Noise Reduction in CT Images Using a Selective Mean Filter

A benchmarked comparison of software packages for time-lapse image processing of monolayer bacterial population dynamics

Time-lapse microscopy offers a powerful approach for analyzing cellular activity. In particular, this technique is valuable for assessing the behavior of bacterial populations, which can exhibit growth and intercellular interactions in a monolayer. Such time-lapse imaging typically generates large quantities of data, limiting the options for manual investigation. Several image-processing software packages have been developed to facilitate analysis. It can thus be a challenge to identify the software package best suited to a particular research goal. Here, we compare four software packages that support the analysis of 2D time-lapse images of cellular populations: CellProfiler, SuperSegger-Omnipose, DeLTA, and FAST. We compare their performance against benchmarked results on time-lapse observations of Escherichia coli populations. Performance varies across the packages, with each of the four outperforming the others in at least one aspect of the analysis. Not surprisingly, the packages that have been in development for longer showed the strongest performance. We found that deep learning-based approaches to object segmentation outperformed traditional approaches, but the opposite was true for frame-to-frame object tracking. We offer these comparisons, together with insight into usability, computational efficiency, and feature availability, as a guide to researchers seeking image-processing solutions.

IMPORTANCE

Time-lapse microscopy provides a detailed window into the world of bacterial behavior. However, the vast amount of data produced by these techniques is difficult to analyze manually. We have analyzed four software tools designed to process such data and compared their performance, using populations of commonly studied bacterial species as our test subjects. Our findings offer a roadmap to scientists, helping them choose the right tool for their research. This comparison bridges a gap between microbiology and computational analysis, streamlining research efforts.

LightGBM is an Effective Predictive Model for Postoperative Complications in Gastric Cancer: A Study Integrating Radiomics with Ensemble Learning

Postoperative complications of radical gastrectomy seriously affect postoperative recovery and require accurate risk prediction. Therefore, this study aimed to develop a prediction model specifically tailored to guide perioperative clinical decision-making for postoperative complications in patients with gastric cancer. A retrospective analysis was conducted on patients who underwent radical gastrectomy at the First Affiliated Hospital of Nanjing Medical University between April 2022 and June 2023. A total of 166 patients were enrolled. Patient demographic characteristics, laboratory examination results, and surgical pathological features were recorded. Preoperative abdominal CT scans were used to segment the visceral fat region of the patients through 3Dslicer, a 3D Convolutional Neural Network (3D-CNN) to extract image features and the LASSO regression model was employed for feature selection. Moreover, an ensemble learning strategy was adopted to train the features and predict postoperative complications of gastric cancer. The prediction performance of the LGBM (Light Gradient Boosting Machine), XGB (XGBoost), RF (Random Forest), and GBDT (Gradient Boosting Decision Tree) models was evaluated through fivefold cross-validation. This study successfully constructed a model for predicting early complications following radical gastrectomy based on the optimal algorithm, LGBM. The LGBM model yielded an AUC value of 0.9232 and an accuracy of 87.28% (95% CI, 75.61-98.95%), surpassing the performance of other models. Through ensemble learning and integration of perioperative clinical data and visceral fat radiomics, a predictive LGBM model was established. This model has the potential to facilitate individualized clinical decision-making and the early recovery of patients with gastric cancer post-surgery.

Synthetization of high-dose images using low-dose CT scans

BACKGROUND

Radiation dose reduction has been the focus of many research activities in x-ray CT. Various approaches were taken to minimize the dose to patients, ranging from the optimization of clinical protocols, refinement of the scanner hardware design, and development of advanced reconstruction algorithms. Although significant progress has been made, more advancements in this area are needed to minimize the radiation risks to patients.

PURPOSE

Reconstruction algorithm-based dose reduction approaches focus mainly on the suppression of noise in the reconstructed images while preserving detailed anatomical structures. Such an approach effectively produces synthesized high-dose images (SHD) from the data acquired with low-dose scans. A representative example is the model-based iterative reconstruction (MBIR). Despite its widespread deployment, its full adoption in a clinical environment is often limited by an undesirable image texture. Recent studies have shown that deep learning image reconstruction (DLIR) can overcome this shortcoming. However, the limited availability of high-quality clinical images for training and validation is often the bottleneck for its development. In this paper, we propose a novel approach to generate SHD with existing low-dose clinical datasets that overcomes both the noise texture issue and the data availability issue.

