AssemblyNet: A large ensemble of CNNs for 3D whole brain MRI segmentation.

Whole brain segmentation of fine-grained structures using deep learning (DL) is a very challenging task since the number of anatomical labels is very high compared to the number of available training images. To address this problem, previous DL methods proposed to use a single convolution neural network (CNN) or few independent CNNs. In this paper, we present a novel ensemble method based on a large number of CNNs processing different overlapping brain areas. Inspired by parliamentary decision-making systems, we propose a framework called AssemblyNet, made of two "assemblies" of U-Nets. Such a parliamentary system is capable of dealing with complex decisions, unseen problem and reaching a relevant consensus. AssemblyNet introduces sharing of knowledge among neighboring U-Nets, an "amendment" procedure made by the second assembly at higher-resolution to refine the decision taken by the first one, and a final decision obtained by majority voting. During our validation, AssemblyNet showed competitive performance compared to state-of-the-art methods such as U-Net, Joint label fusion and SLANT. Moreover, we investigated the scan-rescan consistency and the robustness to disease effects of our method. These experiences demonstrated the reliability of AssemblyNet. Finally, we showed the interest of using semi-supervised learning to improve the performance of our method.

Patch-Based Abnormality Maps for Improved Deep Learning-Based Classification of Huntington's Disease.

Deep learning techniques have demonstrated state-of-the-art performances in many medical imaging applications. These methods can efficiently learn specific patterns. An alternative approach to deep learning is patch-based grading methods, which aim to detect local similarities and differences between groups of subjects. This latter approach usually requires less training data compared to deep learning techniques. In this work, we propose two major contributions: first, we combine patch-based and deep learning methods. Second, we propose to extend the patch-based grading method to a new patch-based abnormality metric. Our method enables us to detect localized structural abnormalities in a test image by comparison to a template library consisting of images from a variety of healthy controls. We evaluate our method by comparing classification performance using different sets of features and models. Our experiments show that our novel patch-based abnormality metric increases deep learning performance from 91.3% to 95.8% of accuracy compared to standard deep learning approaches based on the MRI intensity.

SuperPatchMatch: An Algorithm for Robust Correspondences Using Superpixel Patches.

Superpixels have become very popular in many computer vision applications. Nevertheless, they remain under-exploited, since the superpixel decomposition may produce irregular and nonstable segmentation results due to the dependency to the image content. In this paper, we first introduce a novel structure, a superpixel-based patch, called SuperPatch. The proposed structure, based on superpixel neighborhood, leads to a robust descriptor, since spatial information is naturally included. The generalization of the PatchMatch method to SuperPatches, named SuperPatchMatch, is introduced. Finally, we propose a framework to perform fast segmentation and labeling from an image database, and demonstrate the potential of our approach, since we outperform, in terms of computational cost and accuracy, the results of state-of-the-art methods on both face labeling and medical image segmentation.

CERES: A new cerebellum lobule segmentation method.

The human cerebellum is involved in language, motor tasks and cognitive processes such as attention or emotional processing. Therefore, an automatic and accurate segmentation method is highly desirable to measure and understand the cerebellum role in normal and pathological brain development. In this work, we propose a patch-based multi-atlas segmentation tool called CERES (CEREbellum Segmentation) that is able to automatically parcellate the cerebellum lobules. The proposed method works with standard resolution magnetic resonance T1-weighted images and uses the Optimized PatchMatch algorithm to speed up the patch matching process. The proposed method was compared with related recent state-of-the-art methods showing competitive results in both accuracy (average DICE of 0.7729) and execution time (around 5 minutes).

An Optimized PatchMatch for multi-scale and multi-feature label fusion.

Automatic segmentation methods are important tools for quantitative analysis of Magnetic Resonance Images (MRI). Recently, patch-based label fusion approaches have demonstrated state-of-the-art segmentation accuracy. In this paper, we introduce a new patch-based label fusion framework to perform segmentation of anatomical structures. The proposed approach uses an Optimized PAtchMatch Label fusion (OPAL) strategy that drastically reduces the computation time required for the search of similar patches. The reduced computation time of OPAL opens the way for new strategies and facilitates processing on large databases. In this paper, we investigate new perspectives offered by OPAL, by introducing a new multi-scale and multi-feature framework. During our validation on hippocampus segmentation we use two datasets: young adults in the ICBM cohort and elderly adults in the EADC-ADNI dataset. For both, OPAL is compared to state-of-the-art methods. Results show that OPAL obtained the highest median Dice coefficient (89.9% for ICBM and 90.1% for EADC-ADNI). Moreover, in both cases, OPAL produced a segmentation accuracy similar to inter-expert variability. On the EADC-ADNI dataset, we compare the hippocampal volumes obtained by manual and automatic segmentation. The volumes appear to be highly correlated that enables to perform more accurate separation of pathological populations.

Optimized patchMatch for near real time and accurate label fusion.

Automatic segmentation methods are important tools for quantitative analysis of magnetic resonance images. Recently, patch-based label fusion approaches demonstrated state-of-the-art segmentation accuracy. In this paper, we introduce a new patch-based method using the PatchMatch algorithm to perform segmentation of anatomical structures. Based on an Optimized PAtchMatch Label fusion (OPAL) strategy, the proposed method provides competitive segmentation accuracy in near real time. During our validation on hippocampus segmentation of 80 healthy subjects, OPAL was compared to several state-of-the-art methods. Results show that OPAL obtained the highest median Dice coefficient (89.3%) in less than 1 sec per subject. These results highlight the excellent performance of OPAL in terms of computation time and segmentation accuracy compared to recently published methods.