criteria as indications for needle aspiration biopsy would avoid obviously unnecessary diagnostic interventions in some patients with thyroid nodules classified as EU TIRADS 3–5. However, the S-Detect artificial intelligence program cannot currently fully replace the doctor's intellect, erudition, and experience.