Segmentation of lung lobes and lesions in chest CT for the classification of COVID-19 severity.

To precisely determine the severity of COVID-19-related pneumonia, computed tomography (CT) is an imaging modality beneficial for patient monitoring and therapy planning. Thus, we aimed to develop a deep learning-based image segmentation model to automatically assess lung lesions related to COVID-19 infection and calculate the total severity score (TSS). The entire dataset consisted of 124 COVID-19 patients acquired from Chulabhorn Hospital, divided into 28 cases without lung lesions and 96 cases with lung lesions categorized severity by radiologists regarding TSS. The model used a 3D-UNet along with DenseNet and ResNet models that had already been trained to separate the lobes of the lungs and figure out the percentage of lung involvement due to COVID-19 infection. It also used the Dice similarity coefficient (DSC) to measure TSS. Our final model, consisting of 3D-UNet integrated with DenseNet169, achieved segmentation of lung lobes and lesions with the Dice similarity coefficients of 91.52% and 76.89%, respectively. The calculated TSS values were similar to those evaluated by radiologists, with an R2 of 0.842. The correlation between the ground-truth TSS and model prediction was greater than that of the radiologist, which was 0.890 and 0.709, respectively.

Image Quality Evaluation of a Digital Radiography System Made in Thailand.

BACKGROUND: The National Science and Technology Development Agency (NSTDA) in Thailand researched and prototyped digital radiography systems under the brand name BodiiRay aiming for sustainable development and affordability of medical imaging technology. The image restoration and enhancement were implemented for the systems. PURPOSE: The image quality of the systems was evaluated using images from phantoms and from healthy volunteers. METHODS: The survey phantom images from BodiiRay and other two commercial systems using the exposure settings for the chest, the abdomen, and the extremity were evaluated by three experience observers in terms of the high-contrast image resolution, the low-contrast image detectability, and the grayscale differentiation. The volunteer images of the chests, the abdomens, and the extremities from BodiiRay were evaluated by three specialized radiologists based on visual grading on 5-point scaled questionnaires for the anatomy visibility, the image quality satisfaction, and the diagnosis confidence in using the images. RESULTS: BodiiRay phantom results were similar to those from the commercial systems. The overall performance averaged across the exposure settings showed that BodiiRay was slightly better than Fujifilm FDR Go in the low-contrast detectability (p = 0.033) and in the grayscale differentiation (p = 0.004). It was also slightly better than Siemens YSIO Max in the high-contrast resolution (p = 0.018). The images of chest, pelvis, and hand phantoms illustrated comparable visual quality. For volunteer images, the percentage of the images scored ≥4 ranged from 61% to 99%, 23% to 92%, and 96% to 99% for the chest, abdomen, and extremity images, respectively. The average score ranged from 3.63 to 4.46, 3.18 to 4.21, and 4.41 to 4.51 for the chest, abdomen, and extremity images, respectively. CONCLUSION: The phantom image results showed the comparability of these systems. The clinical evaluation showed BodiiRay images provided sufficient image qualities for digital radiography of these body parts.

Adaptive Multi-Scale Image Enhancement for Digital Radiography.

Digital radiography has been increasingly adopted since it can provide better image quality compared to conventional screen/film method. However, digital radiography can sometimes produce low-quality images because its processing algorithm is unaware of the content. Here, an adaptive multi-scale image enhancement algorithm for digital radiography is demonstrated. The algorithm adapts to the context of the image, thus providing better image quality. The qualitative and quantitative validations of the algorithm in phantoms and in clinical settings showed satisfactory performance.

Toward Optimal Computation of Ultrasound Image Reconstruction Using CPU and GPU.

An ultrasound image is reconstructed from echo signals received by array elements of a transducer. The time of flight of the echo depends on the distance between the focus to the array elements. The received echo signals have to be delayed to make their wave fronts and phase coherent before summing the signals. In digital beamforming, the delays are not always located at the sampled points. Generally, the values of the delayed signals are estimated by the values of the nearest samples. This method is fast and easy, however inaccurate. There are other methods available for increasing the accuracy of the delayed signals and, consequently, the quality of the beamformed signals; for example, the in-phase (I)/quadrature (Q) interpolation, which is more time consuming but provides more accurate values than the nearest samples. This paper compares the signals after dynamic receive beamforming, in which the echo signals are delayed using two methods, the nearest sample method and the I/Q interpolation method. The comparisons of the visual qualities of the reconstructed images and the qualities of the beamformed signals are reported. Moreover, the computational speeds of these methods are also optimized by reorganizing the data processing flow and by applying the graphics processing unit (GPU). The use of single and double precision floating-point formats of the intermediate data is also considered. The speeds with and without these optimizations are also compared.