METHODS

Our approach is based on the observation that noise in the image can be effectively reduced by performing image processing orthogonal to the imaging plane. This process essentially creates an equivalent thick-slice image (TSI), and the characteristics of TSI depend on the nature of the image processing. An advantage of this approach is its potential to reduce impact on the noise texture. The resulting image, however, is likely corrupted by the anatomical structural degradation due to partial volume effects. Careful examination has shown that the differential signal between the original and the processed image contains sufficient information to identify regions where anatomical structures are modified. The differential signal, unfortunately, contains significant noise and has to be removed. The noise removal can be accomplished by performing iterative noise reduction to preserve structural information. The processed differential signal is subsequently subtracted from TSI to arrive at SHD.

RESULTS

The algorithm was evaluated extensively with phantom and clinical datasets. For better visual inspection, difference images between the original and SHD were generated and carefully examined. Negligible residual structure could be observed. In addition to the qualitative inspection, quantitative analyses were performed on clinical images in terms of the CT number consistency and the noise reduction characteristics. Results indicate that no CT number bias is introduced by the proposed algorithm. In addition, noise reduction capability is consistent across different patient anatomical regions. Further, simulated water phantom scans were utilized in the generation of the noise power spectrum (NPS) to demonstrate the preservation of the noise-texture.

CONCLUSIONS

We present a method to generate SHD datasets from regularly acquired low-dose CT scans. Images produced with the proposed approach exhibit excellent noise-reduction with the desired noise-texture. Extensive clinical and phantom studies have demonstrated the efficacy and robustness of our approach. Potential limitations of the current implementation are discussed and further research topics are outlined.

Ensemble classification of integrated CT scan datasets in detecting COVID-19 using feature fusion from contourlet transform and CNN

The COVID-19 disease caused by coronavirus is constantly changing due to the emergence of different variants and thousands of people are dying every day worldwide. Early detection of this new form of pulmonary disease can reduce the mortality rate. In this paper, an automated method based on machine learning (ML) and deep learning (DL) has been developed to detect COVID-19 using computed tomography (CT) scan images extracted from three publicly available datasets (A total of 11,407 images; 7397 COVID-19 images and 4010 normal images). An unsupervised clustering approach that is a modified region-based clustering technique for segmenting COVID-19 CT scan image has been proposed. Furthermore, contourlet transform and convolution neural network (CNN) have been employed to extract features individually from the segmented CT scan images and to fuse them in one feature vector. Binary differential evolution (BDE) approach has been employed as a feature optimization technique to obtain comprehensible features from the fused feature vector. Finally, a ML/DL-based ensemble classifier considering bagging technique has been employed to detect COVID-19 from the CT images. A fivefold and generalization cross-validation techniques have been used for the validation purpose. Classification experiments have also been conducted with several pre-trained models (AlexNet, ResNet50, GoogleNet, VGG16, VGG19) and found that the ensemble classifier technique with fused feature has provided state-of-the-art performance with an accuracy of 99.98%.

Automated development of the contrast-detail curve based on statistical low-contrast detectability in CT images

Purpose

We have developed a software to automatically find the contrast–detail (C–D) curve based on the statistical low‐contrast detectability (LCD) in images of computed tomography (CT) phantoms at multiple cell sizes and to generate minimum detectable contrast (MDC) characteristics.

Methods

A simple graphical user interface was developed to set the initial parameters needed to create multiple grid region of interest of various cell sizes with a 2‐pixel increment. For each cell in the grid, the average CT number was calculated to obtain the standard deviation (SD). Detectability was then calculated by multiplying the SD of the mean CT numbers by 3.29. This process was automatically repeated as many times as the cell size was set at initialization. Based on the obtained LCD, the C–D curve was obtained and the target size at an MDC of 0.6% (i.e., 6‐HU difference) was determined. We subsequently investigated the consistency of the target sizes for a 0.6% MDC at four locations within the homogeneous image. We applied the software to images with six noise levels, images of two modules of the American College of Radiology CT phantom, images of four different phantoms, and images of four different CT scanners. We compared the target sizes at a 0.6% MDC based on the statistical LCD and the results from a human observer.

Results

The developed system was able to measure C–D curves from different phantoms and scanners. We found that the C–D curves follow a power‐law fit. We found that higher noise levels resulted in a higher MDC for a target of the same size. The low‐contrast module image had a slightly higher MDC than the distance module image. The minimum size of an object detected by visual observation was slightly larger than the size using statistical LCD.

