Author(s)

Dingwen Wang

Corresponding Author

Dingwen Wang

ABSTRACT

A pulmonary nodule is a large amount of tissue located in the lungs. Identifying whether it has lung CT medical imaging is of great significance for the discovery of underlying diseases, particularly lung cancer or other risk factors. In real life, radiologists or doctors are prone to misjudgment in the diagnosis of lung CT images due to fatigue and other factors, and traditional machine learning and identification medical images have the problems of time-consuming and labor-intensive manual design features and incomplete feature design. In recent years, with the great success of deep learning technology in the fields of image, voice, video, etc., the author proposes whether the CT image of the lung exists based on the Convolutional Neural Networks (CNN) combined with the LIDC-IDRI data set. The lung nodules are classified. The experimental results show that the classifier reaches 93.25% on the verification set, and the accuracy rate reaches 85.28% on the test set of 1622 pictures, the recall rate is 80.14%, the specificity is 97.46%, and the AUC value under ROC is 0.97 ( The AUC value is between 0-1 and the larger the classifier is, the better the effect is). It achieves the purpose of automatically and accurately identifying lung nodules in lung CT images.

KEYWORDS

Deep learning, Convolutional Neural Network (CNN), pulmonary nodules, lung CT images, LIDC-IDRI data set, lung cancer