Revisiting 3D Context Modeling with Supervised Pre-training for Universal Lesion Detection in CT Slices

Zhang, Shu; Xu, Jincheng; Chen, Yu‐Chun; Ma, Jiechao; Li, Zihao; Wang, Yizhou; Yu, Yizhou

doi:10.1007/978-3-030-59719-1_53

Cited by 20 publications

(12 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this study we demonstrated the reproducibility and generalizability of the model by testing through internal and external validation. To the best of our knowledge, there have been some AI algorithms developed that are capable of identifying multiple abnormalities, [22][23][24] but validations in real-world practice have been scarce, and this study is the first one to develop universal lesion detection model with internal and external validation.…”

Section: Discussionmentioning

confidence: 99%

<strong>External and Internal Validation of a Computer Assisted Diagnostic Model for Detecting Multi-Organ Mass Lesions in CT images</strong>

Yan²,

Lü³

et al. 2019

Chinese Medical Sciences Journal

View full text Add to dashboard Cite

Keywords：lesion detection;| Computer-Aided Diagnosis CAD; convolutional neural network; deep learning Objective：We developed a universal lesion detector (ULDor) which showed good performance in inlab experiments. The study aims to evaluate the performance and its ability to generalize in clinical setting via both external and internal validation.Methods：The ULDor system consists of a convolutional neural network (CNN) trained on around 80K lesion annotations from about 12K CT studies in the DeepLesion dataset and 5 other public organspecific datasets. During the validation process, the test sets include two parts: the external validation datasets were comprised of 164 sets of non-contrasted chest-upper abdomen CT scans collected from a comprehensive hospital, and the internal validation datasets are comprised of 187 sets of low-dose helical CT scans from the National Lung Screening Trial (NLST). We run the model on the two test sets to output lesion detection. Three board-certified radiologists read the CT scans and verify the detection results by ULDor. We used positive predictive value (PPV) and sensitivity to measure the performance of the model in detecting space-occupying lesions at all extra-pulmonary organs, including the liver, kidney, pancreas, thyroid, lymph nodes, body wall, and thoracic spine at visualized CT images.Results：In the external validation, the lesion-level PPV and sensitivity of the model are 58% and 66%, respectively. On average, the model detected 2.1 findings per set, and among them, 0.9 were false positives. ULDor worked well for detecting liver lesions, with a PPV of 79% and a sensitivity of 93%, followed by kidney, with a PPV of 70% and a sensitivity of 58%. In internal validation with NLST test set, ULDor obtained a PPV of 75% and a sensitivity of 52% despite the relatively high noise level of soft tissue in CT images.

show abstract

Section: Discussionmentioning

confidence: 99%

<strong>External and Internal Validation of a Computer Assisted Diagnostic Model for Detecting Multi-Organ Mass Lesions in CT images</strong>

Yan²,

Lü³

et al. 2019

Chinese Medical Sciences Journal

View full text Add to dashboard Cite

show abstract

“…Implementation Details: All the compared methods take 9 consecutive slices as input, which can be represented as a gray-scale 3D tensor of 1 × 9 × 512 × 512. Since only 2D annotations (2D bounding boxes on the key slice) are provided in the DeepLesion dataset, we use a similar architecture as MP3D [32] for lesion detection. MP3D neglects all the downsampling operations for the z-axis to keep the depth of all backbone outputs as 9.…”

Section: E Universal Lesion Detectionmentioning

confidence: 99%

“…Anchor scales of FPN are set to (16,32,64,128,256) to improve detection performance for small lesions. We apply multi-scale training with scales randomly sampled from (384, 448, 512, 576, 640).…”

Section: E Universal Lesion Detectionmentioning

confidence: 99%

“…The idea of pre-training a 3D network with 2D natural images was initially proposed as a trick in our conference paper [32], the objective of which is to construct 3D models for universal lesion detection. This work significantly expands on the preceding conference paper in the following aspects: Firstly, we revisit the mixed usage of RGB images and adjacent slices of 3D volume images in medical problems to shed lights on the root of our proposed idea, and further re-establish the idea as the variable dimension transform based supervised pre-training technique.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Advancing 3D Medical Image Analysis with Variable Dimension Transform based Supervised 3D Pre-training

