A deep-learning semantic segmentation approach to fully automated MRI-based left-ventricular deformation analysis in cardiotoxicity

Left-ventricular (LV) strain measurements with the Displacement Encoding with Stimulated Echoes (DENSE) MRI sequence provide accurate estimates of cardiotoxicity damage related to breast cancer chemotherapy. This study investigated an automated LV chamber quantification tool via segmentation with a supervised deep convolutional neural network (DCNN) before strain analysis with DENSE images. Segmentation for chamber quantification analysis was conducted with a custom DeepLabV3+ DCNN with ResNet-50 backbone on 42 female breast cancer datasets (22 training-sets, eight validation-sets and 12 independent test-sets). Parameters such as LV end-diastolic diameter (LVEDD) and ejection fraction (LVEF) were quantified, and myocardial strains analyzed with the Radial Point Interpolation Method (RPIM). Myocardial classification was validated against ground-truth with sensitivity-specificity analysis, the metrics of Dice, average perpendicular distance (APD) and Hausdorff-distance. Following segmentation, validation was conducted with the Cronbach's Alpha (C-Alpha) intraclass correlation coefficient between LV chamber quantification results with DENSE and Steady State Free Precession (SSFP) acquisitions and a vendor tool-based method to segment the DENSE data, and similarly for myocardial strain analysis in the chambers. The results of myocardial classification from segmentation of the DENSE data were accuracy = 97%, Dice = 0.89 and APD = 2.4 mm in the test-set. The C-Alpha correlations from comparing chamber quantification results between the segmented DENSE and SSFP data and vendor tool-based method were 0.97 for LVEF (56 ± 7% vs 55 ± 7% vs 55 ± 6%, p = 0.6) and 0.77 for LVEDD (4.6 ± 0.4 cm vs 4.5 ± 0.3 cm vs 4.5 ± 0.3 cm, p = 0.8). The validation metrics against ground-truth and equivalent parameters obtained from the SSFP segmentation and vendor tool-based comparisons show that the DCNN approach is applicable for automated LV chamber quantification and subsequent strain analysis in cardiotoxicity.     

DENSE: displacement encoding with stimulated echoes in cardiac functional MRI

Displacement encoding with stimulated echoes (DENSE) was developed for high-resolution myocardial displacement mapping. Pixel phase is modulated by myocardial displacement and data spatial resolution is limited only by pixel size. 2D displacement vector maps were generated for the systolic action in canines with 0.94 x 1.9 mm nominal in-plane resolution and 2.3 mm/pi displacement encoding. A radial strain of 0.208 was measured across the free left ventricular wall over 105 ms during systole. DENSE displacement maps require small first-order gradient moments for encoding. DENSE magnitude images exhibit black-blood contrast which allows for better myocardial definition and reduced motion-related artifacts.     

Reproducibility of cine displacement encoding with stimulated echoes (DENSE) in human subjects

ObjectivesTo test the hypothesis that two-dimensional (2D) displacement encoding via stimulated echoes (DENSE) is a reproducible technique for the depiction of segmental myocardial motion in human subjects.Materials and methodsFollowing the approval of the institutional review board (IRB), 17 healthy volunteers without documented history of cardiovascular disease were recruited. For each participant, 2D DENSE were performed twice (at different days) and the images were obtained at basal, midventricular and apical levels of the left ventricle (LV) with a short-axis view. The radial thickening strain (Err), circumferential strain (Ecc), twist and torsion were calculated. The intra-, inter-observer and inter-study variations of DENSE-derived myocardial motion indices were evaluated using coefficient of variation (CoV) and intra-class correlation coefficient (ICC).ResultsIn total, there are 272 pairs of myocardial segments (data points) for comparison. There is good intra- and inter-observer reproducibility for all DENSE-derived measures in 17 participants. There is good inter-study reproducibility for peak Ecc (CoV = 19.64%, ICC = 0.8896, p < 0.001), twist (CoV = 33.11%, ICC = 0.9135, p < 0.001) and torsion (CoV = 13.96%, ICC = 0.8684, p < 0.001). There is moderate inter-study reproducibility for Err (CoV = 38.89%, ICC = 0.7022, p < 0.001).ConclusionDENSE is a reproducible technique for characterizing LV regional systolic myocardial motion on a per-segment basis in healthy volunteers.     