Toward a practical protocol for human optic nerve DTI with EPI geometric distortion correction.

PURPOSE: To develop a practical protocol for diffusion tensor imaging (DTI) of the human optic nerve with echo planar imaging (EPI) geometric distortion correction. MATERIALS AND METHODS: A conventional DTI protocol was modified to acquire images with fat and cerebrospinal fluid (CSF) suppression and field inhomogeneity maps of contiguous coronal slices covering the whole brain. The technique was applied to healthy volunteers and multiple sclerosis patients with and without a history of unilateral optic neuritis. DTI measures and optic nerve tractography before and after geometric distortion correction were compared. Diffusion measures from left and right or from affected and unaffected eyes in different subject cohorts were reported. RESULTS: The image geometry after correction closely resembled reference anatomical images. Optic nerve tractography became feasible after distortion correction. The diffusion measures from the healthy volunteers were in good agreement with the literature. Statistically significant differences were found in the fractional anisotropy and orthogonal eigenvalues between affected and unaffected eyes in optic neuritis patients with poor recovery. The diffusion measures before and after geometric distortion correction were not significantly different. For cohorts without optic neuritis, the difference between diffusion measures from left and right eyes was not statistically significant. CONCLUSION: The proposed technique could provide a practical DTI protocol to study the human optic nerve.

Estimation of mutual information objective function based on Fourier shift theorem: an application to eddy current distortion correction in diffusion tensor imaging.

Diffusion tensor imaging requires correction of eddy current distortion in diffusion-weighted images. An effective retrospective correction approach is to transform a diffusion-weighted image to maximize the mutual information (MI) between the transformed diffusion-weighted image and the corresponding T2-weighted image. In the literature, either linear interpolation or partial volume interpolation is applied to estimate the MI objective function. However, these interpolation methods induce artifacts to the MI objective function, thus compromising correction results. In this work, the MI objective function is estimated based on interpolation using Fourier shift theorem. This method eliminates the artifacts incurred with the aforementioned interpolation methods. The algorithm is further improved by approximating pixel values using their nearest neighbors in the up-sampled spatial domain, resulting in dramatically increased computational efficiency without compromising the correction results. The effects of varying the number of quantization levels and using Parzen window filtering to smooth the MI objective function are also investigated to obtain optimized algorithm parameters. The diffusion tensor image quality after applying the proposed distortion correction method is significantly improved visually.

Phase labeling using sensitivity encoding (PLUS): data acquisition and image reconstruction for geometric distortion correction in EPI.

Geometric distortion caused by magnetic field inhomogeneity is generally an inevitable tradeoff for fast MRI acquisitions using echo-planar imaging. Most of the existing distortion-correction techniques require separate scans for field maps in order to correct the distortion contained in a measurement. A drawback of these current techniques is that the field map scans and the measurement can capture different patient positions, which invalidates the stationary condition. A new method was developed in this work to correct geometric distortion by using local phase shifts derived directly from the measurement itself, without the need of extra field map scans. This self-sufficient method takes advantage of parallel imaging and k-space trajectory modification to produce multiple images from a single acquisition. The measurement is also used to derive sensitivity maps for parallel imaging reconstruction. The derived phase shifts are retrospectively applied to the measurement for correction of geometric distortion in the measurement itself. The proposed method was successfully demonstrated using experimental data from a phantom and a human brain.

Revisiting anaplastic astrocytomas II: further characterization of an expansive growth pattern with visually enhanced diffusion tensor imaging.

