Search

A new methodology, the hybrid learning system (HLS), based upon semi-supervised learning is proposed. HLS categorizes hyperspectral images into segmented regions with discriminative features using reduced training size. The technique utilizes the modified breaking ties (MBT) algorithm for active learning and unsupervised learning-based regressors, viz. multinomial logistic regression, for hyperspectral image categorization. The probabilities estimated by multinomial logistic regression for each sample helps towards improved segregation. The high dimensionality leads to a curse of dimensionality, which ultimately deteriorates the performance of remote sensing data classification, and the problem aggravates further if labeled training samples are limited. Many studies have tried to address the problem and have employed different methodologies for remote sensing data classification, such as kernelized methods, because of insensitiveness towards the utilization of large dataset information and active learning (AL) approaches (breaking ties as a representative) to choose only prominent samples for training data. The HLS methodology proposed in the current study is a combination of supervised and unsupervised training with generalized composite kernels generating posterior class probabilities for classification. In order to retrieve the best segmentation labels, we employed Markov random fields, which make use of prior labels from the output of the multinomial logistic regression. The comparison of HLS was carried out with known methodologies, using benchmark hyperspectral imaging (HI) datasets, namely "Indian Pines" and "Pavia University". Findings of this study show that the HLS yields the overall accuracy of {99.93% and 99.98%}Indian Pines and {99.14% and 99.42%}Pavia University for classification and segmentation, respectively. [ABSTRACT FROM AUTHOR]

We propose a novel unified point cloud segmentation framework called SRX Net. This framework is composed of our proposed SurRep Group module and X-Net architecture, reconstructing point cloud segmentation models from two aspects: point cloud surface representation and network architecture. The SurRep Group can model the local surface structure of the point cloud through structures such as the Ellipsoid Surface Representation and Umbrella Surface Representation, providing rich geometric information to the network. The XNet establishes an inverse encoding–decoding path to deepen the modeling of shallow semantics, solving the interference of shallow semantics on deep semantics in prediction in the U-Net architecture. It also introduces a GLFEnhancement module to enhance each point's perception of global information. As a unified framework, SRX Net can improve the semantic segmentation effects of various existing models. Extensive experimental results show that our model produces more accurate segmentation edges and significantly reduces discrete predicted points. Based on Point Transformer, our SRX-PT Net achieves state-of-the-art performance of 72.1 mIoU on S3DIS, Scannet V2, and WCS3D datasets, respectively. [Display omitted] • Proposes SRX Net framework to reconstruct segmentation models from surface representation and network architecture. • Uses ellipsoidal and umbrella surfaces in SurRep Group for robust local geometric modeling. • Designs X-Net architecture with inverse path and GLF-Enhancement to address U-Net problems. • Achieves state-of-the-art segmentation performance on S3DIS, ScannetV2, and WCS3D datasets. • Implements SRX Net on Point Transformer and PointNet++, surpassing their baseline results. [ABSTRACT FROM AUTHOR]

Featured Application: Deep learning is an effective strategy for determining skeletal and smooth muscle conditions to help clinic personnel in landmark identification, muscle site, and reliability testing using segmentation or classification via ultrasound images. Deep learning has aided in the improvement of diagnosis identification, evaluation, and the interpretation of muscle ultrasound images, which may benefit clinical personnel. Muscle ultrasound images presents challenges such as low image quality due to noise, insufficient data, and different characteristics between skeletal and smooth muscles that can affect the effectiveness of deep learning results. From 2018 to 2020, deep learning has the improved solutions used to overcome these challenges; however, deep learning solutions for ultrasound images have not been compared to the conditions and strategies used to comprehend the current state of knowledge for handling skeletal and smooth muscle ultrasound images. This study aims to look at the challenges and trends of deep learning performance, especially in regard to overcoming muscle ultrasound image problems such as low image quality, muscle movement in skeletal muscles, and muscle thickness in smooth muscles. Skeletal muscle segmentation presents difficulties due to the regular movement of muscles and resulting noise, recording data through skipped connections, and modified layers required for upsampling. In skeletal muscle classification, the problems faced are area-specific, thus making a cropping strategy useful. Furthermore, there is no need to add additional layer modifications for smooth muscle segmentation as muscle thickness is the main problem in such cases. [ABSTRACT FROM AUTHOR]

