Search

The human imperceptible adversarial examples crafted by l 0 -norm attacks, which aims to minimize l 0 distance from the original image, thereby misleading deep neural network classifiers into the wrong classification. Prior works of tackling l 0 attacks can neither eliminate perturbed pixels nor improve the performance of the classifier in the recovered low-quality images. To address the issue, we propose a novel method, called space transformation pixel defender (STPD), to transform any image into a latent space to separate the perturbed pixels from the normal pixels. In particular, this strategy uses a set of one-class classifiers, including Isolation Forest and Elliptic Envelope, to locate the perturbed pixels from adversarial examples. The value of the neighboring normal pixels is then used to replace the perturbed pixels, which hold more than half of the votes from these one-class classifiers. We use our proposed strategy to successfully defend against well-known l 0 -norm adversarial examples in the image classification settings. We show experimental results under the One-pixel Attack (OPA), the Jacobian-based Saliency Map Attack (JSMA), and the Carlini Wagner (CW) l 0 -norm attack on CIFAR-10, COVID-CT, and ImageNet datasets. Our experimental results show that our approach can effectively defend against l 0 -norm attacks compared with the most popular defense techniques.

PURPOSE Propagation of contours from high-quality magnetic resonance (MR) images to treatment planning ultrasound (US) images with severe needle artifacts is a challenging task, which can greatly aid the organ contouring in high dose rate (HDR) prostate brachytherapy. In this study, a deep learning approach was developed to automatize this registration procedure for HDR brachytherapy practice. METHODS Because of the lack of training labels and difficulty of accurate registration from inferior image quality, a new segmentation-based registration framework was proposed for this multi-modality image registration problem. The framework consisted of two segmentation networks and a deformable registration network, based on the weakly -supervised registration strategy. Specifically, two 3D V-Nets were trained for the prostate segmentation on the MR and US images separately, to generate the weak supervision labels for the registration network training. Besides the image pair, the corresponding prostate probability maps from the segmentation were further fed to the registration network to predict the deformation matrix, and an augmentation method was designed to randomly scale the input and label probability maps during the registration network training. The overlap between the deformed and fixed prostate contours was analyzed to evaluate the registration accuracy. Three datasets were collected from our institution for the MR and US image segmentation networks, and the registration network learning, which contained 121, 104, and 63 patient cases, respectively. RESULTS The mean Dice similarity coefficient (DSC) results of the two prostate segmentation networks are 0.86 ± 0.05 and 0.90 ± 0.03, for MR images and the US images after the needle insertion, respectively. The mean DSC, center-of-mass (COM) distance, Hausdorff distance (HD), and averaged symmetric surface distance (ASSD) results for the registration of manual prostate contours were 0.87 ± 0.05, 1.70 ± 0.89 mm, 7.21 ± 2.07 mm, 1.61 ± 0.64 mm, respectively. By providing the prostate probability map from the segmentation to the registration network, as well as applying the random map augmentation method, the evaluation results of the four metrics were all improved, such as an increase in DSC from 0.83 ± 0.08 to 0.86 ± 0.06 and from 0.86 ± 0.06 to 0.87 ± 0.05, respectively. CONCLUSIONS A novel segmentation-based registration framework was proposed to automatically register prostate MR images to the treatment planning US images with metal artifacts, which not only largely saved the labor work on the data preparation, but also improved the registration accuracy. The evaluation results showed the potential of this approach in HDR prostate brachytherapy practice.

To better understand early brain development in health and disorder, it is critical to accurately segment infant brain magnetic resonance (MR) images into white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF). Deep learning-based methods have achieved state-of-the-art performance; h owever, one of the major limitations is that the learning-based methods may suffer from the multi-site issue, that is, the models trained on a dataset from one site may not be applicable to the datasets acquired from other sites with different imaging protocols/scanners. To promote methodological development in the community, the iSeg-2019 challenge ( http://iseg2019.web.unc.edu ) provides a set of 6-month infant subjects from multiple sites with different protocols/scanners for the participating methods. T raining/validation subjects are from UNC (MAP) and testing subjects are from UNC/UMN (BCP), Stanford University, and Emory University. By the time of writing, there are 30 automatic segmentation methods participated in the iSeg-2019. In this article, 8 top-ranked methods were reviewed by detailing their pipelines/implementations, presenting experimental results, and evaluating performance across different sites in terms of whole brain, regions of interest, and gyral landmark curves. We further pointed out their limitations and possible directions for addressing the multi-site issue. We find that multi-site consistency is still an open issue. We hope that the multi-site dataset in the iSeg-2019 and this review article will attract more researchers to address the challenging and critical multi-site issue in practice.

A common shortfall of supervised deep learning for medical imaging is the lack of labeled data, which is often expensive and time consuming to collect. This article presents a new semisupervised method for medical image segmentation, where the network is optimized by a weighted combination of a common supervised loss only for the labeled inputs and a regularization loss for both the labeled and unlabeled data. To utilize the unlabeled data, our method encourages consistent predictions of the network-in-training for the same input under different perturbations. With the semisupervised segmentation tasks, we introduce a transformation-consistent strategy in the self-ensembling model to enhance the regularization effect for pixel-level predictions. To further improve the regularization effects, we extend the transformation in a more generalized form including scaling and optimize the consistency loss with a teacher model, which is an averaging of the student model weights. We extensively validated the proposed semisupervised method on three typical yet challenging medical image segmentation tasks: 1) skin lesion segmentation from dermoscopy images in the International Skin Imaging Collaboration (ISIC) 2017 data set; 2) optic disk (OD) segmentation from fundus images in the Retinal Fundus Glaucoma Challenge (REFUGE) data set; and 3) liver segmentation from volumetric CT scans in the Liver Tumor Segmentation Challenge (LiTS) data set. Compared with state-of-the-art, our method shows superior performance on the challenging 2-D/3-D medical images, demonstrating the effectiveness of our semisupervised method for medical image segmentation.

Deep convolutional neural networks have significantly boosted the performance of fundus image segmentation when test datasets have the same distribution as the training datasets. However, in clinical practice, medical images often exhibit variations in appearance for various reasons, e.g., different scanner vendors and image quality. These distribution discrepancies could lead the deep networks to over-fit on the training datasets and lack generalization ability on the unseen test datasets. To alleviate this issue, we present a novel Domain-oriented Feature Embedding (DoFE) framework to improve the generalization ability of CNNs on unseen target domains by exploring the knowledge from multiple source domains. Our DoFE framework dynamically enriches the image features with additional domain prior knowledge learned from multi-source domains to make the semantic features more discriminative. Specifically, we introduce a Domain Knowledge Pool to learn and memorize the prior information extracted from multi-source domains. Then the original image features are augmented with domain-oriented aggregated features, which are induced from the knowledge pool based on the similarity between the input image and multi-source domain images. We further design a novel domain code prediction branch to infer this similarity and employ an attention-guided mechanism to dynamically combine the aggregated features with the semantic features. We comprehensively evaluate our DoFE framework on two fundus image segmentation tasks, including the optic cup and disc segmentation and vessel segmentation. Our DoFE framework generates satisfying segmentation results on unseen datasets and surpasses other domain generalization and network regularization methods., Comment: Accepted at IEEE-TMI 2020

