A comparison study on the effect of false positive reduction in deep learning based detection for juxtapleural lung nodules: CNN vs DNN

Accurate detection of juxtapleural lung cancer, which are nodules on the chest wall, has great importance in the early detection of lung cancer. To acquire a good performance of Computer Aided Detection (CAD), both positive (nodule) detection and false positive reduction methods are needed. In this paper, we propose a two-phase deep learning based famework combining both nodule detection and false positive reduction. We applied Convolutional Neural Network to filter out negatives in the first phase. In the second phase, we applied two types of deep learning networks, Deep Neural Network (DNN) and Convolutional Neural Network (CNN), respectively, to reduce the false positive rates for the detected nodules. We used 70 patients from our dataset for training purpose and used another 15 patients for testing. Our CNN based classifier gives the sensitivity of 0.82 for lung juxtapleural detection. For false positive reduction, both CNN and DNN are competent in processing large amount of data with DNN reducing false positive rate from 0.45 to 0.329 and CNN reducing from 0.45 to 0.331.