MedMNIST v2 - A large-scale lightweight benchmark for 2D and 3D biomedical image classification.

We introduce MedMNIST v2, a large-scale MNIST-like dataset collection of standardized biomedical images, including 12 datasets for 2D and 6 datasets for 3D. All images are pre-processed into a small size of 28 × 28 (2D) or 28 × 28 × 28 (3D) with the corresponding classification labels so that no background knowledge is required for users. Covering primary data modalities in biomedical images, MedMNIST v2 is designed to perform classification on lightweight 2D and 3D images with various dataset scales (from 100 to 100,000) and diverse tasks (binary/multi-class, ordinal regression, and multi-label). The resulting dataset, consisting of 708,069 2D images and 9,998 3D images in total, could support numerous research/educational purposes in biomedical image analysis, computer vision, and machine learning. We benchmark several baseline methods on MedMNIST v2, including 2D/3D neural networks and open-source/commercial AutoML tools. The data and code are publicly available at https://medmnist.com/ .

Utility of shape evolution and displacement in the classification of chronic multiple sclerosis lesions.

The accurate recognition of multiple sclerosis (MS) lesions is challenged by the high sensitivity and imperfect specificity of MRI. To examine whether longitudinal changes in volume, surface area, 3-dimensional (3D) displacement (i.e. change in lesion position), and 3D deformation (i.e. change in lesion shape) could inform on the origin of supratentorial brain lesions, we prospectively enrolled 23 patients with MS and 11 patients with small vessel disease (SVD) and performed standardized 3-T 3D brain MRI studies. Bayesian linear mixed effects regression models were constructed to evaluate associations between changes in lesion morphology and disease state. A total of 248 MS and 157 SVD lesions were studied. Individual MS lesions demonstrated significant decreases in volume < 3.75mm3 (p = 0.04), greater shifts in 3D displacement by 23.4% with increasing duration between MRI time points (p = 0.007), and greater transitions to a more non-spherical shape (p < 0.0001). If 62.2% of lesions within a given MRI study had a calculated theoretical radius > 2.49 based on deviation from a perfect 3D sphere, a 92.7% in-sample and 91.2% out-of-sample accuracy was identified for the diagnosis of MS. Longitudinal 3D shape evolution and displacement characteristics may improve lesion classification, adding to MRI techniques aimed at improving lesion specificity.

Hyperbolic Wasserstein Distance for Shape Indexing.

Shape space is an active research topic in computer vision and medical imaging fields. The distance defined in a shape space may provide a simple and refined index to represent a unique shape. This work studies the Wasserstein space and proposes a novel framework to compute the Wasserstein distance between general topological surfaces by integrating hyperbolic Ricci flow, hyperbolic harmonic map, and hyperbolic power Voronoi diagram algorithms. The resulting hyperbolic Wasserstein distance can intrinsically measure the similarity between general topological surfaces. Our proposed algorithms are theoretically rigorous and practically efficient. It has the potential to be a powerful tool for 3D shape indexing research. We tested our algorithm with human face classification and Alzheimer's disease (AD) progression tracking studies. Experimental results demonstrated that our work may provide a succinct and effective shape index.

Efficient Framework for Identifying, Locating, Detecting and Classifying MRI Brain Tumor in MRI Images.

Image processing has plays vital role in today's technological world. It can be applied in numerous application areas such as medical, remote sensing, computer vision etc. Brain tumor is caused due to formation of abnormal tissues within human brain. Therefore, it is necessary to remove affected tumor part from the brain securely. Among various medical imaging techniques Magnetic Resonance Imaging (MRI) employs a vital role to generate images of internal parts of human body. Image segmentation is one of the challenging tasks in today's medical field. An effective segmentation using MRI slices can help to identifying the tumor with its actual size and shape. To meet this requirement, a novel method called Adaptive Convex Region Contour (ACRC) algorithm is presented. Here, Support Vector Machine (SVM) is utilized for slice classification whether it is normal or abnormal. After obtaining SVM results, abnormal slices are involved in segmentation process. Since, human body is having complicated 3D anatomical structure naturally. Unfortunately, MRI slices are yields only 2Dimensional images. The actual shape of tumor cannot be clearly visualized in 2D form. Hence, transformation from 2D to 3D is essential which helps the doctors during surgery. The Rapid Mode Image Matching (RMIM) algorithm has to be followed for 3D reconstruction modeling. After building 3D model, the original volume of the tumor is estimated. The precise experimentation was implemented in MATLAB simulation environment. The obtained results are confirmed that proposed method has better accurate results compared to existing methods.