Purpose Contouring intraprostatic lesions is a prerequisite for dose-escalating these lesions in radiotherapy to improve the local cancer control. In this study, a deep learning-based approach was developed for automatic intraprostatic lesion segmentation in multiparametric magnetic resonance imaging (mpMRI) images contributing to clinical practice. Methods Multiparametric magnetic resonance imaging images from 136 patient cases were collected from our institution, and all these cases contained suspicious lesions with Prostate Imaging Reporting and Data System (PI-RADS) score ≥ 4. The contours of the lesion and prostate were manually created on axial T2-weighted (T2W), apparent diffusion coefficient (ADC) and high b-value diffusion-weighted imaging (DWI) images to provide the ground truth data. Then a multiple branch UNet (MB-UNet) was proposed for the segmentation of an indistinct target in multi-modality MRI images. An encoder module was designed with three branches for the three MRI modalities separately, to fully extract the high-level features provided by different MRI modalities; an input module was added by using three sub-branches for three consecutive image slices, to consider the contour consistency among different image slices; deep supervision strategy was also integrated into the network to speed up the convergency of the network and improve the performance. The probability maps of the background, normal prostate and lesion were output by the network to generate the segmentation of the lesion, and the performance was evaluated using the dice similarity coefficient (DSC) as the main metric. Results A total of 162 lesions were contoured on 652 image slices, with 119 lesions in the peripheral zone, 38 in the transition zone, four in the central zone and one in the anterior fibromuscular stroma. All prostates were also contoured on 1,264 image slices. As for the segmentation of lesions in the testing set, MB-UNet achieved a per case DSC of 0.6333, specificity of 0.9993, sensitivity of 0.7056; and global DSC of 0.7205, specificity of 0.9993, sensitivity of 0.7409. All the three deep learning strategies adopted in this study contributed to the performance promotion of the MB-UNet. Missing the DWI modality would degrade the segmentation performance more markedly compared with the other two modalities. Conclusions A deep learning-based approach with proposed MB-UNet was developed to automatically segment suspicious lesions in mpMRI images. This study makes it feasible to adopt boosting intraprostatic lesions in clinical practice to achieve better outcomes.

The goal of few-shot learning is to recognize new visual concepts with just a few amount of labeled samples in each class. Recent effective metric-based few-shot approaches employ neural networks to learn a feature similarity comparison between query and support examples. However, the importance of feature embedding, i.e., exploring the relationship among training samples, is neglected. In this work, we present a simple yet powerful baseline for few-shot classification by emphasizing the importance of feature embedding. Specifically, we revisit the classical triplet network from deep metric learning, and extend it into a deep K-tuplet network for few-shot learning, utilizing the relationship among the input samples to learn a general representation learning via episode-training. Once trained, our network is able to extract discriminative features for unseen novel categories and can be seamlessly incorporated with a non-linear distance metric function to facilitate the few-shot classification. Our result on the miniImageNet benchmark outperforms other metric-based few-shot classification methods. More importantly, when evaluated on completely different datasets (Caltech-101, CUB-200, Stanford Dogs and Cars) using the model trained with miniImageNet, our method significantly outperforms prior methods, demonstrating its superior capability to generalize to unseen classes.

The success of deep convolutional neural networks is partially attributed to the massive amount of annotated training data. However, in practice, medical data annotations are usually expensive and time-consuming to be obtained. Considering multi-modality data with the same anatomic structures are widely available in clinic routine, in this paper, we aim to exploit the prior knowledge (e.g., shape priors) learned from one modality (aka., assistant modality) to improve the segmentation performance on another modality (aka., target modality) to make up annotation scarcity. To alleviate the learning difficulties caused by modality-specific appearance discrepancy, we first present an Image Alignment Module (IAM) to narrow the appearance gap between assistant and target modality data.We then propose a novel Mutual Knowledge Distillation (MKD) scheme to thoroughly exploit the modality-shared knowledge to facilitate the target-modality segmentation. To be specific, we formulate our framework as an integration of two individual segmentors. Each segmentor not only explicitly extracts one modality knowledge from corresponding annotations, but also implicitly explores another modality knowledge from its counterpart in mutual-guided manner. The ensemble of two segmentors would further integrate the knowledge from both modalities and generate reliable segmentation results on target modality. Experimental results on the public multi-class cardiac segmentation data, i.e., MMWHS 2017, show that our method achieves large improvements on CT segmentation by utilizing additional MRI data and outperforms other state-of-the-art multi-modality learning methods., Accepted by AAAI 2020

Deep learning is becoming an indispensable tool for imaging applications, such as image segmentation, classification, and detection. In this work, we reformulate a standard deep learning problem into a new neural network architecture with multi-output channels, which reflects different facets of the objective, and apply the deep neural network to improve the performance of image segmentation. By adding one or more interrelated auxiliary-output channels, we impose an effective consistency regularization for the main task of pixelated classification (i.e., image segmentation). Specifically, multi-output-channel consistency regularization is realized by residual learning via additive paths that connect main-output channel and auxiliary-output channels in the network. The method is evaluated on the detection and delineation of lung and liver tumors with public data. The results clearly show that multi-output-channel consistency implemented by residual learning improves the standard deep neural network. The proposed framework is quite broad and should find widespread applications in various deep learning problems.

The automatic diagnosis of various conventional ophthalmic diseases from fundus images is important in clinical practice. However, developing such automatic solutions is challenging due to the requirement of a large amount of training data and the expensive annotations for medical images. This paper presents a novel self-supervised learning framework for retinal disease diagnosis to reduce the annotation efforts by learning the visual features from the unlabeled images. To achieve this, we present a rotation-oriented collaborative method that explores rotation-related and rotation-invariant features, which capture discriminative structures from fundus images and also explore the invariant property used for retinal disease classification. We evaluate the proposed method on two public benchmark datasets for retinal disease classification. The experimental results demonstrate that our method outperforms other self-supervised feature learning methods (around 4.2% area under the curve (AUC)). With a large amount of unlabeled data available, our method can surpass the supervised baseline for pathologic myopia (PM) and is very close to the supervised baseline for age-related macular degeneration (AMD), showing the potential benefit of our method in clinical practice.

The presence of metallic implants often introduces severe metal artifacts in the X-ray CT images, which could adversely influence clinical diagnosis or dose calculation in radiation therapy. In this work, we present a novel deep-learning-based approach for metal artifact reduction (MAR). In order to alleviate the need for anatomically identical CT image pairs (i.e., metal artifact-corrupted CT image and metal artifact-free CT image) for network learning, we propose a self-supervised cross-domain learning framework. Specifically, we train a neural network to restore the metal trace region values in the given metal-free sinogram, where the metal trace is identified by the forward projection of metal masks. We then design a novel FBP reconstruction loss to encourage the network to generate more perfect completion results and a residual-learning-based image refinement module to reduce the secondary artifacts in the reconstructed CT images. To preserve the fine structure details and fidelity of the final MAR image, instead of directly adopting CNN-refined images as output, we incorporate the metal trace replacement into our framework and replace the metal-affected projections of the original sinogram with the prior sinogram generated by the forward projection of the CNN output. We then use the filtered backward projection (FBP) algorithms for final MAR image reconstruction. We conduct an extensive evaluation on simulated and real artifact data to show the effectiveness of our design. Our method produces superior MAR results and outperforms other compelling methods. We also demonstrate the potential of our framework for other organ sites., Accepted by PMB

PURPOSE High-performance computed tomography (CT) plays a vital role in clinical decision-making. However, the performance of CT imaging is adversely affected by the nonideal focal spot size of the x-ray source or degraded by an enlarged focal spot size due to aging. In this work, we aim to develop a deep learning-based strategy to mitigate the problem so that high spatial resolution CT images can be obtained even in the case of a nonideal x-ray source. METHODS To reconstruct high-quality CT images from blurred sinograms via joint image and sinogram learning, a cross-domain hybrid model is formulated via deep learning into a modularized data-driven reconstruction (MDR) framework. The proposed MDR framework comprises several blocks, and all the blocks share the same network architecture and network parameters. In essence, each block utilizes two sub-models to generate an estimated blur kernel and a high-quality CT image simultaneously. In this way, our framework generates not only a final high-quality CT image but also a series of intermediate images with gradually improved anatomical details, enhancing the visual perception for clinicians through the dynamic process. We used simulated training datasets to train our model in an end-to-end manner and tested our model on both simulated and realistic experimental datasets. RESULTS On the simulated testing datasets, our approach increases the information fidelity criterion (IFC) by up to 34.2%, the universal quality index (UQI) by up to 20.3%, the signal-to-noise (SNR) by up to 6.7%, and reduces the root mean square error (RMSE) by up to 10.5% as compared with FBP. Compared with the iterative deconvolution method (NSM), MDR increases IFC by up to 24.7%, UQI by up to 16.7%, SNR by up to 6.0%, and reduces RMSE by up to 9.4%. In the modulation transfer function (MTF) experiment, our method improves the MTF50% by 34.5% and MTF10% by 18.7% as compared with FBP, Similarly remarkably, our method improves MTF50% by 14.3% and MTF10% by 0.9% as compared with NSM. Also, our method shows better imaging results in the edge of bony structures and other tiny structures in the experiments using phantom consisting of ham and a bottle of peanuts. CONCLUSIONS A modularized data-driven CT reconstruction framework is established to mitigate the blurring effect caused by a nonideal x-ray source with relatively large focal spot. The proposed method enables us to obtain high-resolution images with less ideal x-ray source.