PURPOSE: To seek to distinguish and visualize the different magnetic resonance imaging (MRI) growth patterns among malignant gliomas utilizing visually enhanced diffusion tensor imaging (DTI). MATERIALS AND METHODS: Nineteen consecutive patients undergoing image-guided resection of a newly diagnosed malignant glioma underwent add-on acquisition of DTI data based on an Institutional Review Board (IRB)-approved imaging protocol during preoperative MRI scans for routine intraoperative image guidance. Tumor growth patterns were assigned to expansive or mixed/infiltrative classes as described in the companion article (24). Infiltrating tumors were WHO Grade IV astrocytomas and all expansive tumors were either WHO Grade III astrocytomas or WHO Grade II astrocytomas. DTI-based white matter tractography was conducted and the DTI data were fused with anatomical images using an in-house software package we developed to enhance the visualization of the tumor/fiber interface. In one case additional analysis was performed with 2D multivoxel (1)H-MRSI utilizing a 2D chemical shift imaging (CSI) technique to corroborate the nature of this interface. RESULTS: Out of the 19 tumor patients studied, 11 had infiltrative tumors and the other 8 had expansive tumors. While less clear with 2D axial diffusion color maps, visually enhanced 3D reconstructions of the tumor/fiber interface successfully corroborated distinctive growth patterns. This was particularly evident when viewed in 3D video loops of each tumor/fiber interface. CONCLUSION: We have successfully developed software that visually enhances the anatomic details of the tumor/fiber interface in patients with anaplastic astrocytomas. These data support the existence of a subgroup of patients within the WHO Grade III classification with expansive tumors and a significantly better prognosis.

Segmentation of elastographic images using a coarse-to-fine active contour model.

Delineation of radiofrequency-ablation-induced coagulation (thermal lesion) boundaries is an important clinical problem that is not well addressed by conventional imaging modalities. Elastography, which produces images of the local strain after small, externally applied compressions, can be used for visualization of thermal coagulations. This paper presents an automated segmentation approach for thermal coagulations on 3-D elastographic data to obtain both area and volume information rapidly. The approach consists of a coarse-to-fine method for active contour initialization and a gradient vector flow, active contour model for deformable contour optimization with the help of prior knowledge of the geometry of general thermal coagulations. The performance of the algorithm has been shown to be comparable to manual delineation of coagulations on elastograms by medical physicists (r = 0.99 for volumes of 36 radiofrequency-induced coagulations). Furthermore, the automatic algorithm applied to elastograms yielded results that agreed with manual delineation of coagulations on pathology images (r = 0.96 for the same 36 lesions). This algorithm has also been successfully applied on in vivo elastograms.

Correlation of RF signals during angular compounding.

A theoretical analysis of the correlation between radio-frequency (RF) echo signal data acquired from the same location but at different angles is presented. The accuracy of the theoretical results is verified with computer simulations. Refinements to previous analyses of the correlation of RF signals originating from the same spatial location at different angular positions are made. We extend the analysis to study correlation of RF signals coming from different spatial locations and eventually correlation of RF signal segments that intersect at the same spatial location. The theory predicts a faster decorrelation with a change in the insonification angle for longer RF echo signal segments. As the RF signal segment becomes shorter, the decorrelation rate with angle is slower and approaches the limit corresponding to the correlation of RF signals originating from the same spatial location. Theoretical results provide a clear understanding of angular compounding techniques used to improve the signal-to-noise ratio in ultrasonic parametric imaging and in elastography.

Improvements in elastographic contrast-to-noise ratio using spatial-angular compounding.

Spatial-angular compounding is a new technique developed for improving the signal-to-noise ratio (SNR) in elastography. Under this method, elastograms of a region-of-interest (ROI) are obtained from a spatially weighted average of local strain estimated along different insonification angles. In this article, we investigate the improvements in the strain contrast and contrast-to-noise ratio (CNR) of the spatially compounded elastograms. Spatial angular compounding is also applied and evaluated in conjunction with global temporal stretching. Quantitative experimental results obtained using a single-inclusion tissue-mimicking phantom demonstrate that the strain contrast reduces slightly but the CNR improves by around 8 to 13 dB. We also present experimental spatial angular compounding results obtained from an in vitro thermal lesion in canine liver tissue embedded in a gelatin phantom that demonstrate the improved visual characteristics (due to the improved CNR) of the compound elastogram. The experimental results provide guidelines for the practical range of maximum insonification angles and estimates of the optimum angular increment.

Monitoring stiffness changes in lesions after radiofrequency ablation at different temperatures and durations of ablation.