Zhang¹,

Li²,

Zhou³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

The difficulties in both data acquisition and annotation substantially restrict the sample sizes of training datasets for 3D medical imaging applications. As a result, constructing high-performance 3D convolutional neural networks from scratch remains a difficult task in the absence of a sufficient pre-training parameter. Previous efforts on 3D pre-training have frequently relied on selfsupervised approaches, which use either predictive or contrastive learning on unlabeled data to build invariant 3D representations. However, because of the unavailability of large-scale supervision information, obtaining semantically invariant and discriminative representations from these learning frameworks remains problematic. In this paper, we revisit an innovative yet simple fully-supervised 3D network pre-training framework to take advantage of semantic supervisions from large-scale 2D natural image datasets. With a redesigned 3D network architecture, reformulated natural images are used to address the problem of data scarcity and develop powerful 3D representations. Comprehensive experiments on four benchmark datasets demonstrate that the proposed pre-trained models can effectively accelerate convergence while also improving accuracy for a variety of 3D medical imaging tasks such as classification, segmentation and detection. In addition, as compared to training from scratch, it can save up to 60% of annotation efforts. On the NIH DeepLesion dataset, it likewise achieves state-ofthe-art detection performance, outperforming earlier selfsupervised and fully-supervised pre-training approaches, as well as methods that do training from scratch. To facilitate further development of 3D medical models, our code and pre-trained model weights are publicly available at https://github.com/urmagicsmine/CSPR.

show abstract

“…Universal Lesion Detection (ULD) in computed tomography (CT) [1][2][3][4][5][6][7][8][9][10][11][12][13], which aims to localize different types of lesions instead of identifying lesion types [14][15][16][17][18][19][20][21][22][23][24][25], plays an essential role in computer-aided diagnosis (CAD). ULD is a challenging task because different lesions have very diverse shapes and sizes, easily leading to false positive and false negative detections.…”

Section: Introductionmentioning

confidence: 99%

Bounding Maps for Universal Lesion Detection

Han

Zhou

2020

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Universal Lesion Detection (ULD) in computed tomography plays an essential role in computer-aided diagnosis. Promising ULD results have been reported by coarse-to-fine two-stage detection approaches, but such two-stage ULD methods still suffer from issues like imbalance of positive v.s. negative anchors during object proposal and insufficient supervision problem during localization regression and classification of the region of interest (RoI) proposals. While leveraging pseudo segmentation masks such as bounding map (BM) can reduce the above issues to some degree, it is still an open problem to effectively handle the diverse lesion shapes and sizes in ULD. In this paper we propose a BM-based conditional training for two-stage ULD, which can (i) reduce positive vs. negative anchor imbalance via a BM-based conditioning (BMC) mechanism for anchor sampling instead of traditional IoU-based rule; and (ii) adaptively compute size-adaptive BM (ABM) from lesion boundingbox, which is used for improving lesion localization accuracy via ABMsupervised segmentation. Experiments with four state-of-the-art methods show that the proposed approach can bring an almost free detection accuracy improvement without requiring expensive lesion mask annotations.

show abstract

Revisiting 3D Context Modeling with Supervised Pre-training for Universal Lesion Detection in CT Slices

Cited by 20 publications

References 16 publications

<strong>External and Internal Validation of a Computer Assisted Diagnostic Model for Detecting Multi-Organ Mass Lesions in CT images</strong>

<strong>External and Internal Validation of a Computer Assisted Diagnostic Model for Detecting Multi-Organ Mass Lesions in CT images</strong>

Advancing 3D Medical Image Analysis with Variable Dimension Transform based Supervised 3D Pre-training

Bounding Maps for Universal Lesion Detection

Contact Info

Product

Resources

About