Learning from external data is an effective and efficient way of training deep networks, which can substantially alleviate the burden on collecting training data and annotations. It is of great significance in improving the performance of CT image segmentation tasks, where collecting a large amount of voxel-wise annotations is expensive or even impractical. In this paper, we propose a generic selective learning method to maximize the performance gains of harnessing external data in CT image segmentation. The key idea is to learn a weight for each external data such that ‘good’ data can have large weights and thus contribute more to the training loss, thereby implicitly encouraging the network to mine more valuable knowledge from informative external data while suppressing to memorize irrelevant patterns from ‘useless’ or even ‘harmful’ data. Particularly, we formulate our idea as a constrained non-linear programming problem, solved by an iterative solution that alternatively conducts weights estimating and network updating. Extensive experiments on abdominal multi-organ CT segmentation datasets show the efficacy and performance gains of our method against existing methods. The code is publicly available (Released at https://github.com/YouyiSong/Codes-for-Selective-Learning).

Annotation scarcity is a long-standing problem in medical image analysis area. To efficiently leverage limited annotations, abundant unlabeled data are additionally exploited in semi-supervised learning, while well-established cross-modality data are investigated in domain adaptation. In this paper, we aim to explore the feasibility of concurrently leveraging both unlabeled data and cross-modality data for annotation-efficient cardiac segmentation. To this end, we propose a cutting-edge semi-supervised domain adaptation framework, namely Dual-Teacher++. Besides directly learning from limited labeled target domain data (e.g., CT) via a student model adopted by previous literature, we design novel dual teacher models, including an inter-domain teacher model to explore cross-modality priors from source domain (e.g., MR) and an intra-domain teacher model to investigate the knowledge beneath unlabeled target domain. In this way, the dual teacher models would transfer acquired inter- and intra-domain knowledge to the student model for further integration and exploitation. Moreover, to encourage reliable dual-domain knowledge transfer, we enhance the inter-domain knowledge transfer on the samples with higher similarity to target domain after appearance alignment, and also strengthen intra-domain knowledge transfer of unlabeled target data with higher prediction confidence. In this way, the student model can obtain reliable dual-domain knowledge and yield improved performance on target domain data. We extensively evaluated the feasibility of our method on the MM-WHS 2017 challenge dataset. The experiments have demonstrated the superiority of our framework over other semi-supervised learning and domain adaptation methods. Moreover, our performance gains could be yielded in bidirections,i.e., adapting from MR to CT, and from CT to MR., Comment: Accepted by TMI

The automatic diagnosis of various retinal diseases from fundus images is important to support clinical decision-making. However, developing such automatic solutions is challenging due to the requirement of a large amount of human-annotated data. Recently, unsupervised/self-supervised feature learning techniques receive a lot of attention, as they do not need massive annotations. Most of the current self-supervised methods are analyzed with single imaging modality and there is no method currently utilize multi-modal images for better results. Considering that the diagnostics of various vitreoretinal diseases can greatly benefit from another imaging modality, e.g., FFA, this paper presents a novel self-supervised feature learning method by effectively exploiting multi-modal data for retinal disease diagnosis. To achieve this, we first synthesize the corresponding FFA modality and then formulate a patient feature-based softmax embedding objective. Our objective learns both modality-invariant features and patient-similarity features. Through this mechanism, the neural network captures the semantically shared information across different modalities and the apparent visual similarity between patients. We evaluate our method on two public benchmark datasets for retinal disease diagnosis. The experimental results demonstrate that our method clearly outperforms other self-supervised feature learning methods and is comparable to the supervised baseline., Comment: IEEE Transactions on Medical Imaging, code is at https://github.com/xmengli999/self_supervised

Although having achieved great success in medical image segmentation, deep learning-based approaches usually require large amounts of well-annotated data, which can be extremely expensive in the field of medical image analysis. Unlabeled data, on the other hand, is much easier to acquire. Semi-supervised learning and unsupervised domain adaptation both take the advantage of unlabeled data, and they are closely related to each other. In this paper, we propose uncertainty-aware multi-view co-training (UMCT), a unified framework that addresses these two tasks for volumetric medical image segmentation. Our framework is capable of efficiently utilizing unlabeled data for better performance. We firstly rotate and permute the 3D volumes into multiple views and train a 3D deep network on each view. We then apply co-training by enforcing multi-view consistency on unlabeled data, where an uncertainty estimation of each view is utilized to achieve accurate labeling. Experiments on the NIH pancreas segmentation dataset and a multi-organ segmentation dataset show state-of-the-art performance of the proposed framework on semi-supervised medical image segmentation. Under unsupervised domain adaptation settings, we validate the effectiveness of this work by adapting our multi-organ segmentation model to two pathological organs from the Medical Segmentation Decathlon Datasets. Additionally, we show that our UMCT-DA model can even effectively handle the challenging situation where labeled source data is inaccessible, demonstrating strong potentials for real-world applications., 19 pages, 6 figures, to appear in Medical Image Analysis. This article is an extension of the conference paper arXiv:1811.12506

Deep learning approaches have demonstrated remarkable progress in automatic Chest X-ray analysis. The data-driven feature of deep models requires training data to cover a large distribution. Therefore, it is substantial to integrate knowledge from multiple datasets, especially for medical images. However, learning a disease classification model with extra Chest X-ray (CXR) data is yet challenging. Recent researches have demonstrated that performance bottleneck exists in joint training on different CXR datasets, and few made efforts to address the obstacle. In this paper, we argue that incorporating an external CXR dataset leads to imperfect training data, which raises the challenges. Specifically, the imperfect data is in two folds: domain discrepancy, as the image appearances vary across datasets; and label discrepancy, as different datasets are partially labeled. To this end, we formulate the multi-label thoracic disease classification problem as weighted independent binary tasks according to the categories. For common categories shared across domains, we adopt task-specific adversarial training to alleviate the feature differences. For categories existing in a single dataset, we present uncertainty-aware temporal ensembling of model predictions to mine the information from the missing labels further. In this way, our framework simultaneously models and tackles the domain and label discrepancies, enabling superior knowledge mining ability. We conduct extensive experiments on three datasets with more than 360,000 Chest X-ray images. Our method outperforms other competing models and sets state-of-the-art performance on the official NIH test set with 0.8349 AUC, demonstrating its effectiveness of utilizing the external dataset to improve the internal classification., Accepted to IEEE Transactions on Medical Imaging

Volumetric ultrasound is rapidly emerging as a viable imaging modality for routine prenatal examinations. Biometrics obtained from the volumetric segmentation shed light on the reformation of precise maternal and fetal health monitoring. However, the poor image quality, low contrast, boundary ambiguity, and complex anatomy shapes conspire toward a great lack of efficient tools for the segmentation. It makes 3-D ultrasound difficult to interpret and hinders the widespread of 3-D ultrasound in obstetrics. In this paper, we are looking at the problem of semantic segmentation in prenatal ultrasound volumes. Our contribution is threefold: 1) we propose the first and fully automatic framework to simultaneously segment multiple anatomical structures with intensive clinical interest, including fetus, gestational sac, and placenta, which remains a rarely studied and arduous challenge; 2) we propose a composite architecture for dense labeling, in which a customized 3-D fully convolutional network explores spatial intensity concurrency for initial labeling, while a multi-directional recurrent neural network (RNN) encodes spatial sequentiality to combat boundary ambiguity for significant refinement; and 3) we introduce a hierarchical deep supervision mechanism to boost the information flow within RNN and fit the latent sequence hierarchy in fine scales, and further improve the segmentation results. Extensively verified on in-house large data sets, our method illustrates a superior segmentation performance, decent agreements with expert measurements and high reproducibilities against scanning variations, and thus is promising in advancing the prenatal ultrasound examinations.

