ABSTRACT
Objective:To investigate the value of conventional MRI and high resolution diffusion weighted imaging (DWI) for preoperative discrimination between nasopharyngeal-skull base osteomyelitis (NP-SBO) and locoregionally advanced nasopharyngeal carcinoma (LA-NPC).Methods:From January 2017 to October 2021, 27 patients of NP-SBO and 32 patients of LA-NPC were retrospectively analyzed at the Eye & ENT Hospital of Fudan University. The clinical characteristics and conventional MRI features were collected, and the apparent diffusion coefficient (ADC) values of polygonal (ADC polygonal) and small circle were measured from readout segmentation of long variable echo-trains (RESOLVE) DWI. MRI features included laterality, margin, signal intensity of T 1WI and T 2WI, enhancement degree, component, abscess, deep mucosal white line, bone invasion, lymph nodes involvement and other accompany symphtoms. The independent sample t test, χ 2 test or Fisher exact test were used to compare the features and ADC values of the NP-SBO and LA-NPC groups. The logistic regression was applied to select independent predictors in the distinguishing LA-NPC from NP-SBO. Then, the conventional MRI model, ADC model and conventional MRI in combination with ADC model were built. The area under the receiver operating characteristic curve (AUC) of models were compared using DeLong test. Results:The age, diabetic status, cranial nerve deficits, inner component, abscess, deep mucosal white line, lymph nodes involvement and ADC polygonal were significantly different between NP-SBO and LA-NPC groups ( P<0.05). The logistic regression analysis showed that ADC polygonal (OR=0.972, 95%CI 0.951-0.993, P=0.011) and abscess (OR=0.101, 95%CI 0.013-0.774, P=0.027) were the independent predictors in the discrimination of NP-SBO and LA-NPC. The AUC (95%CI) of conventional MRI model (abscess), ADC model (ADC polygonal) and combination model were 0.634 (0.499-0.756), 0.870 (0.757-0.943), and 0.925(0.829-0.979), respectively. The AUC of combination model was higher than that of conventional MRI model ( Z=4.77, P<0.001), while there was no difference between combination model and ADC model ( Z=1.87, P=0.062). The AUC of conventional MRI model was lower than that of ADC model ( Z=2.84, P=0.005). Conclusion:Conventional MRI in combination with RESOLVE DWI shows good performance in differentiating between NP-SBO and LA-NPC, especially for abscess in combination with ADC polygonal value.
ABSTRACT
Objective:To build and validate a radiomics and clinical nomogram for preoperative discrimination between low- and high-grade sinonasal squamous cell carcinoma (SNSCC).Methods:From January 2017 to May 2021, 167 SNSCC patients including 78 low-grade (grade Ⅰ or Ⅱ) and 89 high-grade (grade Ⅲ) were retrospectively analyzed at the Eye & ENT Hospital of Fudan University. All patients were randomly divided into a training cohort ( n=117, 64 high-grade and 53 low-grade SNSCC) and a validation cohort ( n=50, 25 high-grade and 25 low-grade SNSCC) in a ratio of 7∶3 using a stratified sampling method. The radiomics features were extracted in contrast enhanced T 1WI with manual segmentation of lesions. The least absolute shrinkage and selection operator regression was used to reduce the dimension of the radiomics features and then the radiomics model was built to predict SNSCC histological grade in training cohort. Independent clinical predicting factors were screened using logistic regression and the clinical model was built. The clinical-radiomics model was built by the radiomics features and clinical factors in the training cohort based on logistic regression and the nomogram was drawn. The receiver operator characteristic curves were drawn to evaluate the performance of clinical model, radiomics model and nomogram. The calibration curve was used to evaluate the consistency between the nomogram prediction and the actual observation risk, and the decision curve analysis (DCA) was used to evaluate the clinical applicability of the nomogram. Results:Using logistic regression analysis, the clinical model was built by the tumor primary site (OR value 7.376, 95%CI 2.517-21.618, P<0.001) and TNM stage (OR value 10.020, 95%CI 3.654-27.472, P<0.001) and the area under the curve (AUC) in the training cohort and validation cohort were 0.798 and 0.784, sensitivity were 84.4% and 84.0%, specificity were 58.5% and 68.0%, respectively. Based on the contrast enhanced T 1WI, a total of 9 radiomics features were screened for establishing the radiomics model. The AUC of radiomics model were 0.833 (sensitivity 82.8%, specificity 73.6%) and 0.851 (sensitivity 92.0%, specificity 68.0%) in the training and validation cohorts. The nomogram based on the clinical-radiomics model predicted histological grade with the highest AUC in the training cohort (AUC 0.920, sensitivity 89.1%, specificity 83.0%) and validation cohort (AUC 0.912, sensitivity 92.0%, specificity 84.0%). The calibration curve of the nomogram was close to the ideal line in both training and validation cohorts. DCA showed that the use of nomogram with a threshold in the range of <85% in training cohort, in the range of 20%-65%, 72%-90% in validation cohort, had a greater clinical application value in predicting the SNSCC histological grade. Nomogram model had a better clinical net benefit than the clinical and radiomics models. Conclusion:Nomogram combining clinical factors (tumor primary site and TNM stage) with radiomics features obtained from contrast enhanced T 1WI has a better ability for predicting histological grade of SNSCC than clinical and radiomics models.