Correction of out-of-FOV motion artifacts using convolutional neural network

PurposeSubject motion during MRI scan can result in severe degradation of image quality. Existing motion correction algorithms rely on the assumption that no information is missing during motions. However, this assumption does not hold when out-of-FOV motion happens. Currently available algorithms are not able to correct for image artifacts introduced by out-of-FOV motion. The purpose of this study is to demonstrate the feasibility of incorporating convolutional neural network (CNN) derived prior image into solving the out-of-FOV motion problem.Methods and materialsA modified U-net network was proposed to correct out-of-FOV motion artifacts by incorporating motion parameters into the loss function. A motion model based data fidelity term was applied in combination with the CNN prediction to further improve the motion correction performance. We trained the CNN on 1113 MPRAGE images with simulated oscillating and sudden motion trajectories, and compared our algorithm to a gradient-based autofocusing (AF) algorithm in both 2D and 3D images. Additional experiment was performed to demonstrate the feasibility of transferring the networks to different dataset. We also evaluated the robustness of this algorithm by adding Gaussian noise to the motion parameters. The motion correction performance was evaluated using mean square error (NMSE), peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM).ResultsThe proposed algorithm outperformed AF-based algorithm for both 2D (NMSE: 0.0066 ± 0.0009 vs 0.0141 ± 0.008, P < .01; PSNR: 29.60 ± 0.74 vs 21.71 ± 0.27, P < .01; SSIM: 0.89 ± 0.014 vs 0.73 ± 0.004, P < .01) and 3D imaging (NMSE: 0.0067 ± 0.0008 vs 0.070 ± 0.021, P < .01; PSNR: 32.40 ± 1.63 vs 22.32 ± 2.378, P < .01; SSIM: 0.89 ± 0.01 vs 0.62 ± 0.03, P < .01). Robust reconstruction was achieved with 20% data missed due to the out-of-FOV motion.ConclusionIn conclusion, the proposed CNN-based motion correction algorithm can significantly reduce out-of-FOV motion artifacts and achieve better image quality compared to AF-based algorithm.     

A robust and accurate center-frequency estimation (RACE) algorithm for improving motion estimation performance of SinMod on tagged cardiac MR images without known tagging parameters

A robust and accurate center-frequency (CF) estimation (RACE) algorithm for improving the performance of the local sine-wave modeling (SinMod) method, which is a good motion estimation method for tagged cardiac magnetic resonance (MR) images, is proposed in this study. The RACE algorithm can automatically, effectively and efficiently produce a very appropriate CF estimate for the SinMod method, under the circumstance that the specified tagging parameters are unknown, on account of the following two key techniques: (1) the well-known mean-shift algorithm, which can provide accurate and rapid CF estimation; and (2) an original two-direction-combination strategy, which can further enhance the accuracy and robustness of CF estimation. Some other available CF estimation algorithms are brought out for comparison. Several validation approaches that can work on the real data without ground truths are specially designed. Experimental results on human body in vivo cardiac data demonstrate the significance of accurate CF estimation for SinMod, and validate the effectiveness of RACE in facilitating the motion estimation performance of SinMod.     

Efficacy of motion artifact reduction in neonatal DW segmented EPI at 3 T using phase correction by numerical optimization and segment data swapping

