AIM: Mediastinal lymph nodes staging in NSCLC (non small cell lung cancer) is of undisputable importance. Although relatively precise, diagnostic modalities, mediastinoscopy and EBUS/EUS - TBNA (endobronchial/endoscopic ultrasound guided--transbronchial needle aspiration) still employ certain level of invasiveness. Artificial Neural Network (ANN) is an established predictor tool which, due to underlying distribution and relationship among the given variables, allow for construction of multidimensional models trained in prognosis of given outcome. Their performance in mediastinal staging based on radiological data only, remains limited to single studies. METHODS: We obtained 467 groups of lymph nodes from 160 patients with primary NSCLC by means of EBUS--TBNA, mediastinoscopy or lymphadenectomy during thoracotomy and analyzed them microscopically. ANN models were created and prospectively validated on unmatched cohort of 50 consecutive patients (158 groups of lymph nodes). To identify factors correlated with nodal involvement single factor tests and logistic regression analyzes were performed. Additionally, logistic regression analysis allowed for construction of scoring model with certain parameters corresponding to risk thresholds of metastatic disease. RESULTS: Size and standard uptake value (SUV) of the node along with primary tumour T characteristics were identified as the most sensitive variables regardless of the analysis conducted. Two ANN models predicted metastatic involvement with 89% and 92% accuracy. Single factor tests maintained high accuracy only for 2 out of 4 most sensitive variables (SUV >2.8 and length >15mm) in prospective validation. CONCLUSIONS: ANN is a repeatable and accurate diagnostic tool in mediastinal staging in NSCLC patients. Before its role in clinical practice will be established in large multi--centre study, findings of this preliminary report should be considered as exploratory only.
AIM: Mediastinal lymph nodes staging in NSCLC (non small cell lung cancer) is of undisputable importance. Although relatively precise, diagnostic modalities, mediastinoscopy and EBUS/EUS - TBNA (endobronchial/endoscopic ultrasound guided--transbronchial needle aspiration) still employ certain level of invasiveness. Artificial Neural Network (ANN) is an established predictor tool which, due to underlying distribution and relationship among the given variables, allow for construction of multidimensional models trained in prognosis of given outcome. Their performance in mediastinal staging based on radiological data only, remains limited to single studies. METHODS: We obtained 467 groups of lymph nodes from 160 patients with primary NSCLC by means of EBUS--TBNA, mediastinoscopy or lymphadenectomy during thoracotomy and analyzed them microscopically. ANN models were created and prospectively validated on unmatched cohort of 50 consecutive patients (158 groups of lymph nodes). To identify factors correlated with nodal involvement single factor tests and logistic regression analyzes were performed. Additionally, logistic regression analysis allowed for construction of scoring model with certain parameters corresponding to risk thresholds of metastatic disease. RESULTS: Size and standard uptake value (SUV) of the node along with primary tumour T characteristics were identified as the most sensitive variables regardless of the analysis conducted. Two ANN models predicted metastatic involvement with 89% and 92% accuracy. Single factor tests maintained high accuracy only for 2 out of 4 most sensitive variables (SUV >2.8 and length >15mm) in prospective validation. CONCLUSIONS: ANN is a repeatable and accurate diagnostic tool in mediastinal staging in NSCLCpatients. Before its role in clinical practice will be established in large multi--centre study, findings of this preliminary report should be considered as exploratory only.
Authors: Daniel P Steinfort; Shankar Siva; Tracy L Leong; Morgan Rose; Dishan Herath; Phillip Antippa; David L Ball; Louis B Irving Journal: Medicine (Baltimore) Date: 2016-02 Impact factor: 1.889