Medical image noise reduction using the Sylvester-Lyapunov equation

Involution-based efficient autoencoder for denoising histopathological images with enhanced hybrid feature extraction

Noise in histopathology images from hardware limitations, preparation artifacts, and environmental factors complicates disease analysis and increases risks. With growing workloads and the complexity of histopathology images, developing efficient and precise histopathology image analysis methods is essential. However, many denoising models struggle to extract spatial features and are computationally expensive, primarily due to the limited capacity of convolutions to capture visual patterns across spatial locations, and tend to occupy the largest share of computational costs. In histopathology, many spatial features, such as anomalies or microorganisms, located sparsely across an image, are crucial for the final diagnosis, and many denoising processes often either blur them or introduce artifacts. In this study, we propose a lightweight autoencoder (43.11 kilobytes) for denoising histopathology images by fusing a single involution layer within a small convolution model, resulting in better denoising performance in a hybrid model, which has both channel-specific and location-specific feature extraction capabilities. Building upon the idea of a shallow autoencoder, our model results in much lower memory and compute overhead requirements, while also not avoiding the generation of artifacts. On Malaria Blood Smear and CRC datasets, SSIM Loss and Peak-Signal-to-Noise-Ratio were used for performance evaluation, with lower SSIM Loss (0.058 and 0.34) in denoising images with an added Gaussian noise of 0.3. Our proposed autoencoder, with low weight parameters of 11,037 and 81,630,000 floating point operations (FLOPs), is over 20 times less computationally expensive than Xception, the second-best performing model, establishing ours as the most efficient denoising autoencoder for histopathology images.

The Venus score for the assessment of the quality and trustworthiness of biomedical datasets

Biomedical datasets are the mainstays of computational biology and health informatics projects, and can be found on multiple data platforms online or obtained from wet-lab biologists and physicians. The quality and the trustworthiness of these datasets, however, can sometimes be poor, producing bad results in turn, which can harm patients and data subjects. To address this problem, policy-makers, researchers, and consortia have proposed diverse regulations, guidelines, and scores to assess the quality and increase the reliability of datasets. Although generally useful, however, they are often incomplete and impractical. The guidelines of Datasheets for Datasets, in particular, are too numerous; the requirements of the Kaggle Dataset Usability Score focus on non-scientific requisites (for example, including a cover image); and the European Union Artificial Intelligence Act (EU AI Act) sets forth sparse and general data governance requirements, which we tailored to datasets for biomedical AI. Against this backdrop, we introduce our new Venus score to assess the data quality and trustworthiness of biomedical datasets. Our score ranges from 0 to 10 and consists of ten questions that anyone developing a bioinformatics, medical informatics, or cheminformatics dataset should answer before the release. In this study, we first describe the EU AI Act, Datasheets for Datasets, and the Kaggle Dataset Usability Score, presenting their requirements and their drawbacks. To do so, we reverse-engineer the weights of the influential Kaggle Score for the first time and report them in this study. We distill the most important data governance requirements into ten questions tailored to the biomedical domain, comprising the Venus score. We apply the Venus score to twelve datasets from multiple subdomains, including electronic health records, medical imaging, microarray and bulk RNA-seq gene expression, cheminformatics, physiologic electrogram signals, and medical text. Analyzing the results, we surface fine-grained strengths and weaknesses of popular datasets, as well as aggregate trends. Most notably, we find a widespread tendency to gloss over sources of data inaccuracy and noise, which may hinder the reliable exploitation of data and, consequently, research results. Overall, our results confirm the applicability and utility of the Venus score to assess the trustworthiness of biomedical data.

Noise reduction by adaptive-SIN filtering for retinal OCT images

Optical coherence tomography (OCT) images is widely used in ophthalmic examination, but their qualities are often affected by noises. Shearlet transform has shown its effectiveness in removing image noises because of its edge-preserving property and directional sensitivity. In the paper, we propose an adaptive denoising algorithm for OCT images. The OCT noise is closer to the Poisson distribution than the Gaussian distribution, and shearlet transform assumes additive white Gaussian noise. We hence propose a square-root transform to redistribute the OCT noise. Different manufacturers and differences between imaging objects may influence the observed noise characteristics, which make predefined thresholding scheme ineffective. We propose an adaptive 3D shearlet image filter with noise-redistribution (adaptive-SIN) scheme for OCT images. The proposed adaptive-SIN is evaluated on three benchmark datasets using quantitative evaluation metrics and subjective visual inspection. Compared with other algorithms, the proposed algorithm better removes noise in OCT images and better preserves image details, significantly outperforming in terms of both quantitative evaluation and visual inspection. The proposed algorithm effectively transforms the Poisson noise to Gaussian noise so that the subsequent shearlet transform could optimally remove the noise. The proposed adaptive thresholding scheme optimally adapts to various noise conditions and hence better remove the noise. The comparison experimental results on three benchmark datasets against 8 compared algorithms demonstrate the effectiveness of the proposed approach in removing OCT noise.

