The value of MRI texture analysis based on gray level co-occurrence matrix to predict cervical lymph node metastasis in tongue carcinoma patients / 中华放射学杂志

ABSTRACT

Objective To evaluate the value of MRI texture analysis based on gray level co-occurrence matrix to predict cervical lymph node metastasis in patients with tongue carcinoma. Methods A total of 70 patients with tongue carcinoma confirmed by pathology were analyzed retrospectively. The patients were divided into cervical lymph node (LN) metastasis group (unilateral LN+, n=18;bilateral LN+,n=22) and no cervical lymph node metastasis(LN-,n=30). T1W, T2W and contrast-enhanced T1W images of the largest section of tumor were selected. ROI of the lesion was manually drew and GLCM texture parameters (energy, contrast, correlation, inverse difference and entropy) were extracted. The tumor length, thickness and para-lingual distance between tumor and tongue midline were also measured.Differences of all parameters were compared between LN+ group and LN- group, unilateral and bilateral cervical lymph node metastasis group, the parameters with statistically significant difference in predicting the efficiency of cervical lymph node metastasis were analyzed. The diagnostic efficiency of lymph node metastasis was calculated. Results The correlation, inverse difference and entropy based on T2WI showed significant difference (Zcor elation=2.97, tinverse dif erence=5.14, tentropy=2.41; P<0.05), entropy showed the best diagnostic efficiency, the area under the ROC curve (AUC) was 0.90, the cut off value was 7.19, the sensitivity was 80.0％, specificity was 86.7％. Only entropy showed significant difference between unilateral LN+group and bilateral LN+group (P<0.05), the AUC was 0.82, the cut off value was 7.47, the sensitivity was 90.9％, specificity was 78.8％. The index of tumor length, thickness and para-lingual distance between tumor and tongue midline all showed significant difference between LN+group and LN-group (P<0.05), the thickness showed the best diagnostic efficiency, the AUC value was 0.81, the cut off value was 11.19, the sensitivity was 78.0％, specificity was 81.7％. The index of tumor length, thickness and para-lingual distance between tumor and tongue midline showed no significant difference between unilateral LN+ group and bilateral LN+ group (P>0.05). The diagnostic sensitivity of radiologists was 65.0％ (26/40), the specificity was 80.0％ (24/32) on cervical lymph node metastasis. Conclusion Texture analysis based on T2WI can predict cervical lymph node metastasis in patients with tongue carcinoma. Entropy has certain value in predicting bilateral cervical lymph node metastasis.