Segmented echoplanar imaging (EPI) is a potentially valuable acquisition method for neonatal diffusion-weighted imaging (DWI) due to the lower acoustic noise levels as well as reduced blurring and distortion associated with it, as compared with single-shot EPI. Reduced acoustic noise may be important for the safety of neonates. However, little information regarding the efficacy of segmented EPI motion correction schemes is available for the neonatal population. We quantitatively assessed the efficacy of a postprocessing technique for motion artifact reduction involving phase correction by nonlinear optimization, alone and in combination with a novel method of utilizing a second data set (referred to as segment data swapping). These methods were applied to three-directional eight-segment echoplanar DW images obtained from 13 sedated neonates and to nine-directional DW images from 3 unsedated neonates. For comparison, the efficacy of the nonlinear optimization method was also evaluated in four adults. Motion correction efficacy was quantified using the motion artifact-to-signal ratio (ASR). The median, 70th percentile and 90th percentile ASR values obtained from neonatal three-directional DWI using nonlinear optimization alone were 2.8%, 4.6% and 9.6%, respectively. Efficacy improved (P<.005), particularly in dealing with the images most difficult to correct, when the phase correction by numerical optimization was combined with segment data swapping (median ASR=1.9%, 70th percentile ASR=2.7%, 90th percentile ASR=4.3%). Similar results were obtained for nine-directional diffusion tensor imaging. Nonlinear optimization alone applied to adult images showed significantly (P<.001) lower ASR values (median ASR=0.9%, 70th percentile ASR=2.1%, 90th percentile ASR=4.1%), demonstrating the greater challenge in DWI of neonates with segmented EPI. In conclusion, phase correction by nonlinear optimization provides effective motion correction for neonatal DW eight-segment EPI, especially when used in conjunction with segment data swapping.     

Three-dimensional myocardial strain analysis based on short- and long-axis magnetic resonance tagged images using a 1D displacement field

A robust algorithm to estimate three-dimensional strain in the left-ventricular heart wall, based on magnetic resonance (MR) grid-tagging in two sets of orthogonal image planes, is presented. Starting-point of this study was to minimize global interpolation and smoothing. Only the longitudinal displacement was interpolated between long-axis images. Homogeneous strain analysis was performed using small tetrahedrons. The method was tested using a stack of short-axis images and three long-axis images in six healthy volunteers. In addition, the method was subjected to an analytical test case, in which the effect of noise in tag point position on the observed strains was explored for normally distributed noise (0.5 mm RMS). In volunteers, the error in the longitudinal displacement due to interpolation between the long-axis image planes was -0.10 +/- 0. 48 mm (mean +/- SD). The resulting error in the longitudinal strain epsilon(l) was -0.003 +/- 0.02. The analytical test case was used to quantify the effects of three sources of errors on the observed strain. The SD of the difference between homogeneous strain and true strain was 0.06 for epsilon(r.) The error due to the 3-D reconstruction was 0.004 for epsilon(r.) The error in epsilon(r) resulting from simulated noise in the tag point position was 0.10. Equivalent results were obtained for all other strain parameters; thus, the error resulting from noise in the tag point position dominates the error introduced by approximations in the method. Because the proposed method uses a minimum of global interpolation and smoothing, it offers the prospect to detect small regions of aberrant contraction.     

Determination of normal regional left ventricular function from cine-MR images using a semi-automated edge detection method

A semi-automated edge detection method for the delineation of the endo- and epicardial borders of the left ventricle from cine MR images has been developed. The feasibility of this was demonstrated by processing end diastolic and end systolic ECG-gated images of four short axis images in 10 healthy subjects. The first derivative method combined with a 2D weighted polynomial fitting procedure was used to determine the endo- and epicardial borders, which then allowed determination of the wall motion, wall thickening, and ejection fraction, of the left ventricle. The results show that the end-systolic radial wall motion varies from (32 ± 8)% to (76 ± 12)%, and wall thickening from (0.60 ± 0.46) cm to (1.26 ± 0.50) cm. An average ejection fraction of (69 ± 6)% was found which agrees well with literature values. The method described, for the delineation of the borders, reduces considerably the long and tedious operator time inherent in manual measurement and greatly increases the reproducibility of the measurements.     

