Search

In recent years, contrastive learning has been a highly favored method for self-supervised representation learning, which significantly improves the unsupervised training of deep image models. Self-supervised learning is a subset of unsupervised learning in which the learning process is supervised by creating pseudolabels from the data themselves. Using supervised final adjustments after unsupervised pretraining is one way to take the most valuable information from a vast collection of unlabeled data and teach from a small number of labeled instances. This study aims firstly to compare contrastive learning with other traditional learning models; secondly to demonstrate by experimental studies the superiority of contrastive learning during classification; thirdly to fine-tune performance using pretrained models and appropriate hyperparameter selection; and finally to address the challenge of using contrastive learning techniques to produce data representations with semantic meaning that are independent of irrelevant factors like position, lighting, and background. Relying on contrastive techniques, the model efficiently captures meaningful representations by discerning similarities and differences between modified copies of the same image. The proposed strategy, involving unsupervised pretraining followed by supervised fine-tuning, improves the robustness, accuracy, and knowledge extraction of deep image models. The results show that even with a modest 5% of data labeled, the semisupervised model achieves an accuracy of 57.72%. However, the use of supervised learning with a contrastive approach and careful hyperparameter tuning increases accuracy to 85.43%. Further adjustment of the hyperparameters resulted in an excellent accuracy of 88.70%.

Medical image registration is an integral part of various clinical applications including image guidance, motion tracking, therapy assessment and diagnosis. We present a robust approach for mono-modal and multi-modal medical image registration. To this end, we propose the novel shape operator based local image distance (SOLID) which estimates the similarity of images by comparing their second-order curvature information. Our similarity metric is rigorously tailored to be suitable for comparing images from different medical imaging modalities or image contrasts. A critical element of our method is the extraction of local features using higher-order shape information, enabling the accurate identification and registration of smaller structures. In order to assess the efficacy of the proposed similarity metric, we have implemented a variational image registration algorithm that relies on the principle of matching the curvature information of the given images. The performance of the proposed algorithm has been evaluated against various alternative state-of-the-art variational registration algorithms. Our experiments involve mono-modal as well as multi-modal and cross-contrast co-registration tasks in a broad variety of anatomical regions. Compared to the evaluated alternative registration methods, the results indicate a very favorable accuracy, precision and robustness of the proposed SOLID method in various highly challenging registration tasks., (Creative Commons Attribution license.)

The non-uniformity of the readout circuit response in the infrared focal plane array unit detector can result in fixed pattern noise with stripe, which seriously affects the quality of the infrared images. Considering the problems of existing non-uniformity correction, such as the loss of image detail and edge blurring, a multi-scale residual network with attention mechanism is proposed for single infrared image stripe noise removal. A multi-scale feature representation module is designed to decompose the original image into varying scales to obtain more image information. The product of the direction structure similarity parameter and the Gaussian weighted Mahalanobis distance is used as the similarity metric; a channel spatial attention mechanism based on similarity (CSAS) ensures the extraction of a more discriminative channel and spatial feature. The method is employed to eliminate the stripe noise in the vertical and horizontal directions, respectively, while preserving the edge texture information of the image. The experimental results show that the proposed method outperforms four state-of-the-art methods by a large margin in terms of the qualitative and quantitative assessments. One hundred infrared images with different simulated noise intensities are applied to verify the performance of our method, and the result shows that the average peak signal-to-noise ratio and average structural similarity of the corrected image exceed 40.08 dB and 0.98, respectively.

