Federated Learning With Privacy-Preserving Ensemble Attention Distillation.

Federated Learning (FL) is a machine learning paradigm where many local nodes collaboratively train a central model while keeping the training data decentralized. This is particularly relevant for clinical applications since patient data are usually not allowed to be transferred out of medical facilities, leading to the need for FL. Existing FL methods typically share model parameters or employ co-distillation to address the issue of unbalanced data distribution. However, they also require numerous rounds of synchronized communication and, more importantly, suffer from a privacy leakage risk. We propose a privacy-preserving FL framework leveraging unlabeled public data for one-way offline knowledge distillation in this work. The central model is learned from local knowledge via ensemble attention distillation. Our technique uses decentralized and heterogeneous local data like existing FL approaches, but more importantly, it significantly reduces the risk of privacy leakage. We demonstrate that our method achieves very competitive performance with more robust privacy preservation based on extensive experiments on image classification, segmentation, and reconstruction tasks.

Pyramid Convolutional RNN for MRI Image Reconstruction.

Fast and accurate MRI image reconstruction from undersampled data is crucial in clinical practice. Deep learning based reconstruction methods have shown promising advances in recent years. However, recovering fine details from undersampled data is still challenging. In this paper, we introduce a novel deep learning based method, Pyramid Convolutional RNN (PC-RNN), to reconstruct images from multiple scales. Based on the formulation of MRI reconstruction as an inverse problem, we design the PC-RNN model with three convolutional RNN (ConvRNN) modules to iteratively learn the features in multiple scales. Each ConvRNN module reconstructs images at different scales and the reconstructed images are combined by a final CNN module in a pyramid fashion. The multi-scale ConvRNN modules learn a coarse-to-fine image reconstruction. Unlike other common reconstruction methods for parallel imaging, PC-RNN does not employ coil sensitive maps for multi-coil data and directly model the multiple coils as multi-channel inputs. The coil compression technique is applied to standardize data with various coil numbers, leading to more efficient training. We evaluate our model on the fastMRI knee and brain datasets and the results show that the proposed model outperforms other methods and can recover more details. The proposed method is one of the winner solutions in the 2019 fastMRI competition.

Learning Hierarchical Attention for Weakly-Supervised Chest X-Ray Abnormality Localization and Diagnosis.

We consider the problem of abnormality localization for clinical applications. While deep learning has driven much recent progress in medical imaging, many clinical challenges are not fully addressed, limiting its broader usage. While recent methods report high diagnostic accuracies, physicians have concerns trusting these algorithm results for diagnostic decision-making purposes because of a general lack of algorithm decision reasoning and interpretability. One potential way to address this problem is to further train these models to localize abnormalities in addition to just classifying them. However, doing this accurately will require a large amount of disease localization annotations by clinical experts, a task that is prohibitively expensive to accomplish for most applications. In this work, we take a step towards addressing these issues by means of a new attention-driven weakly supervised algorithm comprising a hierarchical attention mining framework that unifies activation- and gradient-based visual attention in a holistic manner. Our key algorithmic innovations include the design of explicit ordinal attention constraints, enabling principled model training in a weakly-supervised fashion, while also facilitating the generation of visual-attention-driven model explanations by means of localization cues. On two large-scale chest X-ray datasets (NIH ChestX-ray14 and CheXpert), we demonstrate significant localization performance improvements over the current state of the art while also achieving competitive classification performance.

Towards Contactless Patient Positioning.

The ongoing COVID-19 pandemic, caused by the highly contagious SARS-CoV-2 virus, has overwhelmed healthcare systems worldwide, putting medical professionals at a high risk of getting infected themselves due to a global shortage of personal protective equipment. This has in-turn led to understaffed hospitals unable to handle new patient influx. To help alleviate these problems, we design and develop a contactless patient positioning system that can enable scanning patients in a completely remote and contactless fashion. Our key design objective is to reduce the physical contact time with a patient as much as possible, which we achieve with our contactless workflow. Our system comprises automated calibration, positioning, and multi-view synthesis components that enable patient scan without physical proximity. Our calibration routine ensures system calibration at all times and can be executed without any manual intervention. Our patient positioning routine comprises a novel robust dynamic fusion (RDF) algorithm for accurate 3D patient body modeling. With its multi-modal inference capability, RDF can be trained once and used across different applications (without re-training) having various sensor choices, a key feature to enable system deployment at scale. Our multi-view synthesizer ensures multi-view positioning visualization for the technician to verify positioning accuracy prior to initiating the patient scan. We conduct extensive experiments with publicly available and proprietary datasets to demonstrate efficacy. Our system has already been used, and had a positive impact on, hospitals and technicians on the front lines of the COVID-19 pandemic, and we expect to see its use increase substantially globally.