Principal component analysis of shear strain effects

Shear stresses are always present during quasi-static strain imaging, since tissue slippage occurs along the lateral and elevational directions during an axial deformation. Shear stress components along the axial deformation axes add to the axial deformation while perpendicular components introduce both lateral and elevational rigid motion and deformation artifacts into the estimated axial and lateral strain tensor images. A clear understanding of these artifacts introduced into the normal and shear strain tensor images with shear deformations is essential. In addition, signal processing techniques for improved depiction of the strain distribution is required. In this paper, we evaluate the impact of artifacts introduced due to lateral shear deformations on the normal strain tensors estimated by varying the lateral shear angle during an axial deformation. Shear strains are quantified using the lateral shear angle during the applied deformation. Simulation and experimental validation using uniformly elastic and single inclusion phantoms were performed. Variations in the elastographic signal-to-noise and contrast-to-noise ratios for axial deformations ranging from 0% to 5%, and for lateral deformations ranging from 0 to 5° were evaluated. Our results demonstrate that the first and second principal component strain images provide higher signal-to-noise ratios of 20 dB with simulations and 10 dB under experimental conditions and contrast-to-noise ratio levels that are at least 20 dB higher when compared to the axial and lateral strain tensor images, when only lateral shear deformations are applied. For small axial deformations, the lateral shear deformations significantly reduces strain image quality, however the first principal component provides about a 1–2 dB improvement over the axial strain tensor image. Lateral shear deformations also significantly increase the noise level in the axial and lateral strain tensor images with larger axial deformations. Improved elastographic signal and contrast-to-noise ratios in the first principal component strain image are always obtained for both simulation and experimental data when compared to the corresponding axial strain tensor images in the presence of both axial and lateral shear deformations.     

Tissue-tracking in the assessment of late gadolinium enhancement in myocarditis and myocardial infarction

PurposeTo test the diagnostic performance of cardiovascular magnetic resonance (CMR) tissue-tracking (TT) to detect the presence of late gadolinium enhancement (LGE) in patients with a diagnosis of myocardial infarction (MI) or myocarditis (MYO), preserved left ventricular ejection fraction (LVEF) and no visual regional wall motion abnormalities (RWMA).MethodsWe selected consecutive CMR studies of 50 MI, 50 MYO and 96 controls. Receiving operating characteristic (ROC) curve and net reclassification index (NRI) analyses were used to assess the predictive ability and the incremental diagnostic yield of 2D and 3D TT-derived strain parameters for the detection of LGE and to measure the best cut-off values of strain parameters.ResultsOverall, cases showed significantly reduced 2D global longitudinal strain (2D-GLS) values compared with controls (−20.1 ± 3.1% vs −21.6 ± 2.7%; p = 0.0008). 2D-GLS was also significantly reduced in MYO patients compared with healthy controls (−19.7 ± 2.9% vs −21.9 ± 2.4%; p = 0.0001). 3D global radial strain (3D-GRS) was significantly reduced in MI patients compared with controls with risk factors (34.3 ± 11.8% vs 40.3 ± 12.5%, p = 0.024) Overall, 2D-GLS yielded good diagnostic accuracy for the detection of LGE in the MYO subgroup (AUROC 0.79; NRI (95% CI) = 0.6 (0.3, 1.02) p = 0.0004), with incremental predictive value beyond risk factors and LV function parameters (p for AUROC difference = 0.048). In the MI subgroup, 2D-GRS (AUROC 0.81; NRI (95% CI) = 0.56 (0.17, 0.95) p = 0.004), 3D-GRS (AUROC 0.82; NRI (95% CI) = 0.57 (0.17, 0.97) p = 0.006) and 3D global circumferential strain (3D-GCS) (AUROC 0.81; NRI (95% CI) = 0.62 (0.22, 1.01) p = 0.002) emerged as potential markers of disease. The best cut-off for 2D-GLS was −21.1%, for 2D- and 3D-GRS were 39.1% and 37.7%, respectively, and for 3D-GCS was −16.4%.ConclusionsAt CMR-tissue tracking analysis, 2D-GLS was a significant predictor of LGE in patients with myocarditis but preserved LVEF and no visual RWMA. Both 2D- and 3D-GRS and 2D-GCS yielded good diagnostic accuracy for LGE detection in patients with previous MI but preserved LVEF and no visual RWMA.     