The nature theory of Chinese herbal medicines (CHMs) is the core theory of traditional Chinese medicine (TCM). Cold-hot nature is an important part of CHM nature. It is found that the material basis of cold-hot nature is CHM ingredients. To test the scientific hypothesis that "CHMs with similar cold-hot nature should have similar material basis," we explored an intelligent method for cold-hot nature identification of CHMs based on the feature similarity of CHM ingredients in this work. Sixty one CHMs were selected for cold-hot nature identification. High performance liquid chromatography (HPLC) was used to separate the chemical ingredients of CHMs and extract the feature information of CHM ingredients. A distance metric learning algorithm was then learned to measure the similarity of HPLC fingerprints. With the learned distance metric, cold-hot nature identification scheme (CHNIS) was proposed to build an identification model to evaluate the cold-hot nature of CHMs. A number of experiments were designed to verify the effectiveness and feasibility of the proposed CHNIS model. The total identification accuracy rate of 61 CHMs is 80.3%. The performance of the proposed CHNIS algorithm outperformed that of the compared classical algorithms. The experimental results confirmed our inference that CHMs with similar cold-hot nature had similar composition of substances. The CHNIS model was proved to be effective and feasible., Competing Interests: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest., (Copyright © 2022 Wei, Jia, Kong, Ji and Wang.)

Pedicle screw insertion is considered a complex surgery among Orthopaedics surgeons. Exclusively to prevent postoperative complications associated with pedicle screw insertion, various types of image intensity registration-based navigation systems have been developed. These systems are computation-intensive, have a small capture range and have local maxima issues. On the other hand, deep learning-based techniques lack registration generalizability and have data dependency. To overcome these limitations, a patient-specific hybrid 3D-2D registration principled framework was designed to map a pedicle screw trajectory between intraoperative X-ray image and preoperative CT image. An anatomical landmark-based 3D-2D Iterative Control Point (ICP) registration was performed to register a pedicular marker pose between the X-ray images and axial preoperative CT images. The registration framework was clinically validated by generating projection images possessing an optimal match with intraoperative X-ray images at the corresponding control point registration. The effectiveness of the registered trajectory was evaluated in terms of displacement and directional errors after reprojecting its position on 2D radiographic planes. The mean Euclidean distances for the Head and Tail end of the reprojected trajectory from the actual trajectory in the AP and lateral planes were shown to be 0.6-0.8 mm and 0.5-1.6 mm, respectively. Similarly, the corresponding mean directional errors were found to be 4.90 and 20. The mean trajectory length difference between the actual and registered trajectory was shown to be 2.67 mm. The approximate time required in the intraoperative environment to axially map the marker position for a single vertebra was found to be 3 min. Utilizing the markerless registration techniques, the designed framework functions like a screw navigation tool, and assures the quality of surgery being performed by limiting the need of postoperative CT.

An in-house hybrid deformable image registration (DIR) method, which combines free-form deformation (FFD) and the viscous fluid registration method, is proposed. Its results on the planning computed tomography (CT) and the day 1 treatment cone-beam CT (CBCT) image from 68 head and neck cancer patients are compared with the results of NiftyReg, which uses B-spline FFD alone. Several similarity metrics, the target registration error (TRE) of annotated points, as well as the Dice similarity coefficient (DSC) and Hausdorff distance (HD) of the propagated organs at risk are employed to analyze their registration accuracy. According to quantitative analysis on mutual information, normalized cross-correlation, and the absolute pixel value differences, the results of the proposed DIR are more similar to the CBCT images than the NiftyReg results. Smaller TRE of the annotated points is observed in the proposed method, and the overall mean TRE for the proposed method and NiftyReg was 2.34 and 2.98 mm, respectively (p < 0.001). The mean DSC in the larynx, spinal cord, oral cavity, mandible, and parotid given by the proposed method ranged from 0.78 to 0.91, significantly higher than the NiftyReg results (ranging from 0.77 to 0.90), and the HD was significantly lower compared to NiftyReg. Furthermore, the proposed method did not suffer from unrealistic deformations as the NiftyReg did in the visual evaluation. Meanwhile, the execution time of the proposed method was much higher than NiftyReg (96.98 ± 11.88 s vs. 4.60 ± 0.49 s). In conclusion, the in-house hybrid method gave better accuracy and more stable performance than NiftyReg., (© 2022 The Authors. Journal of Applied Clinical Medical Physics published by Wiley Periodicals, LLC on behalf of The American Association of Physicists in Medicine.)