The variations in the stiffness or stiffness contrast of lesions resulting from radiofrequency (RF) ablation of canine liver tissue at different temperatures and for different ablation durations at a specified temperature are analyzed. Tissue stiffness, in general, increases with temperature; however, an anomaly exists around 80 degrees C, where the stiffness of the lesion is lower than that of the lesion ablated at 70 degrees C. On the other hand, the stiffness increases monotonically with the duration of ablation. Plots illustrating the ratio of mean strains in normal canine liver tissue to mean strains in ablated thermal lesions demonstrate the variation in the stiffness contrast of the thermal lesions. The contrast-to-noise ratio (CNRe) of the lesions, which serves as an indicator of the detectability of the lesions under the different experimental imaging conditions described above, is also presented. The results presented in this paper show that the elastographic depiction of stiffer thermal lesions is better, in terms of the CNRe parameter. An important criterion in the elastographic depiction of RF-ablated regions of tissue is the trade-off between ablation temperature and duration of ablation. Tissue necrosis can occur either by ablating tissue to high temperatures for short durations or to lower temperatures for longer durations. In this paper, we attempt to characterize the elastographic depiction of thermal lesions under these different experimental conditions. This paper provides results that may be utilized by practitioners of RF ablation to decide the ablation temperature and duration, on the basis of the strain images of normal liver tissue and ablated thermal lesions discussed in this paper.

Ultrasonic noninvasive temperature estimation using echoshift gradient maps: simulation results.

Percutaneous ultrasound-image-guided radiofrequency (rf) ablation is an effective treatment for patients with hepatic malignancies that are excluded from surgical resection due to other complications. However, ablated regions are not clearly differentiated from normal untreated regions using conventional ultrasound imaging due to similar echogenic tissue properties. In this paper, we investigate the statistics that govern the relationship between temperature elevation and the corresponding temperature map obtained from the gradient of the echoshifts obtained using consecutive ultrasound radiofrequency signals. A relationship derived using experimental data on the sound speed and tissue expansion variations measured on canine liver tissue samples at different elevated temperatures is utilized to generate ultrasound radiofrequency simulated data. The simulated data set is then utilized to statistically estimate the accuracy and precision of the temperature distributions obtained. The results show that temperature increases between 37 and 67 degrees C can be estimated with standard deviations of +/- 3 degrees C. Our results also indicate that the correlation coefficient between consecutive radiofrequency signals should be greater than 0.85 to obtain accurate temperature estimates.

Elastographic versus x-ray CT imaging of radio frequency ablation coagulations: an in vitro study.

Techniques to image elasticity parameters (i.e., elastography) have recently become of great interest to researchers. In this paper we use conventional ultrasound elastography and x-ray CT to image radio frequency (RF) ablation sites of excised canine liver enclosed in gelatin. Thermal coagulations of different sizes were produced by applying the RF procedure for various times and end point temperatures. Dimensions, areas and volumes computed from CT and elastography were compared with those on whole mount pathology specimens. Ultrasound elastography exhibited high contrast for the thermal coagulations and performed better than CT. The correlation between pathology and elastography for this sample set of 40 thermal coagulations (r = 0.94 for volume estimation, r = 0.87 for area estimation) is better than the correlation between pathology and CT (r = 0.89 for volume estimation, r = 0.82 for area estimation).

Noise reduction using spatial-angular compounding for elastography.

Ultrasound elastography has developed into an imaging modality suitable for detection and diagnosis of cancers in the breast, prostate, and thyroid and for monitoring ablative therapies in the liver, kidneys, and other sites. In this article, a new approach is described that enables the reduction of noise artifacts in elastography without a significant reduction in either the contrast or spatial resolution. The technique uses angular-weighted compounding of local angular strains estimated from echo signals scanned at different insonification angles. Strain estimated along angular insonification directions can be separated into strain tensor components along the axial (direction of compression) and lateral directions. The mechanical stimulus is applied only along one direction. Angular-weighting factors are derived from the relationship between the axial and lateral strains under the assumption of tissue incompressibility. Experimental results using a uniformly elastic, tissue-mimicking phantom demonstrate the improvement in the signal-to-noise ratio obtained with angular-weighted compounding. Variation in the signal-to-noise ratio obtained using different angular increments also is investigated. Elastograms obtained from an inclusion phantom also demonstrate the improvement in contrast detail resolution obtained using spatial-angular compounding.

Wavelet denoising of displacement estimates in elastography.