Automated prostate segmentation in MRI is highly demanded for computer-assisted diagnosis. Recently, a variety of deep learning methods have achieved remarkable progress in this task, usually relying on large amounts of training data. Due to the nature of scarcity for medical images, it is important to effectively aggregate data from multiple sites for robust model training, to alleviate the insufficiency of single-site samples. However, the prostate MRIs from different sites present heterogeneity due to the differences in scanners and imaging protocols, raising challenges for effective ways of aggregating multi-site data for network training. In this paper, we propose a novel multi-site network (MS-Net) for improving prostate segmentation by learning robust representations, leveraging multiple sources of data. To compensate for the inter-site heterogeneity of different MRI datasets, we develop Domain-Specific Batch Normalization layers in the network backbone, enabling the network to estimate statistics and perform feature normalization for each site separately. Considering the difficulty of capturing the shared knowledge from multiple datasets, a novel learning paradigm, i.e., Multi-site-guided Knowledge Transfer, is proposed to enhance the kernels to extract more generic representations from multi-site data. Extensive experiments on three heterogeneous prostate MRI datasets demonstrate that our MS-Net improves the performance across all datasets consistently, and outperforms state-of-the-art methods for multi-site learning., IEEE TMI, 2020

PURPOSE/OBJECTIVE(S) X-ray imaging is a widely used approach to view the internal structure of a subject for clinical diagnosis, image-guided interventions and decision-making. The X-ray projections acquired at different view angles provide complementary information of patient's anatomy and are required for stereoscopic or volumetric imaging of the subject. In reality, obtaining multiple-view projections inevitably increases radiation dose and complicates clinical workflow. Here we investigate a strategy of obtaining the X-ray projection image at a novel view angle from a given projection image at a specific view angle to alleviate the need for actual projection measurement. Specifically, we propose a deep learning-based geometry-integrated projection synthesis framework for generating novel-view X-ray projections. MATERIALS/METHODS We for the first time investigate the novel-view projection synthesis problem for X-ray imaging. We propose a deep learning-based geometry-integrated projection synthesis model to generate novel-view X-ray projections through feature disentanglement and geometry transformation. The proposed deep learning model extracts geometry and texture features from a source-view projection, and then conducts geometry transformation on the geometry features to accommodate the change of view angle. Then the X-ray projection in target view is synthesized from the transformed geometry and shared texture features via an image generator. We validate the proposed approach by experimenting on one-to-one and multi-to-multi X-ray projection synthesis using lung imaging cases across various patients. We conduct experiments on a public dataset containing 1018 lung CT images, where 80% and 20% data are used for training and testing. The projection images are digitally produced from CT images using geometry consistent with a clinical on-board cone-beam CT system for radiation therapy. RESULTS We deploy the trained model on the held-out testing set for novel-view projection synthesis. The generated results are compared with ground truth qualitatively and quantitatively. For AP-to-lateral and lateral-to-AP view synthesis, average NRMSE / SSIM / PSNR values over all testing data are 0.185 / 0.880 / 22.646 and 0.173 / 0.911 / 23.908, respectively. Promising results are also obtained for multi-to-multi view synthesis. By visualizing the generated projections, we observe the proposed model can generate images closely to the targets despite the various anatomic structures of different patients, indicating the potential of the proposed model for X-ray projection synthesis. CONCLUSION This work investigates the synthesis of novel-view X-ray projections and presents a robust model combining the deep learning and geometry transformation. It is shown that the generated X-ray projections reveal the internal anatomy distribution from new viewpoints, which potentially provides a new paradigm for volumetric imaging with substantially reduced efforts in data acquisition.

We propose an analysis of surgical videos that is based on a novel recurrent convolutional network (SV-RCNet), specifically for automatic workflow recognition from surgical videos online, which is a key component for developing the context-aware computer-assisted intervention systems. Different from previous methods which harness visual and temporal information separately, the proposed SV-RCNet seamlessly integrates a convolutional neural network (CNN) and a recurrent neural network (RNN) to form a novel recurrent convolutional architecture in order to take full advantages of the complementary information of visual and temporal features learned from surgical videos. We effectively train the SV-RCNet in an end-to-end manner so that the visual representations and sequential dynamics can be jointly optimized in the learning process. In order to produce more discriminative spatio-temporal features, we exploit a deep residual network (ResNet) and a long short term memory (LSTM) network, to extract visual features and temporal dependencies, respectively, and integrate them into the SV-RCNet. Moreover, based on the phase transition-sensitive predictions from the SV-RCNet, we propose a simple yet effective inference scheme, namely the prior knowledge inference (PKI), by leveraging the natural characteristic of surgical video. Such a strategy further improves the consistency of results and largely boosts the recognition performance. Extensive experiments have been conducted with the MICCAI 2016 Modeling and Monitoring of Computer Assisted Interventions Workflow Challenge dataset and Cholec80 dataset to validate SV-RCNet. Our approach not only achieves superior performance on these two datasets but also outperforms the state-of-the-art methods by a significant margin.

Purpose/Objective(s) Propagation of contours from high-quality magnetic resonance (MR) images to treatment planning ultrasound (US) images with severe needle artifacts is a challenging task, which can greatly aid the organ or intraprostatic lesion contouring in high dose rate (HDR) prostate brachytherapy. In this study, a deep learning approach was developed to automatize this registration procedure for HDR brachytherapy practice. Materials/Methods With the approval from the institutional review board, the MR and US images with manual prostate contours were collected from our clinical database and processed to construct the deep learning dataset. A deep learning framework based on the weakly-supervised strategy was proposed for the registration of prostate MR images and US images after the needle insertion, which consisted of two segmentation networks and a deformable registration network. Specifically, two 3D V-Net was trained for the prostate segmentation on the MR and US images separately, to generate the weak supervision labels automatically for the registration network training. Besides the image pair, the corresponding prostate probability maps from the segmentation were further fed to the registration network to improve the registration accuracy, and the deformation matrix was the output. A random augmentation method was designed to alleviate the map overfitting problem during the network training. The overlap between the deformed and fixed prostate contours was analyzed to evaluate the registration accuracy. Results Clinical data from 121, 104 and 63 patient cases were collected and utilized for the MR and US image segmentation networks, and the registration network learning, respectively. The mean DSC, center-of-mass (COM) distance, Hausdorff distance (HD) and averaged symmetric surface distance (ASSD) results for the registration of prostate contours were 0.87 ± 0.05, 1.70 ± 0.89 mm, 7.21 ± 2.07 mm, 1.61 ± 0.64 mm, respectively. By providing the prostate probability map from the segmentation to the registration network, as well as applying the random map augmentation method, the evaluation results of the four metrics were all improved, such as an increase of DSC from 0.83 ± 0.08 to 0.86 ± 0.06 and from 0.86 ± 0.06 to 0.87 ± 0.05, respectively. Conclusion This study first worked on the deep learning-based registration about the treatment planning US image for HDR prostate brachytherapy. A novel segmentation-based registration framework was proposed for this multi-modality registration problem, which not only largely saved the labor work on the data preparation, but also improved the registration accuracy. The evaluation results showed the potential of this approach in HDR brachytherapy to automatize current clinical procedures.

