Radiomics in Breast Cancer: In-Depth Machine Analysis of MR Images of Metastatic Spine Lesion

Using mathematic criteria for image processing (radiomics) makes it possible to more accurately assess the nature of therapy-associated changes and determine the sites of maximal response. Comparison of the acquired quantitative and clinical data may assist radiologists in making the optimal decision. The aim of the study was to assess the capabilities of software operators for an in-depth analysis of metastatic spine lesion images in breast cancer. MATERIALS AND METHODS: MRI data of three patients with breast cancer T(2)N(2–3)M(1) receiving treatment in accordance with the accepted clinical protocols were used in our work. Spinal metastases were assessed by a radiologist and machine analysis using the Arzela variation operators. Twelve MRI examinations (4 per each patient) excluding the baseline examination have been analyzed with a follow-up period of about 3 months. RESULTS: The structure of the metastatically modified spine was analysed segment by segment in the sagittal and axial projections using machine image analysis operators. Rapid changes in the “complexity” of vertebrae images have been found, allowing one to suggest the efficacy of treatment in one of the three options — stabilization, improvement, progression. Changes in the vertebrae structure with a positive response to the treatment in the form of the formation of bone objects, calderas, reduction of the contrast agent circulation at the microlevel, confirmed by mathematical analysis, have been monitored. A correlation was obtained between the established changes and the level of the CA 15-3 cancer marker. CONCLUSION: The study has shown a high effectiveness of machine image analysis algorithms, high correlation of the obtained results with the radiologist’s report and clinical and laboratory data in 9 cases out of 12. The Pearson correlation coefficient between the classical marker and matrix filter curve was 0.8.