In this paper, we propose and evaluate six methods for the segmentation of skin lesions in dermoscopic images. This set includes some state art techniques which have been successfully used many medical imaging problems (gradient vector flow (GVF) level method Chan et al.[(C-LS)]. It also a developed by authors were tailored to particular application (adaptive thresholding (AT), adaptive snake (AS), EM (EM-LS), fuzzy-based split-and-merge algorithm (FBSM)]. The applied 100 images evaluated...

We present 3DRegNet, a novel deep learning architecture for the registration of 3D scans. Given set point correspondences, we build neural network to address following two challenges: (i) classification correspondences into inliers/outliers, and (ii) regression motion parameters that align scans common reference frame. With regard regression, alternative approaches: Deep Neural Network (DNN) Procrustes approach using SVD estimate transformation. Our correspondence-based achieves higher...

We describe an automated methodology for the analysis of unregistered cranio-caudal (CC) and medio-lateral oblique (MLO) mammography views in order to estimate patient's risk developing breast cancer. The main innovation behind this lies use deep learning models problem jointly classifying mammogram respective segmentation maps lesions (i.e., masses micro-calcifications). This is a holistic that can classify whole mammographic exam, containing CC MLO maps, as opposed classification...

In this paper, we developed BreastScreening-AI within two scenarios for the classification of multimodal beast images: (1) Clinician-Only; and (2) Clinician-AI. The novelty relies on introduction a deep learning method into real clinical workflow medical imaging diagnosis. We attempt to address three high-level goals in above scenarios. Concretely, how clinicians: i) accept interact with these systems, revealing whether are explanations functionalities required; ii) receptive AI-assisted by...

Intelligent agents are showing increasing promise for clinical decision-making in a variety of healthcare settings. While substantial body work has contributed to the best strategies convey these agents' decisions clinicians, few have considered impact personalizing and customizing communications on clinicians' performance receptiveness. This raises question how intelligent should adapt their tone accordance with target audience. We designed two approaches communicate an agent breast cancer...

In this paper, we propose novel methods to evaluate the performance of object detection algorithms in video sequences. This procedure allows us highlight characteristics (e.g., region splitting or merging) which are specific method being used. The proposed framework compares output algorithm with ground truth and measures differences according objective metrics. way it is possible perform a fair comparison among different methods, evaluating their strengths weaknesses allowing user reliable...

We present a new supervised learning model designed for the automatic segmentation of left ventricle (LV) heart in ultrasound images. address following problems inherent to models: 1) need large set training images; 2) robustness imaging conditions not data; and 3) complex search process. The innovations our approach reside formulation that decouples rigid nonrigid detections, deep methods appearance LV, efficient derivative-based algorithms. functionality is evaluated using data diseased...

Precision medicine approaches rely on obtaining precise knowledge of the true state health an individual patient, which results from a combination their genetic risks and environmental exposures. This approach is currently limited by lack effective efficient non-invasive medical tests to define full range phenotypic variation associated with health. Such critical for improved early intervention, better treatment decisions, ameliorating steadily worsening epidemic chronic disease. We present...

We present a new statistical pattern recognition approach for the problem of left ventricle endocardium tracking in ultrasound data. The is formulated as sequential importance resampling algorithm such that expected segmentation current time step estimated based on appearance, shape, and motion models take into account all previous images contours produced by method. appearance shape decouple affine nonrigid segmentations to reduce running complexity. proposed model combines systole diastole...

Abstract Background and objectives: For the planning of surgical procedures involving bony reconstruction mandible, autologous iliac crest graft, along with fibula has become established as a preferred donor region. While computer-assisted methods are increasingly gaining importance, necessary preparation geometric data based on CT imaging remains largely manual process. The aim this work was to develop test method for automated segmentation subsequent planning. Methods: A total 1,398...

Multiplicative noise is often present in medical and biological imaging, such as magnetic resonance imaging (MRI), ultrasound, positron emission tomography (PET), single photon computed (SPECT), fluorescence microscopy. Noise reduction images a difficult task which linear filtering algorithms usually fail. Bayesian have been used with success but they are time consuming computationally demanding. In addition, the increasing importance of 3-D 4-D image analysis diagnosis procedures increases...

This paper proposes an approach for recognizing human activities (more specifically, pedestrian trajectories) in video sequences, a surveillance context. A system automatic processing of information purposes should be capable detecting, recognizing, and collecting statistics activity, reducing intervention as much possible. In the method described this paper, trajectories are modeled concatenation segments produced by set low level dynamical models. These models estimated unsupervised...

Magnetic resonance imaging (MRI) is the recommended modality in diagnosis of breast cancer. However, each MRI scan comprises dozens volumes for radiologist to inspect, providing its own set information on tissues being scanned. This paper proposes a multimodal framework that processes all available data order reach diagnosis, instead relying single volume, mimicking radiologists' workflow. The 3D convolutional neural network modality, whose predictions are then combined using late fusion...

The detection and classification of breast cancer lesions with computer-aided diagnosis systems has seen a huge boost in recent years due to deep learning. However, most works focus on 2D image modalities. Dealing 3D MRI adds new challenges, such as data insufficiency lack local annotations. To handle these issues, this work proposes two-stage framework based multiple instance learning, which requires only global labels (weak supervision) provides: 1) the whole volume each slice; 2)...

This article compares the performance of target detectors based on adaptive background differencing public benchmark data. Five state art methods are described. The is evaluated using measures with respect to ground truth. original points comparison hand labelled truth and evaluation a large database. simpler LOTS SGM more appropriate particular task as MGM complex model.