Vessel tree tracking in angiographic sequences.

We present a method to track vessels in angiography [contrast filled vessels in two-dimensional (2-D) x-ray fluoroscopy]. Finding correspondence of a vessel tree from consecutive angiogram frames provides significant value in computer-aided clinical applications such as fast vessel tree segmentation, three-dimensional (3-D) vessel topology reconstruction from corresponding centerlines, cardiac motion understanding, etc. However, establishing an accurate vessel tree correspondence (vessel tree tracking) is a nontrivial problem due to nonlinear periodic cardiac and breathing motion in 2-D views, foreshortening, false bifurcations due to 3-D to 2-D projection, occlusion from other anatomies, etc. The vessel tree is represented by BSpline curves. The control points of the BSpline curves are landmarks that are the tracking targets. Our method maximizes the appearance similarity while preserving the vessel structure. A directed acyclic graph (DAG) is employed to represent the appearance and shape structure of the vessel tree: nodes from the DAG encode the appearance of the vessel tree landmarks, and the edges encode the relative locations between landmarks. The vessel tree tracking problem turns into finding the most similar tree from the DAG in the next frame, and it is solved using an efficient dynamic programming algorithm. We performed evaluations on 62 x-ray angiography sequences (above 1000 frames). Experiment results show our algorithm is robust to these challenges and delivers better performance, compared to four existing methods.

Automatic Tissue Differentiation Based on Confocal Endomicroscopic Images for Intraoperative Guidance in Neurosurgery.

Diagnosis of tumor and definition of tumor borders intraoperatively using fast histopathology is often not sufficiently informative primarily due to tissue architecture alteration during sample preparation step. Confocal laser microscopy (CLE) provides microscopic information of tissue in real-time on cellular and subcellular levels, where tissue characterization is possible. One major challenge is to categorize these images reliably during the surgery as quickly as possible. To address this, we propose an automated tissue differentiation algorithm based on the machine learning concept. During a training phase, a large number of image frames with known tissue types are analyzed and the most discriminant image-based signatures for various tissue types are identified. During the procedure, the algorithm uses the learnt image features to assign a proper tissue type to the acquired image frame. We have verified this method on the example of two types of brain tumors: glioblastoma and meningioma. The algorithm was trained using 117 image sequences containing over 27 thousand images captured from more than 20 patients. We achieved an average cross validation accuracy of better than 83%. We believe this algorithm could be a useful component to an intraoperative pathology system for guiding the resection procedure based on cellular level information.

Context region discovery for automatic motion compensation in fluoroscopy.

PURPOSE: Image-based tracking for motion compensation is an important topic in image-guided interventions, as it enables physicians to operate in a less complex space. In this paper, we propose an automatic motion compensation scheme to boost image guidence power in transcatheter aortic valve implantation (TAVI). METHODS: The proposed tracking algorithm automatically discovers reliable regions that correlate strongly with the target. These discovered regions can assist to estimate target motion under severe occlusion, even if target tracker fails. RESULTS: We evaluate the proposed method for pigtail tracking during TAVI. We obtain significant improvement (12 %) over the baseline in a clinical dataset. Calcification regions are automatically discovered during tracking, which would aid TAVI processes. CONCLUSION: In this work, we open a new paradigm to provide dynamic real-time guidance for TAVI without user interventions, specially in case of severe occlusion where conventional tracking methods are challenged.

Model-guided extraction of coronary vessel structures in 2D X-ray angiograms.

Analysis of vessel structures in 2D X-ray angiograms is important for pre-operative evaluation and image-guided intervention. However, automated vessel segmentation in angiograms, especially extraction of the topology such as bifurcations and vessel crossings, remains challenging mainly due to the projective nature of angiography and background clutter. In this paper, a novel framework for model-guided coronary vessel extraction in 2D angiograms is presented. In this framework, a graph is constructed using a sparse set of pixels in the angiogram. With a single user-supplied click as the starting point, the vessel tree structure in the angiogram is automatically extracted from the graph. Ambiguities in this tree structure caused by 3D-to-2D projection are then resolved using topological information from the 3D vessel model of the same patient. By incorporating this prior shape information, the proposed method is effective in extraction of vessel topology, and is robust to background clutter and uneven illumination. Through quantitative evaluation on 20 angiograms, it is shown that this model-guided approach significantly improves detection of vessel structures and bifurcations.

Estimating a patient surface model for optimizing the medical scanning workflow.