Wavelet shrinkage denoising of the displacement estimates to reduce noise artefacts, especially at high overlaps in elastography, is presented in this paper. Correlated errors in the displacement estimates increase dramatically with an increase in the overlap between the data segments. These increased correlated errors (due to the increased correlation or similarity between consecutive displacement estimates) generate the so-called "worm" artefact in elastography. However, increases in overlap on the order of 90% or higher are essential to improve axial resolution in elastography. The use of wavelet denoising significantly reduces errors in the displacement estimates, thereby reducing the worm artefacts, without compromising on edge (high-frequency or detail) information in the elastogram. Wavelet denoising is a term used to characterize noise rejection by thresholding the wavelet coefficients. Worm artefacts can also be reduced using a low-pass filter; however, low-pass filtering of the displacement estimates does not preserve local information such as abrupt change in slopes, causing the smoothing of edges in the elastograms. Simulation results using the analytic 2-D model of a single inclusion phantom illustrate that wavelet denoising produces elastograms with the closest correspondence to the ideal mechanical strain image. Wavelet denoising applied to experimental data obtained from an in vitro thermal lesion phantom generated using radiofrequency (RF) ablation also illustrates the improvement in the elastogram noise characteristics.

Impact of gas bubbles generated during interstitial ablation on elastographic depiction of in vitro thermal lesions.

OBJECTIVE: Artifacts from gas bubble formation during radio frequency ablation along with the poor intrinsic contrast between normal and treated regions (zone of necrosis) are considerable problems for the visualization of the necrotic region on conventional sonography. Sonographic elastography is very effective for visualizing the zone of necrosis, but it uses the same echo signals to estimate strain as those used to form gray scale images. Thus, the impact of gas bubbles on strain images or elastograms must be investigated. METHODS: Radio frequency ablation was performed in vitro on liver tissue samples, approximately 40 x 40 x 20 mm, encased in 80-mm cubed gelatin phantoms. Elastograms generated at different instants during the ablation procedures were obtained on a real-time scanner with a 5-MHz linear array. Sequences of elastograms illustrate the growth of the thermal lesion. RESULTS: Degradation of the distal boundary of the thermal lesion was observed. The degradation was confined to the lower-fifth quadrant of the thermal lesion. However, accurate estimates of lesion areas could still be obtained by extrapolation of the thermal lesion boundary. CONCLUSIONS: Elastograms of thermal lesions in vitro can be obtained during radio frequency ablation. Some loss of thermal lesion boundary information on strain images was observed in regions where attenuation due to gas bubbles reduced the signal-noise ratio of the echo signals.

Elastographic measurement of the area and volume of thermal lesions resulting from radiofrequency ablation: pathologic correlation.

OBJECTIVE: Elastography is a promising tool for visualizing the zone of necrosis in liver tissue resulting from radiofrequency tumor ablation. Because heat-ablated tissues are stiffer than normal untreated tissue, elastography may prove useful for following up patients who undergo radiofrequency ablative therapy. We sought to report the initial evaluations of the reliability of elastography for delineating thermal lesion boundaries in liver tissue by comparing lesion dimensions determined by elastography with the findings at whole-mount pathology. MATERIALS AND METHODS: Radiofrequency ablation was performed in vitro on liver tissue specimens encased in gelatin phantoms. The imaging plane for elastography was perpendicular to the axis of the radiofrequency electrode so that the ablated region was around the center of the plane. To obtain three-dimensional visualization of thermal lesions, we reconstructed the lesions from multiple elastograms by linearly translating the elastographic scanning plane. Pathology photographs were obtained in the same image plane used for elastography by slicing through the gelatin and tissue phantom using external markers. We used digitized gross pathology photographs obtained at a specified slice thickness to compute the areas and volumes of the lesions. These measurements were then compared to the measurements obtained from the elastograms. RESULTS: In a sample of 40 thermal lesions, we obtained a correlation between in vitro elastographic and pathologic measurements of r = 0.9371 (p < 0.00001) for area estimates and r = 0.979 (p < 0.00001) for volume estimates. CONCLUSION: We found excellent correlation between the measurements of the dimensions, areas, and volumes of thermal lesions that were based on elastographic images and the measurements that were based on digitized pathologic images. When compared with digitized pathologic measurements, elastographic measurements showed a tendency to slightly underestimate both the areas and volumes of lesions. Nevertheless, elastography is a reliable technique for delineating thermal lesions resulting from radiofrequency ablation.