While deep convolutional neural networks (CNNs) have achieved remarkable success in 2D medical image segmentation, it is still a difficult task for CNNs to segment important organs or structures from 3D medical images owing to several mutually affected challenges, including the complicated anatomical environments in volumetric images, optimization difficulties of 3D networks and inadequacy of training samples. In this paper, we present a novel and efficient 3D fully convolutional network equipped with a 3D deep supervision mechanism to comprehensively address these challenges; we call it 3D DSN. Our proposed 3D DSN is capable of conducting volume-to-volume learning and inference, which can eliminate redundant computations and alleviate the risk of over-fitting on limited training data. More importantly, the 3D deep supervision mechanism can effectively cope with the optimization problem of gradients vanishing or exploding when training a 3D deep model, accelerating the convergence speed and simultaneously improving the discrimination capability. Such a mechanism is developed by deriving an objective function that directly guides the training of both lower and upper layers in the network, so that the adverse effects of unstable gradient changes can be counteracted during the training procedure. We also employ a fully connected conditional random field model as a post-processing step to refine the segmentation results. We have extensively validated the proposed 3D DSN on two typical yet challenging volumetric medical image segmentation tasks: (i) liver segmentation from 3D CT scans and (ii) whole heart and great vessels segmentation from 3D MR images, by participating two grand challenges held in conjunction with MICCAI. We have achieved competitive segmentation results to state-of-the-art approaches in both challenges with a much faster speed, corroborating the effectiveness of our proposed 3D DSN.

Automated melanoma recognition in dermoscopy images is a very challenging task due to the low contrast of skin lesions, the huge intraclass variation of melanomas, the high degree of visual similarity between melanoma and non-melanoma lesions, and the existence of many artifacts in the image. In order to meet these challenges, we propose a novel method for melanoma recognition by leveraging very deep convolutional neural networks (CNNs). Compared with existing methods employing either low-level hand-crafted features or CNNs with shallower architectures, our substantially deeper networks (more than 50 layers) can acquire richer and more discriminative features for more accurate recognition. To take full advantage of very deep networks, we propose a set of schemes to ensure effective training and learning under limited training data. First, we apply the residual learning to cope with the degradation and overfitting problems when a network goes deeper. This technique can ensure that our networks benefit from the performance gains achieved by increasing network depth. Then, we construct a fully convolutional residual network (FCRN) for accurate skin lesion segmentation, and further enhance its capability by incorporating a multi-scale contextual information integration scheme. Finally, we seamlessly integrate the proposed FCRN (for segmentation) and other very deep residual networks (for classification) to form a two-stage framework. This framework enables the classification network to extract more representative and specific features based on segmented results instead of the whole dermoscopy images, further alleviating the insufficiency of training data. The proposed framework is extensively evaluated on ISBI 2016 Skin Lesion Analysis Towards Melanoma Detection Challenge dataset. Experimental results demonstrate the significant performance gains of the proposed framework, ranking the first in classification and the second in segmentation among 25 teams and 28 teams, respectively. This study corroborates that very deep CNNs with effective training mechanisms can be employed to solve complicated medical image analysis tasks, even with limited training data.

In histopathological image analysis, the morphology of histological structures, such as glands and nuclei, has been routinely adopted by pathologists to assess the malignancy degree of adenocarcinomas. Accurate detection and segmentation of these objects of interest from histology images is an essential prerequisite to obtain reliable morphological statistics for quantitative diagnosis. While manual annotation is error-prone, time-consuming and operator-dependant, automated detection and segmentation of objects of interest from histology images can be very challenging due to the large appearance variation, existence of strong mimics, and serious degeneration of histological structures. In order to meet these challenges, we propose a novel deep contour-aware network (DCAN) under a unified multi-task learning framework for more accurate detection and segmentation. In the proposed network, multi-level contextual features are explored based on an end-to-end fully convolutional network (FCN) to deal with the large appearance variation. We further propose to employ an auxiliary supervision mechanism to overcome the problem of vanishing gradients when training such a deep network. More importantly, our network can not only output accurate probability maps of histological objects, but also depict clear contours simultaneously for separating clustered object instances, which further boosts the segmentation performance. Our method ranked the first in two histological object segmentation challenges, including 2015 MICCAI Gland Segmentation Challenge and 2015 MICCAI Nuclei Segmentation Challenge. Extensive experiments on these two challenging datasets demonstrate the superior performance of our method, surpassing all the other methods by a significant margin.

Medical image annotations are prohibitively time-consuming and expensive to obtain. To alleviate annotation scarcity, many approaches have been developed to efficiently utilize extra information, e.g., semi-supervised learning further exploring plentiful unlabeled data, domain adaptation including multi-modality learning and unsupervised domain adaptation resorting to the prior knowledge from additional modality. In this paper, we aim to investigate the feasibility of simultaneously leveraging abundant unlabeled data and well-established cross-modality data for annotation-efficient medical image segmentation. To this end, we propose a novel semi-supervised domain adaptation approach, namely Dual-Teacher, where the student model not only learns from labeled target data (e.g., CT), but also explores unlabeled target data and labeled source data (e.g., MR) by two teacher models. Specifically, the student model learns the knowledge of unlabeled target data from intra-domain teacher by encouraging prediction consistency, as well as the shape priors embedded in labeled source data from inter-domain teacher via knowledge distillation. Consequently, the student model can effectively exploit the information from all three data resources and comprehensively integrate them to achieve improved performance. We conduct extensive experiments on MM-WHS 2017 dataset and demonstrate that our approach is able to concurrently utilize unlabeled data and cross-modality data with superior performance, outperforming semi-supervised learning and domain adaptation methods with a large margin.

Training deep neural networks usually requires a large amount of labeled data to obtain good performance. However, in medical image analysis, obtaining high-quality labels for the data is laborious and expensive, as accurately annotating medical images demands expertise knowledge of the clinicians. In this paper, we present a novel relation-driven semi-supervised framework for medical image classification. It is a consistency-based method which exploits the unlabeled data by encouraging the prediction consistency of given input under perturbations, and leverages a self-ensembling model to produce high-quality consistency targets for the unlabeled data. Considering that human diagnosis often refers to previous analogous cases to make reliable decisions, we introduce a novel sample relation consistency (SRC) paradigm to effectively exploit unlabeled data by modeling the relationship information among different samples. Superior to existing consistency-based methods which simply enforce consistency of individual predictions, our framework explicitly enforces the consistency of semantic relation among different samples under perturbations, encouraging the model to explore extra semantic information from unlabeled data. We have conducted extensive experiments to evaluate our method on two public benchmark medical image classification datasets, i.e.,skin lesion diagnosis with ISIC 2018 challenge and thorax disease classification with ChestX-ray14. Our method outperforms many state-of-the-art semi-supervised learning methods on both single-label and multi-label image classification scenarios., Comment: IEEE Transactions on Medical Imaging, 2020

Rare diseases have extremely low-data regimes, unlike common diseases with large amount of available labeled data. Hence, to train a neural network to classify rare diseases with a few per-class data samples is very challenging, and so far, catches very little attention. In this paper, we present a difficulty-aware meta-learning method to address rare disease classifications and demonstrate its capability to classify dermoscopy images. Our key approach is to first train and construct a meta-learning model from data of common diseases, then adapt the model to perform rare disease classification. To achieve this, we develop the difficulty-aware meta-learning method that dynamically monitors the importance of learning tasks during the meta-optimization stage. To evaluate our method, we use the recent ISIC 2018 skin lesion classification dataset, and show that with only five samples per class, our model can quickly adapt to classify unseen classes by a high AUC of 83.3%. Also, we evaluated several rare disease classification results in the public Dermofit Image Library to demonstrate the potential of our method for real clinical practice.

