Pulmonary nodules detection plays a significant role in the early detection and treatment of lung cancer. And false positive reduction is the one of the major parts of pulmonary nodules detection systems. In this study was provided a framwork that solves following problems: lungs segmentation, left and right lung separation, nodule candidates detection and false positive reduction. Novel methods was proposed aimed at lungs separation and recognizing real pulmonary nodule among a large group of candidates was proposed. The first algorithm via dilation propagation approach, described in this work, gains better performance in term of time complexity in comparison with the State of the Art methods. The latter algorithm consists of three steps: appropriate receptive field selection, feature extraction and a strategy for high level feature fusion and classification. Receptive field's objective is to fit tradeoff between covering 3D nature of nodules appearance and parameters amount. Deep residual 3D CNN acing over such receptive fields prior to fusion part provide an opportunity for spatial merging. Multi-scale information was handled by dimensionality reduction step as part of feature extraction network. The dataset consists of 888 patient's chest volume low dose computer tomography (LDCT) scans, selected from publicly available LIDC-IDRI dataset. This dataset was marked by LUNA16 challenge organizers resulting in 1186 nodules. Trivial data augmentation and dropout were applied in order to avoid overfitting. Proposed method achieved high competition performance metric (CPM) of 0.735 and sensitivities of 78.8% and 83.9% at 1 and 4 false positives per scan, respectively. This study is also accompanied by detailed descriptions and results overview in comparison with the state of the art solutions.
- Numpy, Scipy, Scikit-Image
- SimpleITK, PyDICOM
- Keras, TensorFlow >= 0.10, GPU supported
- Writing tests
- Code review
- Other guidelines