We present new pulmonary nodule segmentation algorithms for computed\u00a0tomography (CT). These include a fully--automated (FA) system, a\u00a0semi-automated (SA) system, and a hybrid system. Like most traditional\u00a0systems, the new FA system requires only a single user-supplied cue\u00a0point. On the other hand, the SA system represents a new algorithm class\u00a0requiring 8 user-supplied control points. This does increase the burden on\u00a0the user, but we show that the resulting system is highly robust and can\u00a0handle a variety of challenging cases. The proposed hybrid system starts\u00a0with the FA system. If improved segmentation results are needed, the SA\u00a0system is then deployed.<\/p>
The FA segmentation engine has 2 free\u00a0parameters, and the SA system has 3. These parameters are adaptively\u00a0determined for each nodule in a search process guided by a regression\u00a0neural network (RNN). The RNN uses a number of features computed for each\u00a0candidate segmentation. We train and test our systems using the new\u00a0Lung\u00a0Image Database Consortium and Image Database Resource Initiative\u00a0(LIDC--IDRI) data. To the best of our knowledge, this is one of the first\u00a0nodule-specific performance benchmarks using the new LIDC--IDRI dataset.\u00a0We also compare the performance of the proposed methods with several\u00a0previously reported results on the same data used by those other methods.\u00a0Our results suggest that the proposed FA system improves upon\u00a0the state-of-the-art, and the SA system offers a considerable boost over\u00a0the FA system.<\/span><\/p> The download links provided below provide easy access to specific subsets of images from our study, which are described in much greater detail in our publication (https:\/\/doi.org\/10.1016\/j.media.2015.02.002<\/a>).\u00a0\u00a0<\/span><\/p>