A New 2.5D Representation for Lymph Node Detection using Random Sets of Deep Convolutional Neural Network Observations

By Holger R. Roth, Le Lu, Ari Seff, Kevin M. Cherry, Joanne Hoffman, Shijun Wang, Jiamin Liu, Evrim B. Turkbey, Ronald M. Summers. International Conference on Medical Image Computing and Computer-Assisted Intervention, 2014

580 citations recorded by the source corpus.

Automated Lymph Node (LN) detection is an important clinical diagnostic task but very challenging due to the low contrast of surrounding structures in Computed Tomography (CT) and to their varying sizes, poses, shapes and sparsely distributed locations. State-of-the-art studies show the performance range of 52.9% sensitivity at 3.1 false-positives per volume (FP/vol.), or 60.9% at 6.1 FP/vol. for mediastinal LN, by one-shot boosting on 3D HAAR features. In this paper, we first operate a preliminary candidate generation stage, towards ~100% sensitivity at the cost of high FP levels (~40 per patient), to harvest volumes of interest (VOI). Our 2.5D approach consequently decomposes any 3D VOI by resampling 2D reformatted orthogonal views N times, via scale, random translations, and rotations with respect to the VOI centroid coordinates. These random views are then used to train a deep Convolutional Neural Network (CNN) classifier. In testing, the CNN is employed to assign LN probabilities for all N random views that can be simply averaged (as a set) to compute the final classification probability per VOI. We validate the approach on two datasets: 90 CT volumes with 388 mediastinal LNs and 86 patients with 595 abdominal LNs. We achieve sensitivities of 70%/83% at 3 FP/vol. and 84%/90% at 6 FP/vol. in mediastinum and abdomen respectively, which drastically improves over the previous state-of-the-art work.

Subfields: Artificial Intelligence, Pulmonary and Respiratory Medicine, Radiation

View Paper: https://www.semanticscholar.org/paper/d3e729ab1c87777a481246130746f84634af021f

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. P. Heng
2. Qi-Yan Dou
3. Jing Qin
4. Hao Chen
5. Lequan Yu
6. Xiaosong Wang
7. Le Lu
8. B. van Ginneken

In Pulmonary and Respiratory Medicine

1. B. van Ginneken
2. Ronald M. Summers
3. Kenji Suzuki
4. Xiaosong Wang
5. P. Heng
6. Lequan Yu
7. Hao Chen
8. Qi-Yan Dou

In Radiation

1. Kenji Suzuki
2. Xiaosong Wang
3. Ronald M. Summers

Citations in the corpus

• DeepOrgan: Multi-level Deep Convolutional Networks for Automated Pancreas Segmentation (2015, unknown): “Studies that applied ConvNets to medical imaging applications also show good promise on detection tasks [8,9].”
• ChestX-Ray8: Hospital-Scale Chest X-Ray Database and Benchmarks on Weakly-Supervised Classification and Localization of Common Thorax Diseases (2017, background): “not limit to) an overview review on the future promise of deep learning [14] and a collection of important medical applications on lymph node and interstitial lung disease detection and classification [37,43]; cerebral microbleed detection [11]; pulmonary nodule detection in CT images [40]; automated”
• 3D deeply supervised network for automated segmentation of volumetric medical images (2017, background): “…they are conceivably not optimal for vol- metric medical image analysis as they cannot take full advanage of the special information encoded in the volumetric data. he 2.5D methods ( Roth et al., 2014 ) have been proposed to inorporate richer spatial information but were still limited to 2D ernels.”
• Multi-crop Convolutional Neural Networks for lung nodule malignancy suspiciousness classification (2017, methodology): “We augmented nodules by random image translation, rotation and fl ip operations as in [26,41,42].”
• DeepNAT: Deep Convolutional Neural Network for Segmenting Neuroanatomy (2017, background): “In medical applications, 2.5D techniques have been proposed (Prasoon et al., 2013; Roth et al., 2014).”
• Improving Computer-aided Detection using Convolutional Neural Networks and Random View Aggregation (2015, methodology): “([20]), effectively circumvents the “curse-of-dimensionality” or “over-fitting” issue in the data-demanding ConvNet training procedures.”
• Deep learning for automated skeletal bone age assessment in X‐ray images (2017, methodology): “Similar methods were proposed in Sahiner et al. (1996); Lo et al.”
• Multi-scale Convolutional Neural Networks for Lung Nodule Classification (2015, methodology): “To enlarge the training samples to train the MCNN, we augmented both benign nodules and malignant nodules by translating the nodule patches along three axes with ±2 pixels as in [16].”
• Convolutional Neural Networks for Medical Image Analysis: Full Training or Fine Tuning? (2016, unknown): “lymph nodes in CT images [20], and computer-aided anatomy”
• Deep Learning Techniques for Medical Image Segmentation: Achievements and Challenges (2019, background): “5D approaches [54, 60, 65] are inspired by the fact that 2.”
• Volumetric ConvNets with Mixed Residual Connections for Automated Prostate Segmentation from 3D MR Images (2017, methodology): “2015), orthogonal planes (Prasoon et al. 2013; Roth et al. 2014) or multi-view planes (Setio et al.”
• Deep learning in radiology: an overview of the concepts and a survey of the state of the art with focus on MRI (2018, methodology): “The requirement for this phase is high sensitivity [87] and therefore it usually produces many false positives.”
• Longitudinal multiple sclerosis lesion segmentation: Resource and challenge (2017, unknown): “MV-CNN is a method based on a Longitudinal Multi-View CNN (Roth et al., 2014).”
• Hough-CNN: Deep learning for segmentation of deep brain regions in MRI and ultrasound (2016, methodology): “Dataset augmentation techniques, such as random patch rotation and mirroring, were also applied when the analysed objects were invariant to these transformations [22], [27], [28], [29].”