Improving the Segmentation of Anatomical Structures in Chest Radiographs Using U-Net with an ImageNet Pre-trained Encoder

Frid-Adar, Maayan; Ben-Cohen, Avi; Amer, Rula; Greenspan, Hayit

doi:10.1007/978-3-030-00946-5_17

Cited by 65 publications

(43 citation statements)

References 19 publications

(47 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…SemiTC reached 87 ± 1.5% mIOU in <20% supervision mode (10 labeled images for training and 10 for validation). Table 2 compares our baseline network and the proposed SemiTC+ with the inter-observer agreement [16] and state-of-the-art chest X-ray segmentation methods [3,8,4]. All these methods are based on fully convolutional networks and are trained in a supervised way using at least 124(123) labeled images from the JSRT dataset.…”

Section: Resultsmentioning

confidence: 99%

“…All these methods are based on fully convolutional networks and are trained in a supervised way using at least 124(123) labeled images from the JSRT dataset. For these comparisons, we post-processed all predicted segmentations as described in [4] (small objects removal, hole filling).…”

Section: Resultsmentioning

confidence: 99%

“…as a regularization in supervised-only learning) to the segmentation performance; 2) the contribution of adding different amounts of unlabeled data into the equivariance learning; 3) how these contributions vary with the size of the labeled portion of the training set. We compare the proposed method trained in the small data (20 labeled images) and full supervision regimes with state-of-the-art methods [14,3,8,4] and the inter-observer agreement [16]. and t out 2 .…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Semi-supervised Medical Image Segmentation via Learning Consistency Under Transformations

Bortsova

Dubost

Hogeweg³

et al. 2019

Lecture Notes in Computer Science

156

113

View full text Add to dashboard Cite

The scarcity of labeled data often limits the application of supervised deep learning techniques for medical image segmentation. This has motivated the development of semi-supervised techniques that learn from a mixture of labeled and unlabeled images. In this paper, we propose a novel semi-supervised method that, in addition to supervised learning on labeled training images, learns to predict segmentations consistent under a given class of transformations on both labeled and unlabeled images. More specifically, in this work we explore learning equivariance to elastic deformations. We implement this through: 1) a Siamese architecture with two identical branches, each of which receives a differently transformed image, and 2) a composite loss function with a supervised segmentation loss term and an unsupervised term that encourages segmentation consistency between the predictions of the two branches. We evaluate the method on a public dataset of chest radiographs with segmentations of anatomical structures using 5-fold crossvalidation. The proposed method reaches significantly higher segmentation accuracy compared to supervised learning. This is due to learning transformation consistency on both labeled and unlabeled images, with the latter contributing the most. We achieve the performance comparable to state-of-the-art chest X-ray segmentation methods while using substantially fewer labeled images.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Resultsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Semi-supervised Medical Image Segmentation via Learning Consistency Under Transformations

Bortsova

Dubost

Hogeweg³

et al. 2019

Lecture Notes in Computer Science

156

113

View full text Add to dashboard Cite

show abstract

“…This is in line with recent studies demonstrating that when using a small dataset, transfer learning is recommended in comparison to full training, and that fine-tuned CNNs are more robust to the size of training. [29][30][31] Due to the high heterogeneity signal of OPGs and unclear borders, in this study we used fuzzy c-means clustering, also referred to as soft k-means clustering, for classification of tumor components. In non-fuzzy clustering, data are divided into hard clusters, where each data point belongs to exactly one cluster; in fuzzy clustering, the data points can belong to more than one cluster, while providing the strength of association for each one of the clusters.…”