From archaeological excavations, huge quantities of material are recovered, usually in the form of fragments. Their correct interpretation and classification are laborious and time-consuming and requires measurement, analysis and comparison of several items. Basing these activities on quantitative methods that process 3D digital data from experimental measurements allows optimizing the entire restoration process, making it faster, more accurate and cheaper. The 3D point clouds, captured by the scanning process, are raw data that must be properly processed to be used in automatic systems for the analysis of archeological finds. This paper focuses on the integration of a shape feature recognizer, able to support the semantic decomposition of the ancient artifact into archaeological features, with a structured database, able to query the large amount of information extracted. Through the automatic measurement of the dimensional attributes of the various features, it is possible to facilitate the comparative analyses between archaeological artifacts and the inferences of the archaeologist and to reduce the routine work. Here, a dedicated database has been proposed, able to store the information extracted from huge quantities of archaeological material using a specific shape feature recognizer. This information is useful for making comparisons but also to improve the archaeological knowledge. The database has been implemented and used for the identification of pottery fragments and the reconstruction of archaeological vessels. Reconstruction, in particular, often requires the solution of complex problems, especially when it involves types of potsherds that cannot be treated with traditional methods.

Comparison of compounds similarity is one of the main strategies of virtual screening protocols. Both similarity and dissimilarity concepts are of great importance during the search for new active compounds. Similarity is important due to the assumption that underlies the process of searching for new drug candidates: structurally similar compounds should induce similar biological response. On the other hand, we are also interested in dissimilarity, as we usually aim to find structurally novel ligands. In the study, we compared several approaches of evaluating compound similarity. Various representations and metrics were applied and we indicated the rate of variation of the results that can occur when shifting from one strategy to another. We compared both general similarity of datasets using different approaches, as well as examined the changes in the set of nearest neighbors when changing one compound representation into another, and the influence of representation/metric settings on the clustering outcome. We hope that the study will be of great help during the preparation of virtual screening experiments, stressing the need for careful selection of the way, the compound similarity is assessed. The differences in the results that can be obtained via the application of particular strategy can significantly influence the outcome of comparison studies; therefore, its settings should be carefully selected beforerunning the comparison., (Copyright © 2020 Elsevier Ltd. All rights reserved.)

Appropriate ways to measure the similarity between single-cell RNA-sequencing (scRNA-seq) data are ubiquitous in bioinformatics, but using single clustering or classification methods to process scRNA-seq data is generally difficult. This has led to the emergence of integrated methods and tools that aim to automatically process specific problems associated with scRNA-seq data. These approaches have attracted a lot of interest in bioinformatics and related fields. In this paper, we systematically review the integrated methods and tools, highlighting the pros and cons of each approach. We not only pay particular attention to clustering and classification methods but also discuss methods that have emerged recently as powerful alternatives, including nonlinear and linear methods and descending dimension methods. Finally, we focus on clustering and classification methods for scRNA-seq data, in particular, integrated methods, and provide a comprehensive description of scRNA-seq data and download URLs., (© The Author(s) 2019. Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oup.com.)

A computer-assisted human peripheral blood leukocyte image classification method based on Siamese network is proposed. Firstly, a Siamese network with two identical convolutional neural network (CNN) sub-networks and a logistic regression for leukocyte five classification is designed, which can learn not only distinguishing features but also a similarity metric. Then for each category of the leukocytes, a typical sample is selected by the hematologist. To train the Siamese network, a leukocyte and a typical sample that belong to the same category make up a genuine pair and the leukocyte with the rest four typical samples respectively make up four impostor pairs. Obviously, the number of the genuine pairs is lesser than that of the impostor pairs. Thus, a data augmentation method suitable for leukocyte is used to enrich the amount of the genuine pairs. By training the Siamese network using the genuine pairs and impostor pairs, the Siamese network can not only shorten the similarity metric between the leukocyte and the same category of the typical sample but also increase the similarity metrics between the leukocyte and the different categories of the typical samples. Experimental results indicate that the proposed method can achieve 98.8% average testing accuracy. Graphical abstract.

