MRI Texture Analysis Reflects Histopathology Parameters in Thyroid Cancer – A First Preliminary Study

Abstract

OBJECT: Thyroid cancer represents the most frequent malignancy of the endocrine system with an increasing incidence worldwide. Novel imaging techniques are able to further characterize tumors and even predict histopathology features. Texture analysis is an emergent imaging technique to extract extensive data from an radiology images. The present study was therefore conducted to identify possible associations between texture analysis and histopathology parameters in thyroid cancer. METHODS: The radiological database was retrospectively reviewed for thyroid carcinoma. Overall, 13 patients (3 females, 23.1%) with a mean age of 61.6 years were identified. The MaZda program was used for texture analysis. The T1-precontrast and T2-weighted images were analyzed and overall 279 texture feature for each sequence was investigated. For every patient cell count, Ki67-index and p53 count were investigated. RESULTS: Several significant correlations between texture features and histopathology were identified. Regarding T1-weighted images, S(0;1)Sum Averg correlated the most with cell count (r = 0.82). An inverse correlations with S(5;0)AngScMom, S(5;0)DifVarnc S(5;0), DiffEntrp and GrNonZeros (r = −0.69, −0.66, −0.69 and −0.63, respectively) was also identified. For T2-weighted images, Variance with r = 0.63 was the highest coefficient, WavEnLL_S3 correlated inversely with cell count (r = −0.57). WavEnLL_S2 derived from T1-weighted images was the highest coefficient r = −0.80, S(0;5)SumVarnc was positively with r = 0.74. Regarding T2-weighted images WavEnHL_s-1 was inverse correlated with Ki67 index (r = −0.77). S(1;0)Correlat was with r = 0.75 the best correlation with Ki67 index. For T1-weighed images S(5;0)SumofSqs was the best with r = 0.65 with p53 count. For T2-weighted images S(1;−1)SumEntrp was the inverse correlation with r = −0.72, whereas S(0;4)AngScMom correlated positively with r = 0.63. CONCLUSIONS: MRI texture analysis derived from conventional sequences reflects histopathology features in thyroid cancer. This technique might be a novel noninvasive modality to further characterize thyroid cancer in clinical oncology.

Exploring Applications of Radiomics in Magnetic Resonance Imaging of Head and Neck Cancer: A Systematic Review

“…The upper limit reflects the choice of one study to maintain their initially derived feature set, which was not reduced in dimensionality. Meyer et al ( 41 ) generated 279 features from T1-weighted and T2-weighted images corresponding to the following categories: gray-level co-occurrence matrix (GLCM), gray-level histogram, gray-level run-length matrix, gray-level absolute gradient, auto-regressive model, and wavelet transform. They then compared the derived T1- or T2-weighted features to cellular density, presence of Ki-67 antigen, or p53 index histopathology in 12 thyroid cancer patients.…”

“…While most included studies detailed selection of extracted radiomic features, Meyer et al ( 41 ) did not reduce their initially derived feature set. Direct and inverse correlations between specified features and classification parameters were discovered, but this presents a challenge to rationalize statistically.…”

Personalized CT‐based radiomics nomogram preoperative predicting Ki‐67 expression in gastrointestinal stromal tumors: a multicenter development and validation cohort

“…Association between radiomic features and Ki‐67 expression in various tumors is well documented [32–38]. Most studies focused on MRI texture features to predict the Ki‐67 expression in thyroid cancer [37], breast cancer [36], liver cancer [34], and glioma [35] with AUC higher than 0.75. It is noteworthy that only two studies have focused on the CE‐CT texture features for the prediction of Ki‐67 expression in lung cancer [32,38].…”

Intravoxel incoherent motion and ADC measurements for differentiating benign from malignant thyroid nodules: utilizing the most repeatable region of interest delineation at 3.0 T

“…Recently, IVIM applications in thyroid and related diseases have been increasingly reported [10][11][12]. Texture analyses of thyroid nodules based on magnetic resonance imaging have also been reported [13]. Based on large-sample and multi-centre research, artificial intelligence and machine learning could gradually be applied to magnetic resonance imaging of thyroid nodules.…”