ViBrism DB: an interactive search and viewer platform for 2D/3D anatomical images of gene expression and co-expression networks.

Understanding anatomical structures and biological functions based on gene expression is critical in a systemic approach to address the complexity of the mammalian brain, where >25 000 genes are expressed in a precise manner. Co-expressed genes are thought to regulate cell type- or region-specific brain functions. Thus, well-designed data acquisition and visualization systems for profiling combinatorial gene expression in relation to anatomical structures are crucial. To this purpose, using our techniques of microtomy-based gene expression measurements and WebGL-based visualization programs, we mapped spatial expression densities of genome-wide transcripts to the 3D coordinates of mouse brains at four post-natal stages, and built a database, ViBrism DB (http://vibrism.neuroinf.jp/). With the DB platform, users can access a total of 172 022 expression maps of transcripts, including coding, non-coding and lncRNAs in the whole context of 3D magnetic resonance (MR) images. Co-expression of transcripts is represented in the image space and in topological network graphs. In situ hybridization images and anatomical area maps are browsable in the same space of 3D expression maps using a new browser-based 2D/3D viewer, BAH viewer. Created images are shareable using URLs, including scene-setting parameters. The DB has multiple links and is expandable by community activity.

Artificial Neural Networks as a powerful numerical tool to classify specific features of a tooth based on 3D scan data.

Chairside manufacturing based on digital image acquisition is gainingincreasing importance in dentistry. For the standardized application of these methods, it is paramount to have highly automated digital workflows that can process acquired 3D image data of dental surfaces. Artificial Neural Networks (ANNs) arenumerical methods primarily used to mimic the complex networks of neural connections in the natural brain. Our hypothesis is that an ANNcan be developed that is capable of classifying dental cusps with sufficient accuracy. This bears enormous potential for an application in chairside manufacturing workflows in the dental field, as it closes the gap between digital acquisition of dental geometries and modern computer-aided manufacturing techniques.Three-dimensional surface scans of dental casts representing natural full dental arches were transformed to range image data. These data were processed using an automated algorithm to detect candidates for tooth cusps according to salient geometrical features. These candidates were classified following common dental terminology and used as training data for a tailored ANN.For the actual cusp feature description, two different approaches were developed and applied to the available data: The first uses the relative location of the detected cusps as input data and the second method directly takes the image information given in the range images. In addition, a combination of both was implemented and investigated.Both approaches showed high performance with correct classifications of 93.3% and 93.5%, respectively, with improvements by the combination shown to be minor.This article presents for the first time a fully automated method for the classification of teeththat could be confirmed to work with sufficient precision to exhibit the potential for its use in clinical practice,which is a prerequisite for automated computer-aided planning of prosthetic treatments with subsequent automated chairside manufacturing.

A method for improved clustering and classification of microscopy images using quantitative co-localization coefficients.

BACKGROUND: The localization of proteins to specific subcellular structures in eukaryotic cells provides important information with respect to their function. Fluorescence microscopy approaches to determine localization distribution have proved to be an essential tool in the characterization of unknown proteins, and are now particularly pertinent as a result of the wide availability of fluorescently-tagged constructs and antibodies. However, there are currently very few image analysis options able to effectively discriminate proteins with apparently similar distributions in cells, despite this information being important for protein characterization. FINDINGS: We have developed a novel method for combining two existing image analysis approaches, which results in highly efficient and accurate discrimination of proteins with seemingly similar distributions. We have combined image texture-based analysis with quantitative co-localization coefficients, a method that has traditionally only been used to study the spatial overlap between two populations of molecules. Here we describe and present a novel application for quantitative co-localization, as applied to the study of Rab family small GTP binding proteins localizing to the endomembrane system of cultured cells. CONCLUSIONS: We show how quantitative co-localization can be used alongside texture feature analysis, resulting in improved clustering of microscopy images. The use of co-localization as an additional clustering parameter is non-biased and highly applicable to high-throughput image data sets.

The Fisher-Markov selector: fast selecting maximally separable feature subset for multiclass classification with applications to high-dimensional data.

