Investigation of Transfer Learning on Pulmonary Nodule Characteristics


KAYA A., KEÇELİ A. S. , CAN A. B.

25th Signal Processing and Communications Applications Conference (SIU), Antalya, Turkey, 15 - 18 May 2017 identifier identifier

Abstract

Studies on the classification of small pulmonary nodules generally focus on the prediction of malignancy of the nodule. In the recent years, publicly available databases provided different types of data to researchers, such as nodule characteristics, apart from the lung image and malignancy degree. In this paper, a study on the classification of pulmonary nodule characteristics using conventional features and deep features obtained from transfer learning method has been proposed. The results were assessed by sensitivity, specificity, and classification accuracy. The results of the study can be used to form multi-level classifiers in predicting malignancy by combining different types of features.