Nowadays, the proficiency of Computer‐Aided Diagnosis systems for early diagnosis of malignant nodules in baseline Computed Tomography (CT) scan of chest crucially depends on the authenticity of the segmented nodule. In this study, the authors introduce a new morphological feature called solidity radius (SR). They then employ this feature in the new segmentation framework for the automatic segmentation of nodule and the attached vessels around the seed point on the nodule, delineated by an expert. In the framework, they extract the SR and the curvature features and employ them to determine the candidate pixels of the nodule. They then use the convex‐hull image of the candidate pixels to surround the nodule area. Afterward, using the region growing on the Hessian‐based vesselness enhancement map, the attached vessels are labelled. Finally, they apply the traditional solidity feature of the segmented nodule and the pattern of the related skeleton to prune the false positive pieces. They validate the introduced approach on two datasets, including 56 and 481 CTs (containing 1205 nodules). They show the proficiency of their SR‐based approach compared to the state‐of‐the‐art methods with average Dice Similarity Coefficients of 77.98 and 77.47% for the two datasets, respectively. [ABSTRACT FROM AUTHOR]

Segmentation of microscopy images is an essential step in most experimental studies of process–structure–property relationships in advanced materials. Currently employed segmentation approaches require the user to identify and string together a sequence of algorithms (and codes) into customized workflows that need extensive tweaking and optimization (often accomplished through repeated trials) for producing the most reliable results for each set of images. Recent advances in materials characterization instruments have significantly increased the throughput and variety of microscopy images that could be generated in the efforts to document and understand the material internal structure. There is a critical need for a guiding framework for the systematic design of segmentation workflows that can eventually lead to fully automated segmentation workflows. In this work, we propose one such modular framework consisting of five sequential steps that is applicable to segmentation of a broad variety of microscopy images. Each step is designed to accomplish a specific subtask in the overall segmentation using available functions in popular software packages. Furthermore, the modular nature of the framework allows the user to explore alternate functions in each step, while systematically comparing their relative efficacies. We describe this new segmentation framework in this paper and demonstrate its value through case studies involving a variety of microstructures. [ABSTRACT FROM AUTHOR]

A new methodology, the hybrid learning system (HLS), based upon semi-supervised learning is proposed. HLS categorizes hyperspectral images into segmented regions with discriminative features using reduced training size. The technique utilizes the modified breaking ties (MBT) algorithm for active learning and unsupervised learning-based regressors, viz. multinomial logistic regression, for hyperspectral image categorization. The probabilities estimated by multinomial logistic regression for each sample helps towards improved segregation. The high dimensionality leads to a curse of dimensionality, which ultimately deteriorates the performance of remote sensing data classification, and the problem aggravates further if labeled training samples are limited. Many studies have tried to address the problem and have employed different methodologies for remote sensing data classification, such as kernelized methods, because of insensitiveness towards the utilization of large dataset information and active learning (AL) approaches (breaking ties as a representative) to choose only prominent samples for training data. The HLS methodology proposed in the current study is a combination of supervised and unsupervised training with generalized composite kernels generating posterior class probabilities for classification. In order to retrieve the best segmentation labels, we employed Markov random fields, which make use of prior labels from the output of the multinomial logistic regression. The comparison of HLS was carried out with known methodologies, using benchmark hyperspectral imaging (HI) datasets, namely “Indian Pines” and “Pavia University”. Findings of this study show that the HLS yields the overall accuracy of {99.93% and 99.98%}Indian Pines and {99.14% and 99.42%}Pavia University for classification and segmentation, respectively.