In this paper, we present the idea of equipping a tomographic medical scanner with a range imaging device (e.g. a 3D camera) to improve the current scanning workflow. A novel technical approach is proposed to robustly estimate patient surface geometry by a single snapshot from the camera. Leveraging the information of the patient surface geometry can provide significant clinical benefits, including automation of the scan, motion compensation for better image quality, sanity check of patient movement, augmented reality for guidance, patient specific dose optimization, and more. Our approach overcomes the technical difficulties resulting from suboptimal camera placement due to practical considerations. Experimental results on more than 30 patients from a real CT scanner demonstrate the robustness of our approach.

Image-based Co-Registration of Angiography and Intravascular Ultrasound Images.

In image-guided cardiac interventions, X-ray imaging and intravascular ultrasound (IVUS) imaging are two often used modalities. Interventional X-ray images, including angiography and fluoroscopy, are used to assess the lumen of the coronary arteries and to monitor devices in real time. IVUS provides rich intravascular information, such as vessel wall composition, plaque, and stent expansions, but lacks spatial orientations. Since the two imaging modalities are complementary to each other, it is highly desirable to co-register the two modalities to provide a comprehensive picture of the coronaries for interventional cardiologists. In this paper, we present a solution for co-registering 2-D angiography and IVUS through image-based device tracking. The presented framework includes learning-based vessel detection and device detections, model-based tracking, and geodesic distance-based registration. The system first interactively detects the coronary branch under investigation in a reference angiography image. During the pullback of the IVUS transducers, the system acquires both ECG-triggered fluoroscopy and IVUS images, and automatically tracks the position of the medical devices in fluoroscopy. The localization of tracked IVUS transducers and guiding catheter tips is used to associate an IVUS imaging plane to a corresponding location on the vessel branch under investigation. The presented image-based solution can be conveniently integrated into existing cardiology workflow. The system is validated with a set of clinical cases, and achieves good accuracy and robustness.

Real time assistance for stent positioning and assessment by self-initialized tracking.

Detailed visualization of stents during their positioning and deployment is critical for the success of an interventional procedure. This paper presents a novel method that relies on balloon markers to enable real-time enhanced visualization and assessment of the stent positioning and expansion, together with the blood flow over the lesion area. The key novelty is an automatic tracking framework that includes a self-initialization phase based on the Viterbi algorithm and an online tracking phase implementing the Bayesian fusion of multiple cues. The resulting motion compensation stabilizes the image of the stent and by compounding multiple frames we obtain a much better stent contrast. Robust results are obtained from more than 350 clinical data sets.

Ultrasound and fluoroscopic images fusion by autonomous ultrasound probe detection.

New minimal-invasive interventions such as transcatheter valve procedures exploit multiple imaging modalities to guide tools (fluoroscopy) and visualize soft tissue (transesophageal echocardiography (TEE)). Currently, these complementary modalities are visualized in separate coordinate systems and on separate monitors creating a challenging clinical workflow. This paper proposes a novel framework for fusing TEE and fluoroscopy by detecting the pose of the TEE probe in the fluoroscopic image. Probe pose detection is challenging in fluoroscopy and conventional computer vision techniques are not well suited. Current research requires manual initialization or the addition of fiducials. The main contribution of this paper is autonomous six DoF pose detection by combining discriminative learning techniques with a fast binary template library. The pose estimation problem is reformulated to incrementally detect pose parameters by exploiting natural invariances in the image. The theoretical contribution of this paper is validated on synthetic, phantom and in vivo data. The practical application of this technique is supported by accurate results (< 5 mm in-plane error) and computation time of 0.5s.

Image-based device tracking for the co-registration of angiography and intravascular ultrasound images.

The accurate and robust tracking of catheters and transducers employed during image-guided coronary intervention is critical to improve the clinical workflow and procedure outcome. Image-based device detection and tracking methods are preferred due to the straightforward integration into existing medical equipments. In this paper, we present a novel computational framework for image-based device detection and tracking applied to the co-registration of angiography and intravascular ultrasound (IVUS), two modalities commonly used in interventional cardiology. The proposed system includes learning-based detections, model-based tracking, and registration using the geodesic distance. The system receives as input the selection of the coronary branch under investigation in a reference angiography image. During the subsequent pullback of the IVUS transducers, the system automatically tracks the position of the medical devices, including the IVUS transducers and guiding catheter tips, under fluoroscopy imaging. The localization of IVUS transducers and guiding catheter tips is used to continuously associate an IVUS imaging plane to the vessel branch under investigation. We validated the system on a set of 65 clinical cases, with high accuracy (mean errors less than 1.5mm) and robustness (98.46% success rate). To our knowledge, this is the first reported system able to automatically establish a robust correspondence between the angiography and IVUS images, thus providing clinicians with a comprehensive view of the coronaries.

Robust and fast contrast inflow detection for 2D X-ray fluoroscopy.

