Automatic lung nodule detection in thoracic CT scans using dilated slice‐wise convolutions

Abstract

Purpose: Most state-of-the-art automated medical image analysis methods for volumetric data rely on adaptations of two-dimensional (2D) and three-dimensional (3D) convolutional neural networks (CNNs). In this paper, we develop a novel unified CNN-based model that combines the benefits of 2D and 3D networks for analyzing volumetric medical images. Methods: In our proposed framework, multiscale contextual information is first extracted from 2D slices inside a volume of interest (VOI). This is followed by dilated 1D convolutions across slices to aggregate in-plane features in a slice-wise manner and encode the information in the entire volume. Moreover, we formalize a curriculum learning strategy for a two-stage system (i.e., a system that consists of screening and false positive reduction), where the training samples are presented to the network in a meaningful order to further improve the performance. Results: We evaluated the proposed approach by developing a computer-aided detection (CADe) system for lung nodules. Our results on 888 CT exams demonstrate that the proposed approach can effectively analyze volumetric data by achieving a sensitivity of > 0.99 in the screening stage and a sensitivity of > 0.96 at eight false positives per case in the false positive reduction stage. Conclusion: Our experimental results show that the proposed method provides competitive results compared to state-of-the-art 3D frameworks. In addition, we illustrate the benefits of curriculum learning strategies in two-stage systems that are of common use in medical imaging applications.

Semi‐supervised training using cooperative labeling of weakly annotated data for nodule detection in chest CT

“…The production of nodule likelihood heatmaps as well as false-positive reduction are performed through the screening network and the false -positive reduction network. 14 The output of this filtering pipeline is fed into the network shown in Figure 4 for classification scans are fed through Lungmask 23 -a network for lung lobe segmentation -to produce lung lobe segmentations. In a separate iteration, the scans are fed through our pre-trained CADe system, producing heatmaps of nodule likelihood across the entire scan.…”

“…In the process of identifying nodule candidate locations, we employ the screening network proposed and evaluated by Farhangi et al 14 This network consists of consecutive 2D and dilated 1D convolutions. The 2D convolution operators extract in-plane features from a stack of slices within CT volumetric data.…”

“…After reducing candidate nodule locations with NMS, a false-positive reduction network further reduces the number of candidate volumes of interest (VOIs) to be extracted from the NLST dataset for training. For this purpose, we use the false -positive reduction network proposed and evaluated in Farhangi et al, 14 After filtering according to the re-weighted candidate scores, a set of positive crops containing nodule candidates at their centers is produced. In addition to these F I G U R E 3 Illustrations of the outputs of the first steps of the cooperative labeling pipeline.…”

“…In these studies, 3D architectures typically outperform 2D architectures as they better capture the 3D structure of the data, but this comes at the cost of greater network complexity and the need for greater computational resources. 13,14 The above mentioned studies employ fully supervised models and as such had to rely on fully annotated data.The process of fully annotating medical data is limited by labor, cost, and expertise involved in annotations. Fully annotating scans sometimes requires multiple clinician experts to annotate each scan in order to ensure a high-quality annotations.…”

Breast Cancer Detection on Histopathological Images Using a Composite Dilated Backbone Network

“…The median filter, which is a nonlinear filter, successfully retains cell edges in the images of breast cancer. The purpose of feature extraction is to enhance the overall classification and prediction performance [ 15 , 16 ]. The process involves producing prospective features using different transformation techniques.…”