The effect of out-of-plane motion on 2D and 3D digital image correlation measurements

The effect of out-of-plane motion (including out-of-plane translation and rotation) on two-dimensional (2D) and three-dimensional (3D) digital image correlation measurements is demonstrated using basic theoretical pinhole image equations and experimentally through synchronized, multi-system measurements. Full-field results obtained during rigid body, out-of-plane motion using a single-camera vision system with (a-1) a standard f55mm Nikon lens and (a-2) a single Schneider–Kreuznach Xenoplan telecentric lens are compared with data obtained using a two-camera stereovision system with standard f55mm Nikon lenses.Results confirm that the theoretical equations are in excellent agreement with experimental measurements. Specifically, results show that (a) a single-camera, 2D imaging system is sensitive to out-of-plane motion, with in-plane strain errors (a-1) due to out-of-plane translation being proportional to ΔZ/Z, where Z is the distance from the object to the pin hole and ΔZ the out-of-plane translation displacement, and (a-2) due to out-of-plane rotation are shown to be a function of both rotation angle and the image distance Z; (b) the telecentric lens has an effective object distance, Z_eff, that is 50× larger than the 55 mm standard lens, with a corresponding reduction in strain errors from 1250 μs/mm of out-of-plane motion to 25 μs/mm; and (c) a stereovision system measures all components of displacement without introducing measurable, full-field, strain errors, even though an object may undergo appreciable out-of-plane translation and rotation.     

Quantification of low-velocity motion using a navigator-echo supported MR velocity-mapping technique: Application to intracranial dynamics in volunteers and patients with brain tumours

Gradient-echo pulse sequences with velocity-encoding gradients of 22.5–25 mT/m, were used for brainmotion and CSF-flow studies. To reduce motion artifacts, a phase-correction technique based on navigator echoes was evaluated. Three patients with right-sided parietal tumours were investigated; one astrocytoma grade III–IV, one astrocytoma grade I–II and one benign meningioma. In healthy volunteers, a maximal brain-tissue velocity of (0.94 ± 0.26) mm/s (mean ± 1SD) was observed, which is consistent with previously presented results. The phase correction was proven useful for reduction of artifacts due to external head movements in modulus and phase images, without loss of phase information related to internal motion. The tissue velocity within the astrocytomas was low during the entire cardiac cycle. An abnormally high rostral velocity component was, however, observed in the brain tissue frontal to the astrocytomas. In all patients, an abnormal CSF flow pattern was observed. The study of brain motion may provide further understanding of the effects of tumours and other pathological conditions in the brain. When considering intracranial motion as a source of error in diffusion/perfusion MRI, the present study suggests that a pathology can alter the properties of brain motion and CSF flow considerably, leading to a more complex impact on diffusion/perfusion images.     

meta-DENSE complex acquisition for reduced intravoxel dephasing

Displacement encoding with stimulated echoes (DENSE) with a meta-DENSE readout and RF phase cycling to suppress the STEAM anti-echo is described for reducing intravoxel dephasing signal loss. This RF phase cycling scheme, when combined with existing meta-DENSE suppression of the T1 recovering signal, yields higher quality DENSE myocardial strain maps. Phantom and human images are provided to demonstrate the technique, which is capable of acquiring phase contrast displacement encoded images at low encoding gradient strengths providing better spatial resolution and less signal loss due to intravoxel dephasing than prior methods.     

Optimal techniques for magnetic resonance imaging of the liver using a respiratory navigator-gated three-dimensional spoiled gradient-recalled echo sequence

Purpose

To optimize the navigator-gating technique for the acquisition of high-quality three-dimensional spoiled gradient-recalled echo (3D SPGR) images of the liver during free breathing.

Materials and methods

Ten healthy volunteers underwent 3D SPGR magnetic resonance imaging of the liver using a conventional navigator-gated 3D SPGR (cNAV-3D-SPGR) sequence or an enhanced navigator-gated 3D SPGR (eNAV-3D-SPGR) sequence. No exogenous contrast agent was used. A 20-ms wait period was inserted between the 3D SPGR acquisition component and navigator component of the eNAV-3D-SPGR sequence to allow T1 recovery. Visual evaluation and calculation of the signal-to-noise ratio were performed to compare image quality between the imaging techniques.

