Representation Learning For Low-quality Medical Image Segmentation

Automatic image segmentation is an essential step for many medical image analysis applications,including computer-aided radiation therapy,disease diagnosis,and treatment effect evaluation.Segmenting target objects fast and efficiently is crucial for the doctors to improve the efficiency and quality of planning and diagnosing.As a consequence,automatic computer-aided image segmentation is becoming an increasingly important research spot in recent years.However,comparing with natural images,medicals images tend to have lower image-contrast,more intense noise,artifacts,larger memory footprint,and more difficulties in acquiring high-quality manual labels.Under these circumstances,how to design specific representation learning algorithms with respect to the properties of medical images to improve the segmentation performance is becoming one of the most important problem in the field.In this paper,we intend to eliminate the adverse effects of the aforementioned characteristics on medical image segmentation by learning better representations with deep learning and multiple kernel clustering in supervised semantic segmentation,instance segmentation,and unsupervised segmentation scenario.Our contributions are listed as follow:1.One of the major challenges for semantic medical image segmentation is the blurry nature of medical images,which can often result in low-contrast and vanishing boundaries.With the recent advances in convolutional neural networks,vast improvements have been made for image segmentation,mainly based on the skip-connection-linked encoder-decoder deep architectures.However,in many applications,these models often fail to accurately locate complex boundaries and properly segment tiny isolated parts.In this paper,we aim to provide a method for blurry medical image segmentation and argue that skip connections are not enough to help accurately locate indistinct boundaries.Accordingly,we propose a novel high-resolution multi-scale encoder-decoder network(MHMSN),in which multi-scale dense connections are introduced for the encoder-decoder structure to finely exploit comprehensive semantic information.Besides skip-connections,extra deeply-supervised high-resolution pathways are integrated to collect high-resolution semantic information for accurate boundary localization.These pathways are paired with a difficulty-guided cross-entropy loss function and a contour regression task to enhance the quality of boundary detection.Extensive experiments show the effectiveness of our method.Our experimental results also show that besides increasing the network complexity,raising the resolution of semantic feature maps can largely affect the overall model performance.2.Existing methods for semantic instance segmentation can be roughly classified into two categories: proposal-based and proposal-free methods.However,we observe that,specific to medical images,the irregular shape-variations(in body organs and cells)and dense target distributions(in nuclei and cells)makes it hard for the detection-oriented proposal-based methods to achieve robust instance localization.On the other hand,ambiguous boundaries caused by the low-contrast nature of medical images(e.g.,CT images)challenges the robustness of the indirect and pixel-centered instance information extraction mechanism in the proposal-free methods.To tackle these issues,we propose a proposal-free segmentation network with discriminative deep supervision.In the proposed network,a novel detection module,termed as discriminative deep supervision(DDS)module,is intertwined with a carefully designed proposal-free segmentation backbone.As a consequence,the features learned by the backbone network become more sensitive to instances.Extensive experiments demonstrate the effectiveness and superior performance of the proposed algorithm.An intermediate version of our proposed method won the second prize in the MICCAI 2018 nuclei segmentation competition,with only a marginal difference with the first prize winner.3.We propose a robust multiple kernel clustering algorithm to relieve the adverse effect of many kinds of noise in unsupervised medical image segmentation.Multiple kernel clustering(MKC)is an important research topic during the last few decades.It optimally combines a group of pre-specified base kernels to improve clustering performance.Though demonstrating promising performance in various applications,this task is still challenging due to lack of reliable discriminative guidance for the base kernel combination.Moreover,noise from either corrupted data or inappropriately selected base kernel parameters would undermine the intrinsic manifold and makes the problem even harder.In this paper,we integrate subspace segmentation into MKC and propose a robust subspace segmentation-based multiple kernel clustering(SS-MKC)algorithm to address these issues.In our formulation,we unify the constrained kernel polarization and subspace segmentation into a single procedure,where the resultant affinity matrix embedded with robust subspace structural information is utilized to guide the linear combination of base kernels.Extensive experiments have been conducted on both synthetic and public benchmark datasets,and the results well demonstrate the superiority of the proposed algorithm when compared with the state-of-the-art MKC methods.4.Super-pixel-based multi-view clustering is an important unsupervised image segmentation algorithm.It aims to group super-pixels into different categories by optimally exploring complementary information from multiple kernel matrices.To preserve the local geometric structure among super-pixels for better discriminative capacity of the learned representation,we propose a neighbor-kernel subspace segmentation-based multiple kernel clustering.Specifically,in our algorithm,we first define a neighbor-kernel,which can be utilized to preserve the block diagonal structure and strengthen the robustness against noise and outliers among base kernels.After that,we linearly combine these base neighbor-kernels to extract a consensus affinity matrix through an exact-rank constrained subspace segmentation.The naturally possessed block diagonal structure of neighbor-kernels better serves the subsequent subspace segmentation,and in turn,the extracted shared structure is further refined through subspace segmentation based on the combined neighbor-kernels.In this manner,the above two learning processes can be seamlessly coupled and negotiate with each other to achieve better clustering.Further,we carefully design an efficient iterative optimization algorithm with proven convergence to address the resultant optimization problem.As a by-product,we reveal an interesting insight into the exact-rank constraint in ridge regression by careful theoretical analysis: it back-projects the solution of the unconstrained counterpart to its principal components.Comprehensive experiments have been conducted on several benchmark datasets,and the results demonstrate the effectiveness of the proposed algorithm.5.To better fuse multiple segmentation results of different algorithms,we propose a multi-view spectral clustering-based algorithm by carefully exploiting the high-level affinity information among data.The existing multi-view spectral clustering methods usually linearly combine a group of pre-specified first-order Laplacian matrices to construct an optimal Laplacian matrix for results integration.However,recent literature finds that this setting may result in limited representation capability and insufficient information exploitation.In this paper,we propose a novel optimal neighborhood multi-view spectral clustering(ONMSC)algorithm to address these issues.Specifically,the proposed algorithm generates an optimal Laplacian matrix by searching the neighborhood of both the linear combination of the first-order and high-order base Laplacian matrices simultaneously.This design enhances the representative capacity of the optimal Laplacian and better utilizes the hidden high-order connection information,leading to improved clustering performance.Extensive experimental results verify the effectiveness of the proposed ONMSC.