A deep Residual U-Net convolutional neural network for automated lung segmentation in computed tomography images

Khanna, Anita; Londhe, Narendra D.; Gupta, Saurabh; Semwal, Ashish

doi:10.1016/j.bbe.2020.07.007

Cited by 98 publications

(64 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In [ 41 ], an algorithm based on random forest, deep convolutional network, and multi-scale super-pixels was proposed for segmenting lungs with interstitial lung disease (IDL) using the ILDs database [ 42 ] with an average DSC of 96.45%. Khanna et al [ 43 ] implemented the residual U-Net with a false-positive removal algorithm using a training set of 173 images from three publicly available benchmark datasets, namely LUNA, VESSEL12, and HUG-ILD. Specifically, they implemented a U-Net with residual block, to overcome the problem of performance degradation, and various data augmentation techniques to improve the generalization capability of the method, obtaining a DSC > 98.63% using the five-fold cross-validation technique.…”

Section: Discussionmentioning

confidence: 99%

Lung Segmentation on High-Resolution Computerized Tomography Images Using Deep Learning: A Preliminary Step for Radiomics Studies

et al. 2020

View full text Add to dashboard Cite

Background: The aim of this work is to identify an automatic, accurate, and fast deep learning segmentation approach, applied to the parenchyma, using a very small dataset of high-resolution computed tomography images of patients with idiopathic pulmonary fibrosis. In this way, we aim to enhance the methodology performed by healthcare operators in radiomics studies where operator-independent segmentation methods must be used to correctly identify the target and, consequently, the texture-based prediction model. Methods: Two deep learning models were investigated: (i) U-Net, already used in many biomedical image segmentation tasks, and (ii) E-Net, used for image segmentation tasks in self-driving cars, where hardware availability is limited and accurate segmentation is critical for user safety. Our small image dataset is composed of 42 studies of patients with idiopathic pulmonary fibrosis, of which only 32 were used for the training phase. We compared the performance of the two models in terms of the similarity of their segmentation outcome with the gold standard and in terms of their resources’ requirements. Results: E-Net can be used to obtain accurate (dice similarity coefficient = 95.90%), fast (20.32 s), and clinically acceptable segmentation of the lung region. Conclusions: We demonstrated that deep learning models can be efficiently applied to rapidly segment and quantify the parenchyma of patients with pulmonary fibrosis, without any radiologist supervision, in order to produce user-independent results.

show abstract

Section: Discussionmentioning

confidence: 99%

Lung Segmentation on High-Resolution Computerized Tomography Images Using Deep Learning: A Preliminary Step for Radiomics Studies

et al. 2020

View full text Add to dashboard Cite

show abstract

“…Finally, the best model was trained and validated five times to ensure that its performance was not due to random factors inherent in the training process such as the initialization of the weights. Using this procedure, the best model configuration identified used a batch size of 8, an equally weighted combined loss function of soft-Dice and categorical cross-entropy, batch normalization, and ReLU activations in the hidden layers [25, 26] (Fig. 3).…”

Section: Methodsmentioning

confidence: 99%

Segmentation of Tissues and Proliferating Cells in Light-Sheet Microscopy Images using Convolutional Neural Networks

Vercio¹,

Rm²,

Robertson³

et al. 2021

Preprint

View full text Add to dashboard Cite

Background and Objective: A variety of genetic mutations are known to affect cell proliferation and apoptosis during organism development, leading to structural birth defects such as facial clefting. Yet, the mechanisms how these alterations influence the development of the face remain unclear. Cell proliferation and its relation to shape variation can be studied in high detail using Light-Sheet Microscopy (LSM) imaging across a range of developmental time points. However, the large number of LSM images captured at cellular resolution precludes manual analysis. Thus, the aim of this work was to develop and evaluate automatic methods to segment tissues and proliferating cells in these images in an accurate and efficient way. Methods: We developed, trained, and evaluated convolutional neural networks (CNNs) for segmenting tissues, cells, and specifically proliferating cells in LSM datasets. We compared the automatically extracted tissue and cell annotations to corresponding manual segmentations for three specific applications: (i) tissue segmentation (neural ectoderm and mesenchyme) in nuclear-stained LSM images, (ii) cell segmentation in nuclear-stained LSM images, and (iii) segmentation of proliferating cells in Phospho-Histone H3 (PHH3)-stained LSM images. Results: The automatic CNN-based tissue segmentation method achieved a macro-average F-score of 0.84 compared to a macro-average F-score of 0.89 comparing corresponding manual segmentations from two observers. The automatic cell segmentation method in nuclear-stained LSM images achieved an F-score of 0.57, while comparing the manual segmentations resulted in an F-score of 0.39. Finally, the automatic segmentation method of proliferating cells in the PHH3-stained LSM datasets achieved an F-score of 0.56 for the automated method, while comparing the manual segmentations resulted in an F-score of 0.45. Conclusions: The proposed automatic CNN-based framework for tissue and cell segmentation leads to results comparable to the inter-observer agreement, accelerating the LSM image analysis. The trained CNN models can also be applied for shape or morphological analysis of embryos, and more generally in other areas of cell biology.