Searching genomic interval sets produced by sequencing methods has been widely and routinely performed; however, existing metrics for quantifying similarities among interval sets are inconsistent. Here we introduce Seqpare, a self-consistent and effective metric of similarity and tool for comparing sequences based on their interval sets. With this metric, the similarity of two interval sets is quantified by a single index, the ratio of their effective overlap over the union: an index of zero indicates unrelated interval sets, and an index of one means that the interval sets are identical. Analysis and tests confirm the effectiveness and self-consistency of the Seqpare metric., Competing Interests: No competing interests were disclosed., (Copyright: © 2021 Feng SC et al.)

Searching genomic interval sets produced by sequencing methods has been widely and routinely performed; however, existing metrics for quantifying similarities among interval sets are inconsistent. Here we introduce Seqpare, a self-consistent and effective metric of similarity and tool for comparing sequences based on their interval sets. With this metric, the similarity of two interval sets is quantified by a single index, the ratio of their effective overlap over the union: an index of zero indicates unrelated interval sets, and an index of one means that the interval sets are identical. Analysis and tests confirm the effectiveness and self-consistency of the Seqpare metric., Competing Interests: No competing interests were disclosed., (Copyright: © 2020 Feng SC et al.)

In this paper, given data with high-dimensional features, we study this problem of how to calculate the similarity between two samples by considering feature interaction network, where a feature interaction network represents the relationship between features. This is different from some traditional methods, those of which learn similarities based on a sample network that represents the relationship between samples. Therefore, we propose a novel network-based similarity metric for computing the similarity between samples, which incorporates the knowledge of feature interaction network, in order to overcome the data sparseness problem. Our similarity metric uses a new Feature Alignment Similarity measure, which does not directly compute the similarities among samples, but projects each sample into a feature interaction network and measures the similarities between two samples using the similarities between the vertices of the samples in the network. As such, when two samples do not share any common features, they are likely to have higher similarity values when their features share the similar network regions. For ensuring that the metric is useful in a real-world application, we apply our metric to discover subtypes in tumor mutational data by incorporating the information of the gene interaction network. Our experimental results from using synthetic data and real-world tumor mutational data show that our approach outperforms the top competitors in cancer subtype discovery. Furthermore, our approach can identify cancer subtypes that cannot be detected by other clustering algorithms in real cancer data.

Advances in high-throughput sequencing on single-cell gene expressions [single-cell RNA sequencing (scRNA-seq)] have enabled transcriptome profiling on individual cells from complex samples. A common goal in scRNA-seq data analysis is to discover and characterise cell types, typically through clustering methods. The quality of the clustering therefore plays a critical role in biological discovery. While numerous clustering algorithms have been proposed for scRNA-seq data, fundamentally they all rely on a similarity metric for categorising individual cells. Although several studies have compared the performance of various clustering algorithms for scRNA-seq data, currently there is no benchmark of different similarity metrics and their influence on scRNA-seq data clustering. Here, we compared a panel of similarity metrics on clustering a collection of annotated scRNA-seq datasets. Within each dataset, a stratified subsampling procedure was applied and an array of evaluation measures was employed to assess the similarity metrics. This produced a highly reliable and reproducible consensus on their performance assessment. Overall, we found that correlation-based metrics (e.g. Pearson's correlation) outperformed distance-based metrics (e.g. Euclidean distance). To test if the use of correlation-based metrics can benefit the recently published clustering techniques for scRNA-seq data, we modified a state-of-the-art kernel-based clustering algorithm (SIMLR) using Pearson's correlation as a similarity measure and found significant performance improvement over Euclidean distance on scRNA-seq data clustering. These findings demonstrate the importance of similarity metrics in clustering scRNA-seq data and highlight Pearson's correlation as a favourable choice. Further comparison on different scRNA-seq library preparation protocols suggests that they may also affect clustering performance. Finally, the benchmarking framework is available at http://www.maths.usyd.edu.au/u/SMS/bioinformatics/software.html., (© The Author(s) 2018. Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oup.com.)