Section: Discussionmentioning

confidence: 99%

Automatic segmentation, classification, and follow‐up of optic pathway gliomas using deep learning and fuzzy c‐means clustering based on MRI

et al. 2020

View full text Add to dashboard Cite

Purpose Optic pathway gliomas (OPG) are low‐grade pilocytic astrocytomas accounting for 3‐5% of pediatric intracranial tumors. Accurate and quantitative follow‐up of OPG using magnetic resonance imaging (MRI) is crucial for therapeutic decision making, yet is challenging due to the complex shape and heterogeneous tissue pattern which characterizes these tumors. The aim of this study was to implement automatic methods for segmentation and classification of OPG and its components, based on MRI. Methods A total of 202 MRI scans from 29 patients with chiasmatic OPG scanned longitudinally were retrospectively collected and included in this study. Data included T2 and post‐contrast T1 weighted images. The entire tumor volume and its components were manually annotated by a senior neuro‐radiologist, and inter‐ and intra‐rater variability of the entire tumor volume was assessed in a subset of scans. Automatic tumor segmentation was performed using deep‐learning method with U‐Net+ResNet architecture. A fivefold cross‐validation scheme was used to evaluate the automatic results relative to manual segmentation. Voxel‐based classification of the tumor into enhanced, non‐enhanced, and cystic components was performed using fuzzy c‐means clustering. Results The results of the automatic tumor segmentation were: mean dice score = 0.736 ± 0.025, precision = 0.918 ± 0.014, and recall = 0.635 ± 0.039 for the validation data, and dice score = 0.761 ± 0.011, precision = 0.794 ± 0.028, and recall = 0.742 ± 0.012 for the test data. The accuracy of the voxel‐based classification of tumor components was 0.94, with precision = 0.89, 0.97, and 0.85, and recall = 1.00, 0.79, and 0.94 for the non‐enhanced, enhanced, and cystic components, respectively. Conclusion This study presents methods for automatic segmentation of chiasmatic OPG tumors and classification into the different components of the tumor, based on conventional MRI. Automatic quantitative longitudinal assessment of these tumors may improve radiological monitoring, facilitate early detection of disease progression and optimize therapy management.

show abstract

“…Islam and Zhang [5] used convolutive networks to segment in the Montgomery and Shenzhen databases. Mayan et al [6] used the architecture U-Net with an ImageNet Pre-trained encoder.…”

Section: Introductionmentioning

confidence: 99%

Lung Region Segmentation in Chest X-Ray Images using Deep Convolutional Neural Networks

Portela

Pereira

Costa

et al. 2020

2020 42nd Annual International Conference of the IEEE Engineering in Medicine &Amp; Biology Society (EMBC)

View full text Add to dashboard Cite

Lung cancer is, by far, the leading cause of cancer death in the world. Tools for automated medical imaging analysis development of a Computer-Aided Diagnosis method comprises several tasks. In general, the first one is the segmentation of region of interest, for example, lung region segmentation from Chest X-ray imaging in the task of detecting lung cancer. Deep Convolutional Neural Networks (DCNN) have shown promising results in the task of segmentation in medical images. In this paper, to implement the lung region segmentation task on chest X-ray images, was evaluated three different DCNN architectures in association with different regularization (Dropout, L2, and Dropout + L2) and optimization methods (SGDM, RMSPROP and ADAM). All networks were applied in the Japanese Society of Radiological Technology (JSRT) database. The best results were obtained using Dropout + L2 as regularization method and ADAM as optimization method. Considering the Jaccard Coefficient obtained (0.97967 ± 0.00232) the proposal outperforms the state of the art. Clinical Relevance-The presented method reduces the time that a professional takes to perform lung segmentation, improving the effectiveness.

show abstract

Improving the Segmentation of Anatomical Structures in Chest Radiographs Using U-Net with an ImageNet Pre-trained Encoder

Cited by 65 publications

References 19 publications

Semi-supervised Medical Image Segmentation via Learning Consistency Under Transformations

Semi-supervised Medical Image Segmentation via Learning Consistency Under Transformations

Automatic segmentation, classification, and follow‐up of optic pathway gliomas using deep learning and fuzzy c‐means clustering based on MRI

Lung Region Segmentation in Chest X-Ray Images using Deep Convolutional Neural Networks

Contact Info

Product

Resources

About