Conclusions

The statistical LCD measurement method can generate a C–D curve automatically, quickly, and objectively.

Elimination of biological and physical artifacts in abdomen and brain computed tomography procedures using filtering techniques

•

Filters reduce the noise at the expense of visual image quality.

•

Digital interpretation of images prevents misinterpretation of the images due to blurring of the images.

•

Dose reduction without compromising the diagnostic.

Introduction

Medical images are usually affected by biological and physical artifacts or noise, which reduces image quality and hence poses difficulties in visual analysis, interpretation and thus requires higher doses and increased radiographs repetition rate.

Objectives

This study aims at assessing image quality during CT abdomen and brain examinations using filtering techniques as well as estimating the radiogenic risk associated with CT abdomen and brain examinations.

Materials and Methods

The data were collected from the Radiology Department at Royal Care International (RCI) Hospital, Khartoum, Sudan. The study included 100 abdominal CT images and 100 brain CT images selected from adult patients. Filters applied are namely: Mean filter, Gaussian filter, Median filter and Minimum filter. In this study, image quality after denoising is measured based on the Mean Squared Error (MSE), Peak Signal-to-Noise Ratio (PSNR), and the Structural Similarity Index Metric (SSIM).

Results

The results show that the images quality parameters become higher after applications of filters. Median filter showed improved image quality as interpreted by the measured parameters: PSNR and SSIM, and it is thus considered as a better filter for removing the noise from all other applied filters.

Discussion

The noise removed by the different filters applied to the CT images resulted in enhancing high quality images thereby effectively revealing the important details of the images without increasing the patients’ risks from higher doses.

Conclusions

Filtering and image reconstruction techniques not only reduce the dose and thus the radiation risks, but also enhances high quality imaging which allows better diagnosis.

Effect Evaluation of Platelet-Rich Plasma Combined with Vacuum Sealing Drainage on Serum Inflammatory Factors in Patients with Pressure Ulcer by Intelligent Algorithm-Based CT Image

This work was to explore the efficacy of intelligent algorithm-based computed tomography (CT) to evaluate platelet-rich plasma (PRP) combined with vacuum sealing drainage (VSD) in the treatment of patients with pressure ulcers. Based on the u-net network structure, an image denoising algorithm based on double residual convolution neural network (Dr-CNN) was proposed to denoise the CT images. A total of 84 patients who were hospitalized in hospital were randomly divided into group A (without any intervention), group B (PRP treatment), group C (VSD treatment), and group D (PRP+VSD treatment). Procalcitonin (PCT) was detected by enzyme-linked immunosorbent assay (ELISA) combined with immunofluorescence method, C-reactive protein (CRP) was detected by rate reflectance turbidimetry (RRT), and interleukin-6 (IL-6) was detected by electrochemiluminescence method. The results showed that after treatment, 44 cases (52.38%) of pressure ulcers patients recovered, 24 cases (28.57%) had no change in stage, and 16 cases (19.04%) developed pressure ulcers. The pain scores of group D at 1 week (3.35 ± 0.56 points) and 2 weeks (2.76 ± 0.55 points) after treatment were significantly lower than those in group C (7.77 ± 0.58 points and 6.34 ± 0.44 points, respectively). The time of complete wound healing in group D (24.5 ± 2.32) was obviously lower in contrast to that in groups A, B, and C (35.54 ± 3.22 days, 30.23 ± 2 days, and 29.34 ± 2.15 days, respectively). In addition, the medical satisfaction of group D (8.74 ± 0.69) was significantly higher than that of groups A, B, and C (4.69 ± 0.85, 5.22 ± 0.31, and 5.18 ± 0.59, respectively). The levels of IL-6 and PCT in group D were lower than those in groups A, B, and C, and the differences were statistically significant (P < 0.01). The average values of peak signal to noise ratio (PSNR) and structural similarity index measure (SSIM) of the Dr-CNN network model were 37.21 ± 1.09 dB and 0.925 ± 0.01, respectively, which were higher than other algorithms. The mean values of root mean square error (MSE) and normalized mean absolute distance (NMAD) of the Dr-CNN network model were 0.022 ± 0.002 and 0.126 ± 0.012, respectively, which were significantly lower than other algorithms (P < 0.05). The experimental results showed that PrP combined with VSD could significantly reduce the inflammatory response of patients with pressure ulcers. PRP combined with VSD could significantly reduce the pain of dressing change for patients. Moreover, the performance model of image denoising algorithm based on double residual convolutional neural network was better than other algorithms.