Selecting features for multiclass classification is a critically important task for pattern recognition and machine learning applications. Especially challenging is selecting an optimal subset of features from high-dimensional data, which typically have many more variables than observations and contain significant noise, missing components, or outliers. Existing methods either cannot handle high-dimensional data efficiently or scalably, or can only obtain local optimum instead of global optimum. Toward the selection of the globally optimal subset of features efficiently, we introduce a new selector--which we call the Fisher-Markov selector--to identify those features that are the most useful in describing essential differences among the possible groups. In particular, in this paper we present a way to represent essential discriminating characteristics together with the sparsity as an optimization objective. With properly identified measures for the sparseness and discriminativeness in possibly high-dimensional settings, we take a systematic approach for optimizing the measures to choose the best feature subset. We use Markov random field optimization techniques to solve the formulated objective functions for simultaneous feature selection. Our results are noncombinatorial, and they can achieve the exact global optimum of the objective function for some special kernels. The method is fast; in particular, it can be linear in the number of features and quadratic in the number of observations. We apply our procedure to a variety of real-world data, including mid--dimensional optical handwritten digit data set and high-dimensional microarray gene expression data sets. The effectiveness of our method is confirmed by experimental results. In pattern recognition and from a model selection viewpoint, our procedure says that it is possible to select the most discriminating subset of variables by solving a very simple unconstrained objective function which in fact can be obtained with an explicit expression.

A standardized measurement technique may improve the reliability of measurements of endometrial thickness and volume.

OBJECTIVE: To investigate whether standardization of the multiplanar view (SMV) when evaluating the uterus using three-dimensional ultrasonography (3D-US) improves intra- and interobserver reliability and agreement with regard to endometrial measurement. METHODS: Two-dimensional (2D) and 3D-US was used to measure endometrial thickness by two observers in 30 women undergoing assisted reproduction treatment. Endometrial volume was measured with Virtual Organ Computer-aided AnaLysis (VOCAL(™)) in the longitudinal (A) and coronal (C) planes using an unmodified multiplanar view (UMV) and a standardized multiplanar view (SMV). Measurement reliability was evaluated by intraclass correlation coefficient (ICC) and agreement was examined using Bland-Altman plots with limits of agreement (LoA). The ease of outlining the endometrial-myometrial interface was compared between the A- and C-planes using subjective assessment. RESULTS: Endometrial volume measurements using the SMV and A-plane were more reliable (intra- and interobserver ICCs, 0.979 and 0.975, respectively) than were measurements of endometrial thickness using 2D-US (intra- and interobserver ICCs, 0.742 and 0.702, respectively) or 3D-US (intra- and interobserver ICCs, 0.890 and 0.784, respectively). The LoAs were narrower for SMV than for UMV. Reliability and agreement were not much different between the A- and C-planes. However the observers agreed that delineating the endometrial-myometrial interface using the A-plane was easier (first and second observer, 50.0 and 46.7%, respectively) or 'comparable' (50 and 53.3%, respectively), but never more difficult than using the C-plane. CONCLUSIONS: Endometrial volume measurements are more reliable than endometrial thickness measurements and are best performed using SMV and the A-plane.

How best to measure the levator hiatus: evidence for the non-Euclidean nature of the 'plane of minimal dimensions'.

OBJECTIVE: To clarify whether the 'plane of minimal dimensions' of the levator hiatus on three-dimensional (3D) ultrasound accurately represents the minimal anatomical transverse hiatal dimension during a Valsalva maneuver. METHODS: In this retrospective study of 3D transperineal ultrasound and magnetic resonance (MR) imaging, datasets from 19 female participants were used to measure the transverse diameter of the levator hiatus using the plane of minimal dimensions on maximum Valsalva maneuver. The term 'apparent minimal transverse diameter' (aMTD) was used to define the transverse diameter measured using axial ultrasound and comparable axial or coronal MR images. Coronal MR images, using the plane of the vagina as a reference, were also obtained on maximum Valsalva. The transverse diameter measured between the caudal margin of the pubococcygeus/puborectalis on the MR coronal image is denoted by the term 'true minimal transverse diameter' (tMTD). Statistical comparisons between the aMTD and tMTD were made using Student's t-test. RESULTS: No significant difference was demonstrated between the aMTD as measured by ultrasonography and MRI. However, there were significant differences found between the aMTD measured by both ultrasound and MRI and the tMTD measured on coronal MR images (both P < 0.01), with mean ( ± SD) values of 4.36 ± 0.85, 4.13 ± 1.09 and 3.23 ± 0.49 cm, respectively. CONCLUSION: This study highlights the complexity and 3D nature of the levator hiatus and pelvic floor muscles. Investigators have previously assumed that the plane of minimal dimensions of the hiatus can be measured in a flat plane, however, the 3D nature of the hiatus means that the true levator hiatus occupies a warped (non-Euclidean) plane. Hiatal measurements on Valsalva may be subject to systematic error if performed in a single section, i.e. using a flat (Euclidean) plane.

