This Article Full Text Full Text (PDF) All Versions of this Article: chest.07-1497v1 133/1/137    most recent Submit a response Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in ISI Web of Science Similar articles in PubMed Alert me to new issues of the journal Add to My Personal Article Archive Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Tagaya, R. Articles by Kobayashi, A. Search for Related Content PubMed PubMed Citation Articles by Tagaya, R. Articles by Kobayashi, A.

Automatic Objective Diagnosis of Lymph Nodal Disease by B-Mode Images From Convex-Type Echobronchoscopy^*

^* From the Department of Surgery (Drs. Tagaya, Kurimoto, and Osada), Division of Chest Surgery, St. Marianna University School of Medicine, Tokyo, Japan; and the Research Center for Charged Particle Therapy (Dr. Kobayashi), National Institute of Radiological Sciences, Chiba, Japan.

Correspondence to: Noriaki Kurimoto, PhD, 2-16-1 Sugao, Miyamae-ku, Kawasaki, Kanagawa 216-8511, Japan; e-mail: kurimoto{at}marianna-u.ac.jp

Abstract

Background: We investigated whether artificial neural networks (ANNs) could diagnose pathology of lymph nodes by feeding B-mode images from convex-type echobronchoscopy to ANNs.

Methods: Subjects comprised 91 patients who had undergone endobronchial ultrasonography transbronchial needle aspiration at our hospital between April 2005 and March 2007. Diagnosis was lymph node metastasis from lung cancer in 66 patients, and sarcoidosis in 25 patients. Layered ANNs consisting of input, middle layers, and output layers were prepared. Back-propagation was chosen as a learning algorithm. For the malignant findings, images obtained from six patients with lymph node metastasis of lung cancer (ie, adenocarcinoma, two patients; squamous cell carcinoma, two patients; small cell carcinoma, two patients) were used. As benign findings, typical images obtained from three patients with sarcoidosis were used. For each image used for supervised training, 5, 10, or 15 regions of interest were randomly selected. Repeated learning comprised either 500,000 or 1,000,000 repetitions. A total of five thoracic surgeons were asked to diagnose the pathology base on the same images. Accuracies were compared between ANNs and thoracic surgeons.

Results: The diagnostic accuracy of the surgeon with 5 years of experience and that of the surgeon with 1 year of experience were 78% and 51%, respectively, compared to 91% for the ANNs.

Conclusion: Assessment of B-mode images by ANNs may offer a useful basis for automatic diagnostic methods.

Key Words: artificial neural network • endobronchial ultrasonography • peribronchial lymph node • transbronchial needle aspiration