Result

The eNAV-3D-SPGR sequence provided better noise properties than the cNAV-3D-SPGR sequence visually and quantitatively. Navigator gating with an acceptance window of 2 mm effectively inhibited respiratory motion artifacts. The widening of the window to 6 mm shortened the acquisition time but increased motion artifacts, resulting in degradation of overall image quality. Neither slice tracking nor incorporation of short breath holding successfully compensated for the widening of the window.

Conclusion

The eNAV-3D-SPGR sequence with an acceptance window of 2 mm provides high-quality 3D SPGR images of the liver.     

Relationship between diffusion-weighted MR images,cerebral blood flow,and energy state in experimental brain infarction

The regional evolution of brain infarction was studied in Wistar rats submitted to remotely controlled thread occlusion of the middle cerebral artery. Occlusion was performed in the magnet of an NMR tomography system to allow continuous recording of diffusion-weighted images. After 30 min (n = 6) or 2 h (n = 9), cerebral blood flow was measured by [¹⁴C] iodoantipyrine autoradiography while the regional distribution of ATP, glucose, lactate, and pH was imaged using pictorial bioluminescence and fluoroscopic methods. In diffusion-weighted images, the hemispheric lesion area (HLA) at the level of caudate-putamen amounted to 54.2 ± 10.9% after 30 min and to 67.0 ± 5.9% after 2 h vascular occlusion. These areas corresponded to the regions exhibiting tissue acidosis (60.8 ± 9.3% and 70.4 ± 4.5%), but were clearly larger than those in which ATP was depleted (22.3 ± 20.8% and 49.6 ± 12.9% after 30 min and 2 h, respectively). The threshold of blood flow for the increase of signal intensity in diffusion-weighted images increased between 30 min and 2 h occlusion from 34 to 41 ml/100 g per minute, the threshold of acidosis from 40 to 47 ml/100 g per minute, and the threshold for ATP depletion from 13 to 19 ml/100 g per minute. Our study demonstrates that diffusion-weighted imaging detects both the core and the penumbra of the evolving infarction but is not able to differentiate between the two parts. It further shows that the ischemic lesion grows during the initial 2 h of vascular occlusion, and that the size of the infarct core increases more rapidly than that of the penumbra.     

2D strain measurement in sub-10 nm SiGe layer with dark-field electron holography

In this paper, we carried out the two-dimensional (2D) strain measurement in sub-10 nm SiGe layer; images were obtained by dark-field electron holography (DFEH). This technique is based on transmission electron microscopy (TEM), in which dark-field holograms were obtained from a (400) diffraction spot. The measured results were compared to the X-ray diffraction (XRD) results in terms of the strain value and the depth of strain distribution in a very thin SiGe layer. Subsequently, we were able to successfully analyze the 2D strain maps along the [100] growth direction of the nanoscale SiGe region. The strain was measured and found to be in the range of 1.8–2.4%. The strain precision was estimated at 2.5 × 10⁻³. As a result, the DFEH technique is truly useful for measuring 2D strain maps in very thin SiGe layers with nanometer resolution and high precision.     

Shear strain estimation and lesion mobility assessment in elastography

Elastography typically measures and images the normal strain component along the insonification/compression axis, i.e., in the axial direction. We have recently shown that, by using interpolation and cross-correlation methods of transversely displaced RF echo segments, it is possible to measure and image displacement and strain transversely to the beam with good precision. This enables the estimation and imaging of all three principal normal strain components. Generally, motion in a direction other than that in which strain is estimated may result in decorrelation noise, severely corrupting the estimates. Therefore, a correction method is applied to correct the displacement and strain estimates for decorrelating motion. In this paper, we show how corrected displacement estimates can also be used to estimate and image the shear strain components. This may allow us to identify regions of decorrelation noise in the normal strain measurement that are due to shear strain. Shear strain estimates provide supplementary information, which can characterize different tissue elements based on their mobility. In the case of breast lesions, low mobility is related to malignancy. Following an in vivo case, we show with 2D simulations how assessment of tumor mobility can be achieved with shear strain estimation.     

