Diagnostic approach to mediastinal pathologies with echoic features of endobronchial ultrasound images

BACKGROUND AND OBJECTIVE: Endobronchial ultrasound (EBUS) is a minimally invasive technique with high diagnostic yield used in the investigation of mediastinal lymph nodes. Recent data suggest that sonographic feature analysis on endobronchial ultrasound (EBUS) imaging can differentiate benign from malignant lymphadenopathy. However, ultrasound is a user dependent technique. The aim of this study was to investigate the diagnostic yield of the sonographic features, investigation of internal and external validation of the sonographic features by making intraobserver- interobserver assessments and to examine the usefulness of numerical datas that allows more objective evaluation of these images. METHODS: Our study was conducted at Karadeniz Technical University, Department of Chest Diseases prospectively between September 2014 and May 2015. We recorded ultrasound images of patients who underwent EBUS and pathology results were followed. Images were interpreted by a radiologist and a clinician who were blind to the pathology results. Then, the image data mixed and one month later was interpreted by observers again. Each data were evaluated by the same observer twice and a total of four times. Additionally, these images were analyzed by an image processing and analysis program. Using this program, histogram analysis and numerical values were obtained. The obtained values were loaded into SPSS and statistical analyzes were performed. RESULTS: 119 patients were included in the study (M/F= 77: 42). The mean age was 58.1 ± 13.3 (23-85). Pathology results in 50.4% (60) were malignant and in 49.6% (59) were benign. In the subtypes; 44.5% (53) were lung cancer, 31.1% (37) were sarcoidosis, 16.8% (20) were benign cytology, 4.2% (5) were other metastases, 2.5% were tuberculosis and 0.8% (1) was lymphoma. Granular sand structure from the sonographic features was found to be statistically significant in determining of benign mediastinal lymph nodes (p=0,028) (81.4% sensitivity and 36.7% specificity; PPV 55.8%, NPV 66.7%). Interobserver agreement was the best in the echogenicity and central hilar structures criteria. Intraobserver agreement scores were significantly higher in observer-2 (radiologist) assessment for shape, border, echogenicity, central hilar structures, coagulation necrosis and granular sand structure. Median histogram mean, median, standard deviation values were statistically significant compared to other numerical data (p=0.037; p=0.048; p=0.018). ROC (Receiver operating characteristic) analysis in distinguishing malignant from benign lymph nodes other than the sarcoidosis and tuberculosis, median numerical value of histogram mean, median and standard deviation was significantly higher (p=0,005; p=0.008; p=0.001). ROC analysis of the quantitative histogram mean, median, standard deviation values found significant in the distinction between malignant lymph nodes and the other benign lymph nodes. There was a statistically significant difference between benign and malignant lymph nodes for quantitative values of minimum gray scale histogram (p=0,035). When subgroup analysis performed in patients between lung cancer and sarcoidosis groups the quantitative values of minimum gray scale, skewness and kurtosis was determined statistically significant (p= 0.029, p=0.024, p=0.22). CONCLUSIONS: This prospective study showed that; the sanding granular structure can be a potential sonographic feature in determining benign lymph nodes. For evaluation of the ultrasound images, assessment of the radiologist was more consistent when compared with clinicians. Because of reader-dependency of sonographic features, quantitative measurement and numerical values is required. Histogram mean, median and standard deviation values and the minimum gray scale value was more meaningful than the other quantitative data and statistically significant to distinguish benign from malignant lesions. As a result; a scoring system can be generated using sonographic features of ultrasound images, histogram data and PET CT findings together. This may be helpful in staging false negative lymph node particularly in malignant staging.