Automatic identification of metastatic lymph nodes in OCT images

Abstract

Lymphatic metastasis is a main pathway of dissemination of malignancies. The diagnosis of metastasis in lymph nodes can help stage cancer or help the surgeons make intraoperative decisions. In addition, lymph nodes are more easily confused with other neck tissues during thyroid surgery. Therefore, identification of lymph nodes is very important. Up to now, the gold standard for identification of metastatic lymph nodes is still histological examination, which can only be performed ex vivo and needs a long time. Optical coherence tomography (OCT) is a non-invasive, high-resolution imaging technology that is capable of detecting microstructures in bio-tissues in real time. In this study, we demonstrated a method to identify metastatic lymph nodes automatically by intraoperative OCT imaging. With a home-made swept source OCT system, we obtained OCT images of different resected neck tissues, including lymph nodes with and without metastasis, thyroid, parathyroid, fat and muscle, from 28 patients undertaking thyroidectomy. The automatic identification algorithm was based on texture analysis and back-propagation artificial neural network (BP-ANN). 66 texture features of OCT images were extracted and 14 were selected and used for automatic identification experiments. The trained BP-ANN has an excellent performance in identifying OCT images of lymph nodes with the sensitivity of 98.9 % and specificity of 98.8 %. The accuracy of lymphatic metastasis diagnosis is 90.1 %.