Applications of mental rotation figures of the Shepard and Metzler type and description of a mental rotation stimulus library.

The 3D cube figures used by Shepard and Metzler [Shepard, R. N., & Metzler, J. (1971). Mental rotation of three-dimensional objects. Science, 171, 701-703] have been applied in a broad range of studies on mental rotation. This note provides a brief background on these figures, their general use in cognitive psychology and their role in studying spatial behavior. In particular, it is pointed out that large sex differences with the 3D mental rotation figures tend to be observed only in particular tasks, such as the Vandenberg and Kuse test [Vandenberg, S. G., & Kuse, A. R. (1978). Mental rotations, a group test of three-dimensional spatial visualization. Perceptual and Motor Skills, 47, 599-604] that involve multiple figures within a single problem. In contrast, pairwise presentation of the same 3D figures yields either small or no significant sex differences. In the context of the very broad range of ongoing research done with 3D figures, and the desirability of uniformity in the stimulus material used, we introduce a library of 16 cube mental rotation figures, each presented in orientations ranging from 0 to 360 degr in 5 degr steps, and with its mirror image, for a total of 2336 figures. This library, freely available to researchers, will help in the creation of mental rotation tasks both for presentation on the computer screen and for pencil and paper applications.

Classification of heterogeneous electron microscopic projections into homogeneous subsets.

The co-existence of different states of a macromolecular complex in samples used by three-dimensional electron microscopy (3D-EM) constitutes a serious challenge. The single particle method applied directly to such heterogeneous sets is unable to provide useful information about the encountered conformational diversity and produces reconstructions with severely reduced resolution. One approach to solving this problem is to partition heterogeneous projection set into homogeneous components and apply existing reconstruction techniques to each of them. Due to the nature of the projection images and the high noise level present in them, this classification task is difficult. A method is presented to achieve the desired classification by using a novel image similarity measure and solving the corresponding optimization problem. Unlike the majority of competing approaches, the presented method employs unsupervised classification (it does not require any prior knowledge about the objects being classified) and does not involve a 3D reconstruction procedure. We demonstrate a fast implementation of this method, capable of classifying projection sets that originate from 3D-EM. The method's performance is evaluated on synthetically generated data sets produced by projecting 3D objects that resemble biological structures.

Classification of continuous diffraction patterns: a numerical study.

The very intense and short pulses of future X-ray free electron lasers may allow the atomic resolution imaging of small, non-periodic objects. Preliminary estimates show that images obtained from single pulses do not contain statistically enough photons to allow successful reconstruction. Therefore multiple exposures of randomly oriented identical replicas have to be taken and the individual images have to be classified according to the object's orientation. The classification has been analytically treated by Huldt et al. [Huldt, G., Szoke, A., Hajdu, J., 2003. J. Struct. Biol. 144, 219.]. In this paper we extend the analytical results with numerical model calculations. This allows us to simulate realistic situations, which we will face in real experiments. We find significant deviations from the analytical expectations, even in the ideal case of spherical particles with random atomic distributions. We introduce a new norm for the individual scattering patterns and describe a criterion to select images belonging to similar orientation, which makes the classification more reliable in practice. We also discuss the effects of particle shape and size, partial orientational ordering, the measurement's resolution and the charge error caused by the Coulomb explosion.

Applying spatial distribution analysis techniques to classification of 3D medical images.

