RESUMO
Deep learning based medical image segmentation is an important step within diagnosis, which relies strongly on capturing sufficient spatial context without requiring too complex models that are hard to train with limited labelled data. Training data is in particular scarce for segmenting infection regions of CT images of COVID-19 patients. Attention models help gather contextual information within deep networks and benefit semantic segmentation tasks. The recent criss-cross-attention module aims to approximate global self-attention while remaining memory and time efficient by separating horizontal and vertical self-similarity computations. However, capturing attention from all non-local locations can adversely impact the accuracy of semantic segmentation networks. We propose a new Dynamic Deformable Attention Network (DDANet) that enables a more accurate contextual information computation in a similarly efficient way. Our novel technique is based on a deformable criss-cross attention block that learns both attention coefficients and attention offsets in a continuous way. A deep U-Net (Schlemper et al., 2019) segmentation network that employs this attention mechanism is able to capture attention from pertinent non-local locations and also improves the performance on semantic segmentation tasks compared to criss-cross attention within a U-Net on a challenging COVID-19 lesion segmentation task. Our validation experiments show that the performance gain of the recursively applied dynamic deformable attention blocks comes from their ability to capture dynamic and precise attention context. Our DDANet achieves Dice scores of 73.4% and 61.3% for Ground-glass opacity and consolidation lesions for COVID-19 segmentation and improves the accuracy by 4.9% points compared to a baseline U-Net and 24.4% points compared to current state of art methods (Fan et al., 2020).
Assuntos
COVID-19 , Humanos , Processamento de Imagem Assistida por Computador , Redes Neurais de Computação , SARS-CoV-2 , Semântica , Tomografia Computadorizada por Raios XRESUMO
Deep learning-based image segmentation models rely strongly on capturing sufficient spatial context without requiring complex models that are hard to train with limited labeled data. For COVID-19 infection segmentation on CT images, training data are currently scarce. Attention models, in particular the most recent self-attention methods, have shown to help gather contextual information within deep networks and benefit semantic segmentation tasks. The recent attention-augmented convolution model aims to capture long range interactions by concatenating self-attention and convolution feature maps. This work proposes a novel attention-augmented convolution U-Net (AA-U-Net) that enables a more accurate spatial aggregation of contextual information by integrating attention-augmented convolution in the bottleneck of an encoder-decoder segmentation architecture. A deep segmentation network (U-Net) with this attention mechanism significantly improves the performance of semantic segmentation tasks on challenging COVID-19 lesion segmentation. The validation experiments show that the performance gain of the attention-augmented U-Net comes from their ability to capture dynamic and precise (wider) attention context. The AA-U-Net achieves Dice scores of 72.3% and 61.4% for ground-glass opacity and consolidation lesions for COVID-19 segmentation and improves the accuracy by 4.2% points against a baseline U-Net and 3.09% points compared to a baseline U-Net with matched parameters. Supplementary Information: The online version contains supplementary material available at 10.1007/s11760-022-02302-3.
RESUMO
Constructing a 3-D surface model from sparse-point data is a nontrivial task. Here, we report an accurate and robust approach for reconstructing a surface model of the proximal femur from sparse-point data and a dense-point distribution model (DPDM). The problem is formulated as a three-stage optimal estimation process. The first stage, affine registration, is to iteratively estimate a scale and a rigid transformation between the mean surface model of the DPDM and the sparse input points. The estimation results of the first stage are used to establish point correspondences for the second stage, statistical instantiation, which stably instantiates a surface model from the DPDM using a statistical approach. This surface model is then fed to the third stage, kernel-based deformation, which further refines the surface model. Handling outliers is achieved by consistently employing the least trimmed squares (LTS) approach with a roughly estimated outlier rate in all three stages. If an optimal value of the outlier rate is preferred, we propose a hypothesis testing procedure to automatically estimate it. We present here our validations using four experiments, which include 1) leave-one-out experiment, 2) experiment on evaluating the present approach for handling pathology, 3) experiment on evaluating the present approach for handling outliers, and 4) experiment on reconstructing surface models of seven dry cadaver femurs using clinically relevant data without noise and with noise added. Our validation results demonstrate the robust performance of the present approach in handling outliers, pathology, and noise. An average 95-percentile error of 1.7-2.3 mm was found when the present approach was used to reconstruct surface models of the cadaver femurs from sparse-point data with noise added.
