Correlation Of Radiomics Features from Thyroid Planar Scan with 99mTc Uptake– A Preliminary Study

Background: Use of radiomics features (RF) is being increasingly applied for the computer aided risk stratification of malignancy in thyroid nodules, using two- and three-dimensional imaging data. We explored the characteristic of radiomic features extracted from 99mTc thyroid planar scan (TTPS) images. Patients and methods: Anterior planar images from patients who underwent TTPS with quantification of percent 99mTc-uptake (%TU) were analyzed in LIFEX software to generate RF data from two regions of interest (ROI): right lobe and left lobe per patient. Correlation of the RF from multiple categories were checked against the TU of each lobe. After discovering RF with significant correlation with demographic and clinical variables, the RF in each category with smallest p-value (i.e. the best correlation) were shortlisted. Results: A total of 140 RF were extracted from 44 lobes in 22 consecutive patients. Among the RF, 21 of 32 morphological, 21 of 23 intensity based, 24 of 29 intensity histogram derived, 22 of 24 Gray-Level Co-occurrence Matrix (GLCM) based, 8 of 11 Gray-Level Run-Length Matrix (GLRLM) based, four of five Neighboring Gray Tone Difference Matrix (NGTDM) based and 10 of 16 Gray Level Size Zone (GLSZM) based, showed significant correlation with %TU (p < 0.05). The RF with lowest p-values in each category were, morphological integrated intensity (R = 0.89, p = 3.59e-16), intensity-based energy (R = 0.94, p = 5.48e-21), intensity histogram maximum histogram gradient (R = 0.9, p = 1.56e-16), GLCM angular second moment (R 0.92, p = 5.13e-19), GLRLM long runs emphasis (R = 0.95, p = 3.36e-23), NGTDM coarseness (R = -0.59, p = 2.59e-05), GLSZM large zone emphasis (R = 0.96, p = 8.06e-25), GLSZM large zone high grey level emphasis (R = 0.96, p = 6.46e-25), GLSZM zone size variance (R = 0.96, p = 6.8e-25). Conclusions: Correlations of RF with %TU indicate the possibility of finding RF as biomarkers for nodular and generalized parenchymal thyroid diseases from TTPS images as well as possible usefulness of these RF in the prediction of disease outcome. Bangladesh J. Nuclear Med. 27(2): 169-172, 2024