Detection of muscle layer invasion with submillimeter pixel MR images: Staging of bladder carcinoma

Conventional magnetic resonance (MR) images used for the pelvic organs generally have a pixel size of 1.3 mm × 1.3 mm. We obtained images with a higher resolution than conventional images, and evaluated the usefulness of this type of image in staging urinary bladder carcinoma. Twenty-six patients having either transurethral resection of bladder tumor (TUR-BT) or cystectomy were retrospectively studied. T₂-weighted images of the bladder were acquired with a 20 cm field-of-view, a matrix size of 224 × 224 (pixel size: 0.9 mm × 0.9 mm), and a slice thickness of 7 mm using a 0.5 T system. MR appearances of the carcinoma were divided into the following five categories: no abnormality found on the inner surface of the bladder wall (0), high signal layer or protrusion without breakage of the wall (I), partial disruption of the wall (II), transmural disruption of the wall (III), and complete disruption with mass formations in the perivesicular region (IV). These findings were correlated with the TNM pathologic staging determined from each tissue specimen. A prediction for muscle layer invasion was calculated by selecting pathologic stage pT2 and MR category III as a boundary measure. The accuracy was 96.2%, the sensitivity 100.0%, and the specificity 91.7%. The results obtained indicate that submillimeter pixel MR imaging shows promise as a noninvasive method for the preoperative staging of bladder cancers.     

A novel CNN algorithm for pathological complete response prediction using an I-SPY TRIAL breast MRI database

PurposeTo apply our convolutional neural network (CNN) algorithm to predict neoadjuvant chemotherapy (NAC) response using the I-SPY TRIAL breast MRI dataset.MethodsFrom the I-SPY TRIAL breast MRI database, 131 patients from 9 institutions were successfully downloaded for analysis. First post-contrast MRI images were used for 3D segmentation using 3D slicer. Our CNN was implemented entirely of 3 × 3 convolutional kernels and linear layers. The convolutional kernels consisted of 6 residual layers, totaling 12 convolutional layers. Dropout with a 0.5 keep probability and L2 normalization was utilized. Training was implemented by using the Adam optimizer. A 5-fold cross validation was used for performance evaluation. Software code was written in Python using the TensorFlow module on a Linux workstation with one NVidia Titan X GPU.ResultsOf 131 patients, 40 patients achieved pCR following NAC (group 1) and 91 patients did not achieve pCR following NAC (group 2). Diagnostic accuracy of our CNN two classification model distinguishing patients with pCR vs non-pCR was 72.5 (SD ± 8.4), with sensitivity 65.5% (SD ± 28.1) and specificity of 78.9% (SD ± 15.2). The area under a ROC Curve (AUC) was 0.72 (SD ± 0.08).ConclusionIt is feasible to use our CNN algorithm to predict NAC response in patients using a multi-institution dataset.     

Reduction in radiation dose with reconstruction technique in the brain perfusion CT

The principal objective of this study was to verify the utility of the reconstruction imaging technique in the brain perfusion computed tomography (PCT) scan by assessing reductions in the radiation dose and analyzing the generated images. The setting used for image acquisition had a detector coverage of 40 mm, a helical thickness of 0.625 mm, a helical shuttle mode scan type and a rotation time of 0.5 s as the image parameters used for the brain PCT scan. Additionally, a phantom experiment and an animal experiment were carried out. In the phantom and animal experiments, noise was measured in the scanning with the tube voltage fixed at 80 kVp (kilovolt peak) and the level of the adaptive statistical iterative reconstruction (ASIR) was changed from 0% to 100% at 10% intervals. The standard deviation of the CT coefficient was measured three times to calculate the mean value. In the phantom and animal experiments, the absorbed dose was measured 10 times under the same conditions as the ones for noise measurement before the mean value was calculated. In the animal experiment, pencil-type and CT-dedicated ionization chambers were inserted into the central portion of pig heads for measurement. In the phantom study, as the level of the ASIR changed from 0% to 100% under identical scanning conditions, the noise value and dose were proportionally reduced. In our animal experiment, the noise value was lowest when the ASIR level was 50%, unlike in the phantom study. The dose was reduced as in the phantom study.