Deep convolutional neural networks (CNNs) have achieved commendable results on a variety of medical image segmentation tasks. However, CNNs usually require a large amount of training samples with accurate annotations, which are extremely difficult and expensive to obtain in medical image analysis field. In practice, we notice that the junior trainees after training can label medical images in some medical image segmentation applications. These non-expert annotations are more easily accessible and can be regarded as a source of weak annotation to guide network learning. In this paper, we propose a novel Tri-network learning framework to alleviate the problem of insufficient accurate annotations in medical segmentation tasks by utilizing the non-expert annotations. To be specific, we maintain three networks in our framework, and each pair of networks alternatively select informative samples for the third network learning, according to the consensus and difference between their predictions. The three networks are jointly optimized in such a collaborative manner. We evaluated our method on real and simulated non-expert annotated datasets. The experiment results show that our method effectively mines informative information from the non-expert annotations for improved segmentation performance and outperforms other competing methods.

Computed tomography (CT) has been widely used for medical diagnosis, assessment, and therapy planning and guidance. In reality, CT images may be affected adversely in the presence of metallic objects, which could lead to severe metal artifacts and influence clinical diagnosis or dose calculation in radiation therapy. In this paper, we propose a generalizable framework for metal artifact reduction (MAR) by simultaneously leveraging the advantages of image domain and sinogram domain-based MAR techniques. We formulate our framework as a sinogram completion problem and train a neural network (SinoNet) to restore the metal-affected projections. To improve the continuity of the completed projections at the boundary of metal trace and thus alleviate new artifacts in the reconstructed CT images, we train another neural network (PriorNet) to generate a good prior image to guide sinogram learning, and further design a novel residual sinogram learning strategy to effectively utilize the prior image information for better sinogram completion. The two networks are jointly trained in an end-to-end fashion with a differentiable forward projection (FP) operation so that the prior image generation and deep sinogram completion procedures can benefit from each other. Finally, the artifact-reduced CT images are reconstructed using the filtered backward projection (FBP) from the completed sinogram. Extensive experiments on simulated and real artifacts data demonstrate that our method produces superior artifact-reduced results while preserving the anatomical structures and outperforms other MAR methods., Comment: Accepted by IEEE Transactions on Medical Imaging

Emerging self-ensembling methods have achieved promising semi-supervised segmentation performances on medical images through forcing consistent predictions of unannotated data under different perturbations. However, the consistency only penalizes on independent pixel-level predictions, making structure-level information of predictions not exploited in the learning procedure. In view of this, we propose a novel structure-aware entropy regularized mean teacher model to address the above limitation. Specifically, we firstly introduce the entropy minimization principle to the student network, thereby adjusting itself to produce high-confident predictions of unannotated images. Based on this, we design a local structural consistency loss to encourage the consistency of inter-voxel similarities within the same local region of predictions from teacher and student networks. To further capture local structural dependencies, we enforce the global structural consistency by matching the weighted self-information maps between two networks. In this way, our model can minimize the prediction uncertainty of unannotated images, and more importantly that it can capture local and global structural information and their complementarity. We evaluate the proposed method on a publicly available 3D left atrium MR image dataset. Experimental results demonstrate that our method achieves outstanding segmentation performances than the state-of-the-art approaches in scenes with limited annotated images.

The generalization capability of neural networks across domains is crucial for real-world applications. We argue that a generalized object recognition system should well understand the relationships among different images and also the images themselves at the same time. To this end, we present a new domain generalization framework (called EISNet) that learns how to generalize across domains simultaneously from extrinsic relationship supervision and intrinsic self-supervision for images from multi-source domains. To be specific, we formulate our framework with feature embedding using a multi-task learning paradigm. Besides conducting the common supervised recognition task, we seamlessly integrate a momentum metric learning task and a self-supervised auxiliary task to collectively integrate the extrinsic and intrinsic supervisions. Also, we develop an effective momentum metric learning scheme with the K-hard negative mining to boost the network generalization ability. We demonstrate the effectiveness of our approach on two standard object recognition benchmarks VLCS and PACS, and show that our EISNet achieves state-of-the-art performance.

Automated polyp detection in colonoscopy videos has been demonstrated to be a promising way for colorectal cancer prevention and diagnosis. Traditional manual screening is time consuming, operator dependent, and error prone; hence, automated detection approach is highly demanded in clinical practice. However, automated polyp detection is very challenging due to high intraclass variations in polyp size, color, shape, and texture, and low interclass variations between polyps and hard mimics. In this paper, we propose a novel offline and online three-dimensional (3-D) deep learning integration framework by leveraging the 3-D fully convolutional network (3D-FCN) to tackle this challenging problem. Compared with the previous methods employing hand-crafted features or 2-D convolutional neural network, the 3D-FCN is capable of learning more representative spatio-temporal features from colonoscopy videos, and hence has more powerful discrimination capability. More importantly, we propose a novel online learning scheme to deal with the problem of limited training data by harnessing the specific information of an input video in the learning process. We integrate offline and online learning to effectively reduce the number of false positives generated by the offline network and further improve the detection performance. Extensive experiments on the dataset of MICCAI 2015 Challenge on Polyp Detection demonstrated the better performance of our method when compared with other competitors.

Cerebral microbleeds (CMBs) are small haemorrhages nearby blood vessels. They have been recognized as important diagnostic biomarkers for many cerebrovascular diseases and cognitive dysfunctions. In current clinical routine, CMBs are manually labelled by radiologists but this procedure is laborious, time-consuming, and error prone. In this paper, we propose a novel automatic method to detect CMBs from magnetic resonance (MR) images by exploiting the 3D convolutional neural network (CNN). Compared with previous methods that employed either low-level hand-crafted descriptors or 2D CNNs, our method can take full advantage of spatial contextual information in MR volumes to extract more representative high-level features for CMBs, and hence achieve a much better detection accuracy. To further improve the detection performance while reducing the computational cost, we propose a cascaded framework under 3D CNNs for the task of CMB detection. We first exploit a 3D fully convolutional network (FCN) strategy to retrieve the candidates with high probabilities of being CMBs, and then apply a well-trained 3D CNN discrimination model to distinguish CMBs from hard mimics. Compared with traditional sliding window strategy, the proposed 3D FCN strategy can remove massive redundant computations and dramatically speed up the detection process. We constructed a large dataset with 320 volumetric MR scans and performed extensive experiments to validate the proposed method, which achieved a high sensitivity of 93.16% with an average number of 2.74 false positives per subject, outperforming previous methods using low-level descriptors or 2D CNNs by a significant margin. The proposed method, in principle, can be adapted to other biomarker detection tasks from volumetric medical data.

Standard plane localization is crucial for ultrasound (US) diagnosis. In prenatal US, dozens of standard planes are manually acquired with a 2D probe. It is time-consuming and operator-dependent. In comparison, 3D US containing multiple standard planes in one shot has the inherent advantages of less user-dependency and more efficiency. However, manual plane localization in US volume is challenging due to the huge search space and large fetal posture variation. In this study, we propose a novel reinforcement learning (RL) framework to automatically localize fetal brain standard planes in 3D US. Our contribution is two-fold. First, we equip the RL framework with a landmark-aware alignment module to provide warm start and strong spatial bounds for the agent actions, thus ensuring its effectiveness. Second, instead of passively and empirically terminating the agent inference, we propose a recurrent neural network based strategy for active termination of the agent’s interaction procedure. This improves both the accuracy and efficiency of the localization system. Extensively validated on our in-house large dataset, our approach achieves the accuracy of 3.4 mm/9.6\(^{\circ }\) and 2.7 mm/9.1\(^{\circ }\) for the transcerebellar and transthalamic plane localization, respectively. Our proposed RL framework is general and has the potential to improve the efficiency and standardization of US scanning.