2D X-ray fluoroscopy is widely used in computer assisted and image guided interventions because of the real time visual guidance it can provide to the physicians. During cardiac interventions, acquisitions of angiography are often used to assist the physician in visualizing the blood vessel structures, guide wires, or catheters, localizing bifurcations, estimating severity of a lesion, or observing the blood flow. Computational algorithms often need to process differently to frames with or without contrast medium. In order to automate this process and streamline the clinical workflow, a fully automatic contrast inflow detection algorithm is proposed. The robustness of the algorithm is validated by more than 1300 real fluoroscopic scenes. The algorithm is computationally efficient; a sequence with 100 frames can be processed within a second.

Combined cardiac and respiratory motion compensation for atrial fibrillation ablation procedures.

Catheter ablation of atrial fibrillation has become an accepted treatment option if a patient no longer responds to or tolerates drug therapy. A main goal is the electrical isolation of the pulmonary veins attached to the left atrium. Catheter ablation may be performed under fluoroscopic image guidance. Due to the rather low soft-tissue contrast of X-ray imaging, the heart is not visible in these images. To overcome this problem, overlay images from pre-operative 3-D volumetric data can be used to add anatomical detail. Unfortunately, this overlay is compromised by respiratory and cardiac motion. In the past, two methods have been proposed to perform motion compensation. The first approach involves tracking of a circumferential mapping catheter placed at an ostium of a pulmonary vein. The second method relies on a motion estimate obtained by localizing an electrode of the coronary sinus (CS) catheter. We propose a new motion compensation scheme which combines these two methods. The effectiveness of the proposed method is verified using 19 real clinical data sets. The motion in the fluoroscopic images was estimated with an overall average error of 0.55 mm by tracking the circumferential mapping catheter. By applying an algorithm involving both the CS catheter and the circumferential mapping catheter, we were able to detect motion of the mapping catheter from one pulmonary vein to another with a false positive rate of 5.8 %.

Graph based interactive detection of curve structures in 2D fluoroscopy.

An accurate and robust method to detect curve structures, such as a vessel branch or a guidewire, is essential for many medical imaging applications. A fully automatic method, although highly desired, is prone to detection errors that are caused by image noise and curve-like artifacts. In this paper, we present a novel method to interactively detect a curve structure in a 2D fluoroscopy image with a minimum requirement of human corrections. In this work, a learning based method is used to detect curve segments. Based on the detected segment candidates, a graph is built to search a curve structure as the best path passing through user interactions. Furthermore, our method introduces a novel hyper-graph based optimization method to allow for imposing geometric constraints during the path searching, and to provide a smooth and quickly converged result. With minimum human interactions involved, the method can provide accurate detection results, and has been used in different applications for guidewire and vessel detections.

Dynamic layer separation for coronary DSA and enhancement in fluoroscopic sequences.

This paper presents a new technique of coronary digital subtraction angiography which separates layers of moving background structures from dynamic fluoroscopic sequences of the heart and obtains moving layers of coronary arteries. A Bayeisan framework combines dense motion estimation, uncertainty propagation and statistical fusion to achieve reliable background layer estimation and motion compensation for coronary sequences. Encouraging results have been achieved on clinically acquired coronary sequences, where the proposed method considerably improves the visibility and perceptibility of coronary arteries undergoing breathing and cardiac movements. Perceptibility improvement is significant especially for very thin vessels. Clinical benefit is expected in the context of obese patients and deep angulation, as well as in the reduction of contrast dose in normal size patients.

Total variation models for variable lighting face recognition.

In this paper, we present the logarithmic total variation (LTV) model for face recognition under varying illumination, including natural lighting conditions, where we rarely know the strength, direction, or number of light sources. The proposed LTV model has the ability to factorize a single face image and obtain the illumination invariant facial structure, which is then used for face recognition. Our model is inspired by the SQI model but has better edge-preserving ability and simpler parameter selection. The merit of this model is that neither does it require any lighting assumption nor does it need any training. The LTV model reaches very high recognition rates in the tests using both Yale and CMU PIE face databases as well as a face database containing 765 subjects under outdoor lighting conditions.

Background correction for cDNA microarray images using the TV+L1 model.

MOTIVATION: Background correction is an important preprocess in cDNA microarray data analysis. A variety of methods have been used for this purpose. However, many kinds of backgrounds, especially inhomogeneous ones, cannot be estimated correctly using any of the existing methods. In this paper, we propose the use of the TV+L1 model, which minimizes the total variation (TV) of the image subject to an L1-fidelity term, to correct background bias. We demonstrate its advantages over the existing methods by both analytically discussing its properties and numerically comparing it with morphological opening. RESULTS: Experimental results on both synthetic data and real microarray images demonstrate that the TV+L1 model gives the restored intensity that is closer to the true data than morphological opening. As a result, this method can serve an important role in the preprocessing of cDNA microarray data.