Nonrigid multimodal image registration remains a challenging task in medical image processing and analysis. The structural representation (SR)-based registration methods have attracted much attention recently. However, the existing SR methods cannot provide satisfactory registration accuracy due to the utilization of hand-designed features for structural representation. To address this problem, the structural representation method based on the improved version of the simple deep learning network named PCANet is proposed for medical image registration. In the proposed method, PCANet is firstly trained on numerous medical images to learn convolution kernels for this network. Then, a pair of input medical images to be registered is processed by the learned PCANet. The features extracted by various layers in the PCANet are fused to produce multilevel features. The structural representation images are constructed for two input images based on nonlinear transformation of these multilevel features. The Euclidean distance between structural representation images is calculated and used as the similarity metrics. The objective function defined by the similarity metrics is optimized by L-BFGS method to obtain parameters of the free-form deformation (FFD) model. Extensive experiments on simulated and real multimodal image datasets show that compared with the state-of-the-art registration methods, such as modality-independent neighborhood descriptor (MIND), normalized mutual information (NMI), Weber local descriptor (WLD), and the sum of squared differences on entropy images (ESSD), the proposed method provides better registration performance in terms of target registration error (TRE) and subjective human vision.

Similarity measurement of lung nodules is a critical component in content-based image retrieval (CBIR), which can be useful in differentiating between benign and malignant lung nodules on computer tomography (CT). This paper proposes a new two-step CBIR scheme (TSCBIR) for computer-aided diagnosis of lung nodules. Two similarity metrics, semantic relevance and visual similarity, are introduced to measure the similarity of different nodules. The first step is to search for K most similar reference ROIs for each queried ROI with the semantic relevance metric. The second step is to weight each retrieved ROI based on its visual similarity to the queried ROI. The probability is computed to predict the likelihood of the queried ROI depicting a malignant lesion. In order to verify the feasibility of the proposed algorithm, a lung nodule dataset including 366 nodule regions of interest (ROIs) is assembled from LIDC-IDRI lung images on CT scans. Three groups of texture features are implemented to represent a nodule ROI. Our experimental results on the assembled lung nodule dataset show good performance improvement over existing popular classifiers.

Purpose: The goal of medical case-based image retrieval (M-CBIR) is to assist radiologists in the clinical decision-making process by finding medical cases in large archives that most resemble a given case. Cases are described by radiology reports comprised of radiological images and textual information on the anatomy and pathology findings. The textual information, when available in standardized terminology, e.g., the RadLex ontology, and used in conjunction with the radiological images, provides a substantial advantage for M-CBIR systems., Methods: We present a new method for incorporating textual radiological findings from medical case reports in M-CBIR. The input is a database of medical cases, a query case, and the number of desired relevant cases. The output is an ordered list of the most relevant cases in the database. The method is based on a new case formulation, the Augmented RadLex Graph and an Anatomy-Pathology List. It uses a new case relatedness metric [Formula: see text] that prioritizes more specific medical terms in the RadLex tree over less specific ones and that incorporates the length of the query case., Results: An experimental study on 8 CT queries from the 2015 VISCERAL 3D Case Retrieval Challenge database consisting of 1497 volumetric CT scans shows that our method has accuracy rates of 82 and 70% on the first 10 and 30 most relevant cases, respectively, thereby outperforming six other methods., Conclusions: The increasing amount of medical imaging data acquired in clinical practice constitutes a vast database of untapped diagnostically relevant information. This paper presents a new hybrid approach to retrieving the most relevant medical cases based on textual and image information.