Fluorescence microscopy image noise reduction using a stochastically-connected random field model

Fluorescence microscopy is an essential part of a biologist’s toolkit, allowing assaying of many parameters like subcellular localization of proteins, changes in cytoskeletal dynamics, protein-protein interactions, and the concentration of specific cellular ions. A fundamental challenge with using fluorescence microscopy is the presence of noise. This study introduces a novel approach to reducing noise in fluorescence microscopy images. The noise reduction problem is posed as a Maximum A Posteriori estimation problem, and solved using a novel random field model called stochastically-connected random field (SRF), which combines random graph and field theory. Experimental results using synthetic and real fluorescence microscopy data show the proposed approach achieving strong noise reduction performance when compared to several other noise reduction algorithms, using quantitative metrics. The proposed SRF approach was able to achieve strong performance in terms of signal-to-noise ratio in the synthetic results, high signal to noise ratio and contrast to noise ratio in the real fluorescence microscopy data results, and was able to maintain cell structure and subtle details while reducing background and intra-cellular noise.

Automatic 3-D segmentation of endocardial border of the left ventricle from ultrasound images

The segmentation of the left ventricle (LV) is an important task to assess the cardiac function in ultrasound images of the heart. This paper presents a novel methodology for the segmentation of the LV in three-dimensional (3-D) echocardiographic images based on the probabilistic data association filter (PDAF). The proposed methodology begins by initializing a 3-D deformable model either semiautomatically, with user input, or automatically, and it comprises the following feature hierarchical approach: 1) edge detection in the vicinity of the surface (low-level features); 2) edge grouping to obtain potential LV surface patches (mid-level features); and 3) patch filtering using a shape-PDAF framework (high-level features). This method provides good performance accuracy in 20 echocardiographic volumes, and compares favorably with the state-of-the-art segmentation methodologies proposed in the recent literature.

Medical image restoration with different types of noise

The images obtained by X-Ray or computed tomography (CT) in adverse conditions may be contaminated with noise that can affect the detection of diseases. A large number of image processing techniques (filters) have been proposed to remove noise. These techniques depend on the type of noise present in the image. In this work, we propose a method designed to reduce the Gaussian, the impulsive and speckle noise and combined noise. This filter, called PGNDF, combines a non-linear diffusive filter with a peer group with fuzzy metric technique. The proposed filter is able to reduce efficiently the image noise without any information about what kind of noise might be present. To evaluate the filter performance, we use mammographic images from the mini- MIAS database which we have damaged by adding Gaussian, impulsive and speckle noises of different magnitudes. As a result, the proposed method obtains a good performance in most of the different types of noise.

Fast and Automatic Ultrasound Simulation from CT Images

Ultrasound is currently widely used in clinical diagnosis because of its fast and safe imaging principles. As the anatomical structures present in an ultrasound image are not as clear as CT or MRI. Physicians usually need advance clinical knowledge and experience to distinguish diseased tissues. Fast simulation of ultrasound provides a cost-effective way for the training and correlation of ultrasound and the anatomic structures. In this paper, a novel method is proposed for fast simulation of ultrasound from a CT image. A multiscale method is developed to enhance tubular structures so as to simulate the blood flow. The acoustic response of common tissues is generated by weighted integration of adjacent regions on the ultrasound propagation path in the CT image, from which parameters, including attenuation, reflection, scattering, and noise, are estimated simultaneously. The thin-plate spline interpolation method is employed to transform the simulation image between polar and rectangular coordinate systems. The Kaiser window function is utilized to produce integration and radial blurring effects of multiple transducer elements. Experimental results show that the developed method is very fast and effective, allowing realistic ultrasound to be fast generated. Given that the developed method is fully automatic, it can be utilized for ultrasound guided navigation in clinical practice and for training purpose.

Supervised restoration of degraded medical images using multiple-point geostatistics

Reducing noise in medical images has been an important issue of research and development for medical diagnosis, patient treatment, and validation of biomedical hypotheses. Noise inherently exists in medical and biological images due to the acquisition and transmission in any imaging devices. Being different from image enhancement, the purpose of image restoration is the process of removing noise from a degraded image in order to recover as much as possible its original version. This paper presents a statistically supervised approach for medical image restoration using the concept of multiple-point geostatistics. Experimental results have shown the effectiveness of the proposed technique which has potential as a new methodology for medical and biological image processing.

