SESSION TYPE: Endobronchial Ultrasound
PRESENTED ON: Monday, October 22, 2012 at 04:00 PM - 05:30 PM
PURPOSE: The aim of this study is to develop an objective image analysis system for prediction of lymph node metastasis in patients with lung cancer during endobronchial ultrasound-guided transbronchial needle aspiration (EBUS-TBNA).
METHODS: Ultrasound spectrums of lymph nodes during EBUS-TBNA were retrospectively analyzed using an ultrasound scanner (EU-ME1, Olympus, Tokyo, Japan) with software capable of recording the raw data. A linear regression of frequency spectrum and three ultrasonic spectral parameters were calculated: 1) Midband-fit (dB, value of the regression line at the center frequency), 2) Slope (dB/MHz), 3) Intercept (dB, extrapolation to zero frequency). These parameters were calculated by changing the area and creating spectrum parameter images. A mean value of these parameters within lymph nodes were computed and the cut-off values for each parameter to distinguish metastatic and benign lymph nodes were decided based on receiver operating characteristic of the training set. These cut-off values were applied to the testing set for validation.
RESULTS: 362 lymph nodes (112 metastatic and 250 benign) were analyzed as a training set and 284 lymph nodes (73 metastatic and 211 benign) were evaluated as a testing set. The spectrum images showed different characters from the B-mode images. In the training set, all 3 parameters showed significant difference between metastatic and benign lymph nodes (p<0.001). The metastatic lymph nodes tended to show low midband-fit, high slope, and low intercept. When we combine midband-fit and intercept, the diagnostic yield was maximized (sensitivity 83.9%, specificity 74.4%, positive predictive value (PPV) 59.5%, negative predictive value (NPV) 91.2%). The diagnostic yield of this cut-off value for testing set were sensitivity 78.1%, specificity 75.4%, PPV 52.3%, NPV 90.9%, respectively.
CONCLUSIONS: Spectrum analysis of EBUS images showed high negative predictive values with fair sensitivity in this cohort.
CLINICAL IMPLICATIONS: Automated real-time diagnostic image guidance for EBUS-TBNA may reduce unnecessary biopsies and shorten examination time.
DISCLOSURE: Kazuhiro Yasufuku: Grant monies (from industry related sources): Grants from Olympus Medical Systems for continuing medical education and research
The following authors have nothing to disclose: Takahiro Nakajima, Masato Shingyoji, Takashi Anayama, Hideki Kimura, Ichiro Yoshino
We used dedicated software capable of recording the raw data of endobronchial ultrasound.Division of Thoracic Surgery, Toronto General Hospital, University Health Network, Toronto, ON, Canada