Central focused convolutional neural networks: Developing a data-driven model for lung nodule segmentation

Abstract

Accurate lung nodule segmentation from computed tomography (CT) images is of great importance for image-driven lung cancer analysis. However, the heterogeneity of lung nodules and the presence of similar visual characteristics between nodules and their surroundings make it difficult for robust nodule segmentation. In this study, we propose a data-driven model, termed the Central Focused Convolutional Neural Networks (CF-CNN), to segment lung nodules from heterogeneous CT images. Our approach combines two key insights: 1) the proposed model captures a diverse set of nodule-sensitive features from both 3-D and 2-D CT images simultaneously; 2) when classifying an image voxel, the effects of its neighbor voxels can vary according to their spatial locations. We describe this phenomenon by proposing a novel central pooling layer retaining much information on voxel patch center, followed by a multi-scale patch learning strategy. Moreover, we design a weighted sampling to facilitate the model training, where training samples are selected according to their degree of segmentation difficulty. The proposed method has been extensively evaluated on the public LIDC dataset including 893 nodules and an independent dataset with 74 nodules from Guangdong General Hospital (GDGH). We showed that CF-CNN achieved superior segmentation performance with average dice scores of 82.15% and 80.02% for the two datasets respectively. Moreover, we compared our results with the inter-radiologists consistency on LIDC dataset, showing a difference in average dice score of only 1.98%.

Automated delineation of non‐small cell lung cancer: A step toward quantitative reasoning in medical decision science

“…They could literally have very small attachments to the wall as shown by subject 035, which is simply denoted as pleural‐tail nodules by physicians, or could also be severely adhered to more than one surrounding structure, as shown by subjects 008 and 224. Farag et al and Wang et al have demonstrated successful delineation of pleural‐tail nodules with level set (93% accuracy) and convolutional neural network algorithms (92.7% sensitivity), respectively. Our proposed model demonstrated sensitivity, specificity, and accuracy ranging between 88% and 99%.…”

“…Wang et al demonstrated that energy-based models (level set, graph-cut) and machine learning-based models (U-Net, CF-CNN) have efficiently segmented solitary masses, with sensitivity ranging from 64%-92%. 48 Zheng et al investigated four classes of segmentation algorithms, namely, intensitybased (Otsu), region-based (region growing, watershed transform), edge-based (Canny detector) and energybased (active contour); and have qualitatively shown that all methods yield between good to excellent results; except for the watershed transform, which had the highest false negative rate 49 . We present two examples from our datasets, as depicted in Figures 3A and 3B, for the solitary mass segmentation from subjects 024 and 051.…”

High‐performance dynamic magnetic resonance image reconstruction and synthesis employing deep feature learning convolutional networks

“…The concept for deep learning algorithms is first lower level features is learned and higherlevel complex are presented to classify. 22 Moreover, the shallow models, deep learning engineering can encode complex data from easy to complex learning. Subsequently, for image enlistment, deep learning is promising in light of the fact that it (a) it does not require prior knowledge as it is an unsupervised model, (b) utilizes a layered design to derive complex nonlinear connections, (c) is totally data driven and not depends on considered features, and (d) can rapidly and productively process the presented testing input data.…”

Cardiac or Infectious? Transfer Learning with Chest X-Rays for ER Patient Classification

“…The CheXNet [14] model was able to accurately identify 14 categories of abnormalities in chest X-ray images. Deep learning techniques have shown promise for automated detection and diagnosis of lung cancer [19][20][21][22], breast cancer [23,24], skin cancer [25][26][27], and other diseases. Most of these approaches use deep neural networks [28] especially convolutional neural networks [29,30].…”