Assuntos
Cabeça do Fêmur/diagnóstico por imagem , Cabeça do Fêmur/cirurgia , Modelos Biológicos , Intensificação de Imagem Radiográfica/métodos , Interpretação de Imagem Radiográfica Assistida por Computador/métodos , Cirurgia Assistida por Computador/métodos , Algoritmos , Cadáver , Simulação por Computador , Humanos , Modelos Estatísticos , Tamanho da Amostra , Distribuições Estatísticas , Tomografia Computadorizada por Raios X/métodosRESUMO
A majority of pre-operative planning and navigational guidance during computer assisted orthopaedic surgery routinely uses three-dimensional models of patient anatomy. These models enhance the surgeon's capability to decrease the invasiveness of surgical procedures and increase their accuracy and safety. A common approach for this is to use computed tomography (CT) or magnetic resonance imaging (MRI). These have the disadvantages that they are expensive and/or induce radiation to the patient. In this paper we propose a novel method to construct a patient-specific three-dimensional model that provides an appropriate intra-operative visualization without the need for a pre or intra-operative imaging. The 3D model is reconstructed by fitting a statistical deformable model to minimal sparse 3D data consisting of digitized landmarks and surface points that are obtained intra-operatively. The statistical model is constructed using Principal Component Analysis from training objects. Our deformation scheme efficiently and accurately computes a Mahalanobis distance weighted least square fit of the deformable model to the 3D data. Relaxing the Mahalanobis distance term as additional points are incorporated enables our method to handle small and large sets of digitized points efficiently. Formalizing the problem as a linear equation system helps us to provide real-time updates to the surgeons. Incorporation of M-estimator based weighting of the digitized points enables us to effectively reject outliers and compute stable models. We present here our evaluation results using leave-one-out experiments and extended validation of our method on nine dry cadaver bones.
Assuntos
Osso e Ossos/anatomia & histologia , Processamento de Imagem Assistida por Computador/métodos , Modelos Anatômicos , Algoritmos , Cadáver , Humanos , Imageamento Tridimensional , Modelos Estatísticos , UltrassomRESUMO
This paper addresses the problem of surface reconstruction from partial data consisting of digitized landmarks and surface points that are obtained intraoperatively. The surface is derived by deforming a template so that the reconstructed surface matches the digitized points. Two techniques are employed to address such an ill-posed problem. First, a patient-specific template is used, which is computed by optimally fitting a statistical deformable model to partial data. Second, the estimated patient specific template is deformed using a regression technique by carefully designing a regularization term in kernel space. The proposed method is especially useful for accurate and stable surface construction from partial data when only a small sample size of training set is available. It adapts gradually to use more information derived from the statistical shape model when larger data are available. The proposed reconstruction method has been successfully tested on femoral heads, yielding very promising results.
RESUMO
The correspondence problem is of high relevance in the construction and use of statistical models. Statistical models are used for a variety of medical application, e.g. segmentation, registration and shape analysis. In this paper, we present comparative studies in three anatomical structures of four different correspondence establishing methods. The goal in all of the presented studies is a model-based application. We have analyzed both the direct correspondence via manually selected landmarks as well as the properties of the model implied by the correspondences, in regard to compactness, generalization and specificity. The studied methods include a manually initialized subdivision surface (MSS) method and three automatic methods that optimize the object parameterization: SPHARM, MDL and the covariance determinant (DetCov) method. In all studies, DetCov and MDL showed very similar results. The model properties of DetCov and MDL were better than SPHARM and MSS. The results suggest that for modeling purposes the best of the studied correspondence method are MDL and DetCov.