End-to-End Supervised Lung Lobe Segmentation

Abstract

The segmentation and characterization of the lung lobes are important tasks for Computer Aided Diagnosis (CAD) systems related to pulmonary disease. The detection of the fissures that divide the lung lobes is non-trivial when using classical methods that rely on anatomical information like the localization of the airways and vessels. This work presents a fully automatic and supervised approach to the problem of the segmentation of the five pulmonary lobes from a chest Computer Tomography (CT) scan using a Fully Regularized V-Net (FRV-Net), a 3D Fully Convolutional Neural Network trained end-toend. Our network was trained and tested in a custom dataset that we make publicly available. It can correctly separate the lobes even in cases when the fissure is not well delineated, achieving 0.93 in per-lobe Dice Coefficient and 0.85 in the inter-lobar Dice Coefficient in the test set. Both quantitative and qualitative results show that the proposed method can learn to produce correct lobe segmentations even when trained on a reduced dataset.

Automatic Pulmonary Lobe Segmentation Using Deep Learning

“…In this work, we release our manual annotation by radiologist for 50 CT scans collected from the LUNA16 challenge and present a practical and robust framework for robust pulmonary lobe segmentation. We believe the public availabil- Step-wise performance gains of using hybrid loss and pre-processing using convex hull as compared to a baseline model trained only with dice loss and previous state-ofart method [7]. RU, RM, RL, LU, LL and AVG represent the dice coefficient of right upper lobe, right middle lobe, right lower lobe, left upper lobe, left lower lobe and their average respectively.…”

“…More recently, FJS Bragman et al applied probabilistic model in enhanced fissure detection using fissure prior which yields accurate results under various fissure incompleteness. However, the attempt of using deep learning in this task is still rare [7] because of the need for a large number of annotated training examples. Moreover, publicly available annotations for pulmonary lobe segmentation can hardly be found for supervised training of deep neural network.…”

Relational Modeling for Robust and Efficient Pulmonary Lobe Segmentation in CT Scans

“…There are two learning objectives for each RU-Net: lobe segmentation and lobe border segmentation, inspired by [12], [15]. Therefore, the final loss function is a summation of four terms, and each is the generalized Dice loss [36].…”

“…Recent advances in convolution neural networks (CNN) provide a data-driven approach for more robust feature extraction in an end-to-end optimization process. Many works have successfully adopted CNNs in their lobe segmentation framework [12]- [15]. In [12], deep supervision was extensively used in the up-sampling path based on their V-Net design [16] along with the multi-tasking that segments lobe and lobe borders at the same time.…”

“…Many works have successfully adopted CNNs in their lobe segmentation framework [12]- [15]. In [12], deep supervision was extensively used in the up-sampling path based on their V-Net design [16] along with the multi-tasking that segments lobe and lobe borders at the same time. [13] uses a relatively deeper architecture based on Dense-Net [17] to ensure a sufficient receptive field of extracted features.…”