Glaucoma is a leading cause of irreversible blindness. Accurate segmentation of the optic disc (OD) and cup (OC) from fundus images is beneficial to glaucoma screening and diagnosis. Recently, convolutional neural networks demonstrate promising progress in joint OD and OC segmentation. However, affected by the domain shift among different datasets, deep networks are severely hindered in generalizing across different scanners and institutions. In this paper, we present a novel patchbased Output Space Adversarial Learning framework (pOSAL) to jointly and robustly segment the OD and OC from different fundus image datasets. We first devise a lightweight and efficient segmentation network as a backbone. Considering the specific morphology of OD and OC, a novel morphology-aware segmentation loss is proposed to guide the network to generate accurate and smooth segmentation. Our pOSAL framework then exploits unsupervised domain adaptation to address the domain shift challenge by encouraging the segmentation in the target domain to be similar to the source ones. Since the whole-segmentationbased adversarial loss is insufficient to drive the network to capture segmentation details, we further design the pOSAL in a patch-based fashion to enable fine-grained discrimination on local segmentation details. We extensively evaluate our pOSAL framework and demonstrate its effectiveness in improving the segmentation performance on three public retinal fundus image datasets, i.e., Drishti-GS, RIM-ONE-r3, and REFUGE. Furthermore, our pOSAL framework achieved the first place in the OD and OC segmentation tasks in MICCAI 2018 Retinal Fundus Glaucoma Challenge., Comment: IEEE Transactions on Medical Imaging (In press)

Training deep convolutional neural networks usually requires a large amount of labeled data. However, it is expensive and time-consuming to annotate data for medical image segmentation tasks. In this paper, we present a novel uncertainty-aware semi-supervised framework for left atrium segmentation from 3D MR images. Our framework can effectively leverage the unlabeled data by encouraging consistent predictions of the same input under different perturbations. Concretely, the framework consists of a student model and a teacher model, and the student model learns from the teacher model by minimizing a segmentation loss and a consistency loss with respect to the targets of the teacher model. We design a novel uncertainty-aware scheme to enable the student model to gradually learn from the meaningful and reliable targets by exploiting the uncertainty information. Experiments show that our method achieves high performance gains by incorporating the unlabeled data. Our method outperforms the state-of-the-art semi-supervised methods, demonstrating the potential of our framework for the challenging semi-supervised problems.

Accurate segmentation of the optic disc (OD) and cup (OC) in fundus images from different datasets is critical for glaucoma disease screening. The cross-domain discrepancy (domain shift) hinders the generalization of deep neural networks to work on different domain datasets. In this work, we present an unsupervised domain adaptation framework, called Boundary and Entropy-driven Adversarial Learning (BEAL), to improve the OD and OC segmentation performance, especially on the ambiguous boundary regions. In particular, our proposed BEAL framework utilizes the adversarial learning to encourage the boundary prediction and mask probability entropy map (uncertainty map) of the target domain to be similar to the source ones, generating more accurate boundaries and suppressing the high uncertainty predictions of OD and OC segmentation. We evaluate the proposed BEAL framework on two public retinal fundus image datasets (Drishti-GS and RIM-ONE-r3), and the experiment results demonstrate that our method outperforms the state-of-the-art unsupervised domain adaptation methods. Our code is available at https://github.com/EmmaW8/BEAL.

This paper presents a novel approach to learn and detect distinctive regions on 3D shapes. Unlike previous works, which require labeled data, our method is unsupervised. We conduct the analysis on point sets sampled from 3D shapes, then formulate and train a deep neural network for an unsupervised shape clustering task to learn local and global features for distinguishing shapes with respect to a given shape set. To drive the network to learn in an unsupervised manner, we design a clustering-based nonparametric softmax classifier with an iterative re-clustering of shapes, and an adapted contrastive loss for enhancing the feature embedding quality and stabilizing the learning process. By then, we encourage the network to learn the point distinctiveness on the input shapes. We extensively evaluate various aspects of our approach and present its applications for distinctiveness-guided shape retrieval, sampling, and view selection in 3D scenes., Comment: Accepted by ACM TOG

In this paper, we develop a 3D convolutional neural network to predict the fluid intelligence from T1-weighted MRI images by adding an encoder-decoder regularization. Considering that cerebellar volume is often highly correlated to intelligence of an individual, we propose to incorporate this morphological information into the framework for fluid intelligence prediction by utilizing an encoder-decoder regularization for brain structure segmentation simultaneously. Specifically, we first train an encoder-decoder network to generate the brain segmentation mask, where the discriminative morphological feature of the brain volume can be learned. Then, we reuse the encoder path of the network as the prediction network backbone for final fluid intelligence prediction by adding an additional regression part to predict the fluid intelligence value. The proposed framework is able to learn the discriminative relationship between the morphological information of brain structures and the intelligence score for more accurate prediction.

While making a tremendous impact in various fields, deep neural networks usually require large amounts of labeled data for training which are expensive to collect in many applications, especially in the medical domain. Unlabeled data, on the other hand, is much more abundant. Semi-supervised learning techniques, such as co-training, could provide a powerful tool to leverage unlabeled data. In this paper, we propose a novel framework, uncertainty-aware multi-view co-training (UMCT), to address semi-supervised learning on 3D data, such as volumetric data from medical imaging. In our work, co-training is achieved by exploiting multi-viewpoint consistency of 3D data. We generate different views by rotating or permuting the 3D data and utilize asymmetrical 3D kernels to encourage diversified features in different sub-networks. In addition, we propose an uncertainty-weighted label fusion mechanism to estimate the reliability of each view's prediction with Bayesian deep learning. As one view requires the supervision from other views in co-training, our self-adaptive approach computes a confidence score for the prediction of each unlabeled sample in order to assign a reliable pseudo label. Thus, our approach can take advantage of unlabeled data during training. We show the effectiveness of our proposed semi-supervised method on several public datasets from medical image segmentation tasks (NIH pancreas & LiTS liver tumor dataset). Meanwhile, a fully-supervised method based on our approach achieved state-of-the-art performances on both the LiTS liver tumor segmentation and the Medical Segmentation Decathlon (MSD) challenge, demonstrating the robustness and value of our framework, even when fully supervised training is feasible., Accepted to WACV 2020

CT and MR are dominant imaging modalities in cardiovascular inspection. Segmenting the whole heart from CT and MR volumes, and parsing it into distinctive substructures are highly desired in clinic. However, traditional methods tend to be degraded by the large variances of heart and image, and also the high requirement in simultaneously distinguishing several substructures. In this paper, we start with the well-founded Fully Convolutional Network (FCN), and closely couple the FCN with 3D operators, transfer learning and deep supervision mechanism to distill 3D contextual information and attack potential difficulties in training deep neural networks. We then focus on a main concern in our enhanced FCN. As the number of substructures to be distinguished increases, the imbalance among different classes will emerge and bias the training towards major classes and therefore should be tackled seriously. Class-balanced loss function is useful in addressing the problem but at the risk of sacrificing the segmentation details. For a better trade-off, in this paper, we propose a hybrid loss which takes advantage of different kinds of loss functions to guide the training procedure to equally treat all classes, and at the same time preserve boundary details, like the branchy structure of great vessels. We verified our method on the MM-WHS Challenge 2017 datasets, which contain both CT and MR. Our hybrid loss guided model presents superior results in concurrently labeling 7 substructures of heart (ranked as second in CT segmentation Challenge). Our framework is robust and efficient on different modalities and can be extended to other volumetric segmentation tasks.