show abstract

“…Residual connections were originally introduced with the ResNet architecture [51] and were frequently incorporated into CNN architectures as ResNet models converge better thus enable deeper networks to perform considerably better than their shallow counterparts [31,51]. U-Net architectures with residual connections were designed, e.g., for road extraction from aerial remote sensing imagery [42], for urban land cover classification in aerial and satellite imagery [52,53], for sea-land segmentation [54,55], for tree species classification in airborne imagery [56], for semantic segmentation of ships in optical remote sensing imagery [57], and also for biomedical image segmentation [58] (also see [31]). The subsequent max pooling operation is performed by taking the maximum value over a 2 × 2 window in order to halve the image size before entering the next downsampling block.…”

Section: Post-processingmentioning

confidence: 99%

A Novel Method for Automated Supraglacial Lake Mapping in Antarctica Using Sentinel-1 SAR Imagery and Deep Learning

et al. 2021

View full text Add to dashboard Cite

Supraglacial meltwater accumulation on ice sheets can be a main driver for accelerated ice discharge, mass loss, and global sea-level-rise. With further increasing surface air temperatures, meltwater-induced hydrofracturing, basal sliding, or surface thinning will cumulate and most likely trigger unprecedented ice mass loss on the Greenland and Antarctic ice sheets. While the Greenland surface hydrological network as well as its impacts on ice dynamics and mass balance has been studied in much detail, Antarctic supraglacial lakes remain understudied with a circum-Antarctic record of their spatio-temporal development entirely lacking. This study provides the first automated supraglacial lake extent mapping method using Sentinel-1 synthetic aperture radar (SAR) imagery over Antarctica and complements the developed optical Sentinel-2 supraglacial lake detection algorithm presented in our companion paper. In detail, we propose the use of a modified U-Net for semantic segmentation of supraglacial lakes in single-polarized Sentinel-1 imagery. The convolutional neural network (CNN) is implemented with residual connections for optimized performance as well as an Atrous Spatial Pyramid Pooling (ASPP) module for multiscale feature extraction. The algorithm is trained on 21,200 Sentinel-1 image patches and evaluated in ten spatially or temporally independent test acquisitions. In addition, George VI Ice Shelf is analyzed for intra-annual lake dynamics throughout austral summer 2019/2020 and a decision-level fused Sentinel-1 and Sentinel-2 maximum lake extent mapping product is presented for January 2020 revealing a more complete supraglacial lake coverage (~770 km2) than the individual single-sensor products. Classification results confirm the reliability of the proposed workflow with an average Kappa coefficient of 0.925 and a F1-score of 93.0% for the supraglacial water class across all test regions. Furthermore, the algorithm is applied in an additional test region covering supraglacial lakes on the Greenland ice sheet which further highlights the potential for spatio-temporal transferability. Future work involves the integration of more training data as well as intra-annual analyses of supraglacial lake occurrence across the whole continent and with focus on supraglacial lake development throughout a summer melt season and into Antarctic winter.

show abstract

A deep Residual U-Net convolutional neural network for automated lung segmentation in computed tomography images

Cited by 98 publications

References 16 publications

Lung Segmentation on High-Resolution Computerized Tomography Images Using Deep Learning: A Preliminary Step for Radiomics Studies

Lung Segmentation on High-Resolution Computerized Tomography Images Using Deep Learning: A Preliminary Step for Radiomics Studies

Segmentation of Tissues and Proliferating Cells in Light-Sheet Microscopy Images using Convolutional Neural Networks

A Novel Method for Automated Supraglacial Lake Mapping in Antarctica Using Sentinel-1 SAR Imagery and Deep Learning

Contact Info

Product

Resources

About