OBJECTIVE: The objective of this paper is to classify 3D medical images by analyzing spatial distributions to model and characterize the arrangement of the regions of interest (ROIs) in 3D space. METHODS AND MATERIAL: Two methods are proposed for facilitating such classification. The first method uses measures of similarity, such as the Mahalanobis distance and the Kullback-Leibler (KL) divergence, to compute the difference between spatial probability distributions of ROIs in an image of a new subject and each of the considered classes represented by historical data (e.g., normal versus disease class). A new subject is predicted to belong to the class corresponding to the most similar dataset. The second method employs the maximum likelihood (ML) principle to predict the class that most likely produced the dataset of the new subject. RESULTS: The proposed methods have been experimentally evaluated on three datasets: synthetic data (mixtures of Gaussian distributions), realistic lesion-deficit data (generated by a simulator conforming to a clinical study), and functional MRI activation data obtained from a study designed to explore neuroanatomical correlates of semantic processing in Alzheimer's disease (AD). CONCLUSION: Performed experiments demonstrated that the approaches based on the KL divergence and the ML method provide superior accuracy compared to the Mahalanobis distance. The later technique could still be a method of choice when the distributions differ significantly, since it is faster and less complex. The obtained classification accuracy with errors smaller than 1% supports that useful diagnosis assistance could be achieved assuming sufficiently informative historic data and sufficient information on the new subject.

Boosting accuracy of automated classification of fluorescence microscope images for location proteomics.

BACKGROUND: Detailed knowledge of the subcellular location of each expressed protein is critical to a full understanding of its function. Fluorescence microscopy, in combination with methods for fluorescent tagging, is the most suitable current method for proteome-wide determination of subcellular location. Previous work has shown that neural network classifiers can distinguish all major protein subcellular location patterns in both 2D and 3D fluorescence microscope images. Building on these results, we evaluate here new classifiers and features to improve the recognition of protein subcellular location patterns in both 2D and 3D fluorescence microscope images. RESULTS: We report here a thorough comparison of the performance on this problem of eight different state-of-the-art classification methods, including neural networks, support vector machines with linear, polynomial, radial basis, and exponential radial basis kernel functions, and ensemble methods such as AdaBoost, Bagging, and Mixtures-of-Experts. Ten-fold cross validation was used to evaluate each classifier with various parameters on different Subcellular Location Feature sets representing both 2D and 3D fluorescence microscope images, including new feature sets incorporating features derived from Gabor and Daubechies wavelet transforms. After optimal parameters were chosen for each of the eight classifiers, optimal majority-voting ensemble classifiers were formed for each feature set. Comparison of results for each image for all eight classifiers permits estimation of the lower bound classification error rate for each subcellular pattern, which we interpret to reflect the fraction of cells whose patterns are distorted by mitosis, cell death or acquisition errors. Overall, we obtained statistically significant improvements in classification accuracy over the best previously published results, with the overall error rate being reduced by one-third to one-half and with the average accuracy for single 2D images being higher than 90% for the first time. In particular, the classification accuracy for the easily confused endomembrane compartments (endoplasmic reticulum, Golgi, endosomes, lysosomes) was improved by 5-15%. We achieved further improvements when classification was conducted on image sets rather than on individual cell images. CONCLUSIONS: The availability of accurate, fast, automated classification systems for protein location patterns in conjunction with high throughput fluorescence microscope imaging techniques enables a new subfield of proteomics, location proteomics. The accuracy and sensitivity of this approach represents an important alternative to low-resolution assignments by curation or sequence-based prediction.

La imagen digital, una mirada interna / The digital image, an internal look

Se realiza una amplia revisión a fin de documentarnos sobre los aspectos básicos referentes a la imagen digital. Se abordan antecedentes históricos, clasificación, procesos implicados en la puesta a punto de imágenes tales como la compresión y sus diferentes tipos, así como la optimización, entre otros; por otra parte, se sugieren herramientas de diseño para llevar a cabo un buen tratamiento de la imagen, ya sea dentro o fuera de la red, a fin de lograr mejor calidad en el resultado de las mismas(AU)

Three-dimensional tracking of coronary arteries from biplane angiographic sequences using parametrically deformable models.

A new method for coronary artery tracking in biplane digital subtraction is presented. The dynamic tracking of nonrigid objects from two views is achieved using a generalization of parametrically deformable models. Three-dimensional (3-D) Fourier descriptors used for shape representation are obtained from the two-dimensional (2-D) descriptors of the projections. A new constraint inferred from epipolar geometry is applied to the contour model. Direct 3-D tracking is compared with the classical approach in two steps: independent 2-D tracking in each of the two projection planes; 3-D reconstruction using the epipolar constraint. Convergence quality and accuracy of the 3-D reconstruction are analyzed for several sequences showing different displacement amplitudes, deformation rates and image contrasts.