Comparing to Learn: Surpassing ImageNet Pretraining on Radiographs by Comparing Image Representations

Zhou, Hongyu; Yu, Shuang; Bian, Cheng; Hu, Yifan; Ma, Kai; Zheng, Yefeng

doi:10.1007/978-3-030-59710-8_39

Cited by 74 publications

(67 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The network is initialized using the pre-trained weights of ImageNet dataset [46] and eliminated the fully connected layers. Many studies reported the pre-training model using the ImageNet can improve the ability of medical image classification tasks [47][48].…”

Section: ) Deep Feature Extractionmentioning

confidence: 99%

Genotype-Guided Radiomics Signatures for Recurrence Prediction of Non-Small Cell Lung Cancer

et al. 2021

View full text Add to dashboard Cite

Non-small cell lung cancer (NSCLC) is a serious disease and has a high recurrence rate after surgery. Recently, many machine learning methods have been proposed for recurrence prediction. The methods using gene expression data achieve high accuracy rates but expensive. While, the radiomics features using computer tomography (CT) image is a cost-effective method, but their accuracy is not competitive. In this paper, we propose a genotype-guided radiomics method (GGR) for obtaining high prediction accuracy at a low cost. We used a public radiogenomics dataset of NSCLC, which includes CT images and gene expression data. Our proposed method is two steps method that uses two models. The first model is a gene estimation model, which is used to estimate the gene expression from radiomics features and deep features extracted from CT images. The second model is used to predict the recurrence using the estimated gene. The proposed GGR method is designed based on hybrid features which is the fusion of handcrafted-and deep learning-based features. The experiments demonstrated that the prediction accuracy can be improved significantly from 78.61% (existing radiomics method) and 79.09% (ResNet50) to 83.28% by the proposed GGR.

show abstract

Section: ) Deep Feature Extractionmentioning

confidence: 99%

Genotype-Guided Radiomics Signatures for Recurrence Prediction of Non-Small Cell Lung Cancer

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Using EMA weights has been shown to help obtain more accurate predictions while stabilizing the training process [35,44]. Compared to CSD, we replace the siamese network with a student-teacher architecture to promote the detection performance.…”

Section: Investigation Of Ema Factormentioning

confidence: 99%

“…Consistency-based semi-supervised learning methods [14] mainly utilize self-supervision [46,5,44] and usually consist of two procedures. First, synthesize a pair of input images via some data augmentation strategies.…”

mentioning

confidence: 99%

SSMD: Semi-Supervised Medical Image Detection with Adaptive Consistency and Heterogeneous Perturbation

Wang¹,

Wang³

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

Semi-Supervised classification and segmentation methods have been widely investigated in medical image analysis. Both approaches can improve the performance of fully-supervised methods with additional unlabeled data. However, as a fundamental task, semi-supervised object detection has not gained enough attention in the field of medical image analysis. In this paper, we propose a novel Semi-Supervised Medical image Detector (SSMD). The motivation behind SSMD is to provide free yet effective supervision for unlabeled data, by regularizing the predictions at each position to be consistent. To achieve the above idea, we develop a novel adaptive consistency cost function to regularize different components in the predictions. Moreover, we introduce heterogeneous perturbation strategies that work in both feature space and image space, so that the proposed detector is promising to produce powerful image representations and robust predictions. Extensive experimental results show that the proposed SSMD achieves the state-of-the-art performance at a wide range of settings. We also demonstrate the strength of each proposed module with comprehensive ablation studies.

show abstract

“…Nonetheless, despite their effectiveness, unsupervised learning algorithms, including self-supervised learning, are frequently overshadowed by supervised learning algorithms. For example, as demonstrated in the most recent 3D pre-training studies [24], [42], even state-of-the-art self-supervised learning algorithms could not outperform ImageNet pre-trained models. This might be because, due to a lack of supervised signals, semantically discriminative representations are difficult to be mined from un/self-supervised learning algorithms.…”

Section: Introductionmentioning

confidence: 99%

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

Comparing to Learn: Surpassing ImageNet Pretraining on Radiographs by Comparing Image Representations

Cited by 74 publications

References 9 publications

Genotype-Guided Radiomics Signatures for Recurrence Prediction of Non-Small Cell Lung Cancer

Genotype-Guided Radiomics Signatures for Recurrence Prediction of Non-Small Cell Lung Cancer

SSMD: Semi-Supervised Medical Image Detection with Adaptive Consistency and Heterogeneous Perturbation

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

Contact Info

Product

Resources

About