Automatic lesion segmentation in dermoscopy images is an essential step for computer-aided diagnosis of melanoma. The dermoscopy images exhibits rotational and reflectional symmetry, however, this geometric property has not been encoded in the state-of-the-art convolutional neural networks based skin lesion segmentation methods. In this paper, we present a deeply supervised rotation equivariant network for skin lesion segmentation by extending the recent group rotation equivariant network. Specifically, we propose the G-upsampling and G-projection operations to adapt the rotation equivariant classification network for our skin lesion segmentation problem. To further increase the performance, we integrate the deep supervision scheme into our proposed rotation equivariant segmentation architecture. The whole framework is equivariant to input transformations, including rotation and reflection, which improves the network efficiency and thus contributes to the segmentation performance. We extensively evaluate our method on the ISIC 2017 skin lesion challenge dataset. The experimental results show that our rotation equivariant networks consistently excel the regular counterparts with the same model complexity under different experimental settings. Our best model also outperforms the state-of-the-art challenging methods, which further demonstrate the effectiveness of our proposed deeply supervised rotation equivariant segmentation network.

Segmenting cardiovascular volumes plays a crucial role for clinical applications, especially parsing the whole heart into fine-grained structures. However, conquering fuzzy boundaries and differentiating branchy structures in cardiovascular volume images still remain a challenging task. In this paper, we propose a general and fully automatic solution for fine-grained whole heart partition. The proposed framework originates from the 3D Fully Convolutional Network, and is reinforced in the following aspects: (1) By inheriting the knowledge from a pre-trained C3D Network, our network launches with a good initialization and gains capabilities in coping with overfitting. (2) We triggered several auxiliary loss functions on shallow layers to promote gradient flow and thus alleviate the training difficulties associated with deep neural networks. (3) Considering the obvious volume imbalance among different substructures, we introduced a Multi-class Dice Similarity Coefficient based metric to efficiently balance the training for all classes. We evaluated our method on the MM-WHS Challenge 2017 datasets. Extensive experimental results demonstrated the promising performance of our method. Our framework achieves promising results across different modalities and is general to be referred in other volumetric segmentation tasks.

Segmenting ventricular structures from cardiovascular MR scan is important for quantitative evaluation of heart. Manual delineation is time-consuming and tedious and lack of reproductivity. Considering MR image quality, heart variance, spatial inconsistency and motion artifacts during scanning, it is still a non-trivial task for automatic segmentation methods. In this paper, we propose a general and fully automatic solution to concurrently segment three important ventricular structures. Rooting in the deep learning trend, our method starts from 3D Fully Convolutional Network (3D FCN). We then enhance the 3D FCN with two well-verified blocks: (1) we conduct transfer learning between a pre-trained C3D model and our 3D FCN to get good initialization and thus suppress overfitting. (2) since boosting the gradient flow in network is beneficial to promote segmentation performance, we attach several auxiliary loss functions so as to expose early layers to better supervision. Because the volume size imbalance among different ventricular structures often biases the training of our 3D FCN, to this end, we investigate the capacity of different loss functions and propose a Multi-class Dice Similarity Coefficient (mDSC) based loss function to re-weight the training for all classes. We verified our method, especially the significance of mDSC, on the Automated Cardiac Diagnosis Challenge 2017 datasets for MR image segmentation. Extensive experimental results demonstrate the promising performance of our method.

Point clouds obtained from 3D scans are typically sparse, irregular, and noisy, and required to be consolidated. In this paper, we present the first deep learning based edge-aware technique to facilitate the consolidation of point clouds. We design our network to process points grouped in local patches, and train it to learn and help consolidate points, deliberately for edges. To achieve this, we formulate a regression component to simultaneously recover 3D point coordinates and point-to-edge distances from upsampled features, and an edge-aware joint loss function to directly minimize distances from output points to 3D meshes and to edges. Compared with previous neural network based works, our consolidation is edge-aware. During the synthesis, our network can attend to the detected sharp edges and enable more accurate 3D reconstructions. Also, we trained our network on virtual scanned point clouds, demonstrated the performance of our method on both synthetic and real point clouds, presented various surface reconstruction results, and showed how our method outperforms the state-of-the-arts.

The whole slide histopathology images (WSIs) play a critical role in gastric cancer diagnosis. However, due to the large scale of WSIs and various sizes of the abnormal area, how to select informative regions and analyze them are quite challenging during the automatic diagnosis process. The multi-instance learning based on the most discriminative instances can be of great benefit for whole slide gastric image diagnosis. In this paper, we design a recalibrated multi-instance deep learning method (RMDL) to address this challenging problem. We first select the discriminative instances, and then utilize these instances to diagnose diseases based on the proposed RMDL approach. The designed RMDL network is capable of capturing instance-wise dependencies and recalibrating instance features according to the importance coefficient learned from the fused features. Furthermore, we build a large whole-slide gastric histopathology image dataset with detailed pixel-level annotations. Experimental results on the constructed gastric dataset demonstrate the significant improvement on the accuracy of our proposed framework compared with other state-of-the-art multi-instance learning methods. Moreover, our method is general and can be extended to other diagnosis tasks of different cancer types based on WSIs., Comment: Accepted at Medical Image Analysis. Code: https://github.com/EmmaW8/RMDL

Objective : False positive reduction is one of the most crucial components in an automated pulmonary nodule detection system, which plays an important role in lung cancer diagnosis and early treatment. The objective of this paper is to effectively address the challenges in this task and therefore to accurately discriminate the true nodules from a large number of candidates. Methods : We propose a novel method employing three-dimensional (3-D) convolutional neural networks (CNNs) for false positive reduction in automated pulmonary nodule detection from volumetric computed tomography (CT) scans. Compared with its 2-D counterparts, the 3-D CNNs can encode richer spatial information and extract more representative features via their hierarchical architecture trained with 3-D samples. More importantly, we further propose a simple yet effective strategy to encode multilevel contextual information to meet the challenges coming with the large variations and hard mimics of pulmonary nodules. Results : The proposed framework has been extensively validated in the LUNA16 challenge held in conjunction with ISBI 2016, where we achieved the highest competition performance metric (CPM) score in the false positive reduction track. Conclusion : Experimental results demonstrated the importance and effectiveness of integrating multilevel contextual information into 3-D CNN framework for automated pulmonary nodule detection in volumetric CT data. Significance : While our method is tailored for pulmonary nodule detection, the proposed framework is general and can be easily extended to many other 3-D object detection tasks from volumetric medical images, where the targeting objects have large variations and are accompanied by a number of hard mimics.

Automatic and accurate whole-heart and great vessel segmentation from 3D cardiac magnetic resonance (MR) images plays an important role in the computer-assisted diagnosis and treatment of cardiovascular disease. However, this task is very challenging due to ambiguous cardiac borders and large anatomical variations among different subjects. In this paper, we propose a novel densely-connected volumetric convolutional neural network, referred as DenseVoxNet, to automatically segment the cardiac and vascular structures from 3D cardiac MR images. The DenseVoxNet adopts the 3D fully convolutional architecture for effective volume-to-volume prediction. From the learning perspective, our DenseVoxNet has three compelling advantages. First, it preserves the maximum information flow between layers by a densely-connected mechanism and hence eases the network training. Second, it avoids learning redundant feature maps by encouraging feature reuse and hence requires fewer parameters to achieve high performance, which is essential for medical applications with limited training data. Third, we add auxiliary side paths to strengthen the gradient propagation and stabilize the learning process. We demonstrate the effectiveness of DenseVoxNet by comparing it with the state-of-the-art approaches from HVSMR 2016 challenge in conjunction with MICCAI, and our network achieves the best dice coefficient. We also show that our network can achieve better performance than other 3D ConvNets but with fewer parameters.