Deep learning has aided in the improvement of diagnosis identification, evaluation, and the interpretation of muscle ultrasound images, which may benefit clinical personnel. Muscle ultrasound images presents challenges such as low image quality due to noise, insufficient data, and different characteristics between skeletal and smooth muscles that can affect the effectiveness of deep learning results. From 2018 to 2020, deep learning has the improved solutions used to overcome these challenges; however, deep learning solutions for ultrasound images have not been compared to the conditions and strategies used to comprehend the current state of knowledge for handling skeletal and smooth muscle ultrasound images. This study aims to look at the challenges and trends of deep learning performance, especially in regard to overcoming muscle ultrasound image problems such as low image quality, muscle movement in skeletal muscles, and muscle thickness in smooth muscles. Skeletal muscle segmentation presents difficulties due to the regular movement of muscles and resulting noise, recording data through skipped connections, and modified layers required for upsampling. In skeletal muscle classification, the problems faced are area-specific, thus making a cropping strategy useful. Furthermore, there is no need to add additional layer modifications for smooth muscle segmentation as muscle thickness is the main problem in such cases.

The accurate segmentation of the optic disc (OD) is important in diagnosing and evaluating many retinal diseases. However, the OD boundary is unclear, making the task of automatic OD segmentation very challenging. Recently,many researchers have applied convolutional neural network (CNN) technology to the automatic segmentation of OD, and the network has been widened and deepened. It can effectively improves the accuracy of segmentation but also requires high computational complexity and large memory consumption. To overcome the above defects, we propose a segmentation framework of a lightweight cascade CNN. It consists of a designed‐to‐be‐lightweight segmentation network and a shape‐refinement network cascade, cascading a shape‐refined network behind a segmentation network to compensate for the degraded performance of the segmentation network after lightweight design. We tested our framework on three databases, DRIVE, DIARETDB1, and DRIONS‐DB, and found that its segmentation performance is slightly better than that of u‐net, and the trainable parameters are approximately 1/35 that of u‐net. After verified by the DRIVE dataset, the memory used for training and testing is only about 1/3 of u‐net. The method proposed in this paper can greatly reduce trainable parameters and computational resource consumption while guaranteeing satisfactory segmentation performance of the model. [ABSTRACT FROM AUTHOR]

Segmenting structures of interest in medical images is an important step in different tasks such as visualization, quantitative analysis, simulation, and image-guided surgery, among several other clinical applications. Numerous segmentation methods have been developed in the past three decades for extraction of anatomical or functional structures on medical imaging. Deformable models, which include the active contour models or snakes, are among the most popular methods for image segmentation combining several desirable features such as inherent connectivity and smoothness. Even though different approaches have been proposed and significant work has been dedicated to the improvement of such algorithms, there are still challenging research directions as the simultaneous extraction of multiple objects and the integration of individual techniques. This paper presents a novel open-source framework called deformable model array (DMA) for the segmentation of multiple and complex structures of interest in different imaging modalities. While most active contour algorithms can extract one region at a time, DMA allows integrating several deformable models to deal with multiple segmentation scenarios. Moreover, it is possible to consider any existing explicit deformable model formulation and even to incorporate new active contour methods, allowing to select a suitable combination in different conditions. The framework also introduces a control module that coordinates the cooperative evolution of the snakes and is able to solve interaction issues toward the segmentation goal. Thus, DMA can implement complex object and multi-object segmentations in both 2D and 3D using the contextual information derived from the model interaction. These are important features for several medical image analysis tasks in which different but related objects need to be simultaneously extracted. Experimental results on both computed tomography and magnetic resonance imaging show that the proposed framework has a wide range of applications especially in the presence of adjacent structures of interest or under intra-structure inhomogeneities giving excellent quantitative results. [ABSTRACT FROM AUTHOR]