Optimal sampling and estimation in PASL perfusion imaging

Pulsed arterial spin labeling (PASL) techniques potentially allow the absolute, noninvasive quantification of brain perfusion using MRI. This can be achieved by fitting a kinetic model to the data acquired at a number of sampling times. However, the intrinsically low signal-to-noise ratio of PASL measurements usually requires substantial signal averaging, which may result in undesirably long scanning times. A judicious choice of the sampling points is, therefore, crucial in order to minimize scanning time, while optimizing estimation accuracy. On the other hand, a priori information regarding the model parameters may improve estimation performance. Here, we propose a Bayesian framework to determine an optimal sampling strategy and estimation method for the measurement of brain perfusion and arterial transit time (ATT). A Bayesian Fisher information criterion is used to determine the optimal sampling points and a MAP criterion is employed for the estimation of the model parameters, both taking into account the uncertainty in the model parameters as well as the amount of noise in the data. By Monte Carlo simulations, we show that using optimal compared to uniform sampling strategies, as well as the Bayesian estimator relative to a standard least squares approach, improves the accuracy of perfusion and ATT measurements. Moreover, we also demonstrate the applicability of the proposed approach to real data, with the advantage of reduced intersubject variability relative to conventional sampling and estimation approaches.

Ultrasound despeckling for contrast enhancement

Images produced by ultrasound systems are adversely hampered by a stochastic process known as speckle. A despeckling method based upon removing outlier is proposed. The method is developed to contrast enhance B-mode ultrasound images. The contrast enhancement is with respect to decreasing pixel variations in homogeneous regions while maintaining or improving differences in mean values of distinct regions. A comparison of the proposed despeckling filter is compared with the other well known despeckling filters. The evaluations of despeckling performance are based upon improvements to contrast enhancement, structural similarity, and segmentation results on a Field II simulated image and actual B-mode cardiac ultrasound images captured in vivo.

Novel Bayesian deringing method in image interpolation and compression using a SGLI prior

This paper provides a novel Bayesian deringing method to reduce ringing artifacts caused by image interpolation and JPEG compression. To remove the ringing artifacts, the proposed method uses a Bayesian framework based on a SGLI (spatial-gradient-local-inhomogeneity) prior. The SGLI prior employs two complementary discontinuity measures: spatial gradient and local inhomogeniety. The spatial gradient measure effectively detects strong edge components in images. In addition, the local inhomogeniety measure successfully detects locations of the significant discontinuities by taking uniformity of small regions into consideration. The two complementary measures are elaborately combined to create prior probabilities of the Bayesian deringing framework. Thus, the proposed deringing method can effectively preserve the significant discontinuities such as textures of objects as well as the strong edge components in images while reducing the ringing artifacts. Experimental results show that the proposed deringing method achieves average PSNR gains of 0.09 dB in image interpolation artifact reduction and 0.21 dB in JPEG compression artifact reduction.

Efficient, adaptive estimation of two-dimensional firing rate surfaces via Gaussian process methods

Estimating two-dimensional firing rate maps is a common problem, arising in a number of contexts: the estimation of place fields in hippocampus, the analysis of temporally nonstationary tuning curves in sensory and motor areas, the estimation of firing rates following spike-triggered covariance analyses, etc. Here we introduce methods based on Gaussian process nonparametric Bayesian techniques for estimating these two-dimensional rate maps. These techniques offer a number of advantages: the estimates may be computed efficiently, come equipped with natural errorbars, adapt their smoothness automatically to the local density and informativeness of the observed data, and permit direct fitting of the model hyperparameters (e.g., the prior smoothness of the rate map) via maximum marginal likelihood. We illustrate the method's flexibility and performance on a variety of simulated and real data.

A 3-D ultrasound-based framework to characterize the echo morphology of carotid plaques

Carotid atherosclerosis is the main cause of brain stroke, which is the most common life-threatening neurological disease. Nearly all methods aiming at assessing the risk of plaque rupture are based on its characterization from 2-D ultrasound images, which depends on plaque geometry, degree of stenosis, and echo morphology (intensity and texture). The computation of these indicators is, however, usually affected by inaccuracy and subjectivity associated with data acquisition and operator-dependent image selection. To circumvent these limitations, a novel and simple method based on 3-D freehand ultrasound is proposed that does not require any expensive equipment except the common scanner. This method comprises the 3-D reconstruction of carotids and plaques to provide clinically meaningful parameters not available in 2-D ultrasound imaging, namely diagnostic views not usually accessible via conventional techniques and local 3-D characterization of plaque echo morphology. The labeling procedure, based on graph cuts, allows us to identify, locate, and quantify potentially vulnerable foci within the plaque. Validation of the characterization method was made with synthetic data. Results of plaque characterization with real data are encouraging and consistent with the results from conventional methods and after inspection of surgically removed plaques.