Guk Bae Kim

Deep Learning in Medical Imaging: General Overview

The artificial neural network (ANN)–a machine learning technique inspired by the human neuronal synapse system–was introduced in the 1950s. However, the ANN was previously limited in its ability to solve actual problems, due to the vanishing gradient and overfitting problems with training of deep architecture, lack of computing power, and primarily the absence of sufficient data to train the computer system. Interest in this concept has lately resurfaced, due to the availability of big data, enhanced computing power with the current graphics processing units, and novel algorithms to train the deep neural network. Recent studies on this technology suggest its potentially to perform better than humans in some visual and auditory recognition tasks, which may portend its applications in medicine and healthcare, especially in medical imaging, in the foreseeable future. This review article offers perspectives on the history, development, and applications of deep learning technology, particularly regarding its applications in medical imaging.

Three-Dimensional Printing: Basic Principles and Applications in Medicine and Radiology

The advent of three-dimensional printing (3DP) technology has enabled the creation of a tangible and complex 3D object that goes beyond a simple 3D-shaded visualization on a flat monitor. Since the early 2000s, 3DP machines have been used only in hard tissue applications. Recently developed multi-materials for 3DP have been used extensively for a variety of medical applications, such as personalized surgical planning and guidance, customized implants, biomedical research, and preclinical education. In this review article, we discuss the 3D reconstruction process, touching on medical imaging, and various 3DP systems applicable to medicine. In addition, the 3DP medical applications using multi-materials are introduced, as well as our recent results.

Multi-VENC acquisition of four-dimensional phase-contrast MRI to improve precision of velocity field measurement

The present study aims to improve precision of four‐dimensional (4D) phase‐contrast (PC) MRI technique by using multiple velocity encoding (VENC) parameters.

Turbulent Kinetic Energy Measurement Using Phase Contrast MRI for Estimating the Post-Stenotic Pressure Drop: In Vitro Validation and Clinical Application

Background Although the measurement of turbulence kinetic energy (TKE) by using magnetic resonance imaging (MRI) has been introduced as an alternative index for quantifying energy loss through the cardiac valve, experimental verification and clinical application of this parameter are still required. Objectives The goal of this study is to verify MRI measurements of TKE by using a phantom stenosis with particle image velocimetry (PIV) as the reference standard. In addition, the feasibility of measuring TKE with MRI is explored. Methods MRI measurements of TKE through a phantom stenosis was performed by using clinical 3T MRI scanner. The MRI measurements were verified experimentally by using PIV as the reference standard. In vivo application of MRI-driven TKE was explored in seven patients with aortic valve disease and one healthy volunteer. Transvalvular gradients measured by MRI and echocardiography were compared. Results MRI and PIV measurements of TKE are consistent for turbulent flow (0.666 < R2 < 0.738) with a mean difference of −11.13 J/m3 (SD = 4.34 J/m3). Results of MRI and PIV measurements differ by 2.76 ± 0.82 cm/s (velocity) and −11.13 ± 4.34 J/m3 (TKE) for turbulent flow (Re > 400). The turbulence pressure drop correlates strongly with total TKE (R2 = 0.986). However, in vivo measurements of TKE are not consistent with the transvalvular pressure gradient estimated by echocardiography. Conclusions These results suggest that TKE measurement via MRI may provide a potential benefit as an energy-loss index to characterize blood flow through the aortic valve. However, further clinical studies are necessary to reach definitive conclusions regarding this technique.

Post-stenotic plug-like jet with a vortex ring demonstrated by 4D flow MRI

PURPOSE To investigate the details of the flow structure of a plug-like jet that had a vortex ring in pulsatile stenotic phantoms using 4D flow MRI. METHOD Pulsatile Newtonian flows in two stenotic phantoms with 50% and 75% reductions in area were scanned by 4D flow MRI. Blood analog working fluid was circulated via the stenotic phantom using a pulsatile pump at a constant pulsating frequency of 1Hz. The velocity and vorticity fields of the plug-like jet with a vortex ring were quantitatively analyzed in the spatial and temporal domains. RESULTS Pulsatile stenotic flow showed a plug-like jet at the specific stenotic degree of 50% in our pulsatile waveform design. This plug-like jet was found at the decelerating period in the post-stenotic region of 26.4mm (1.2 D). It revealed a vortex ring structure with vorticity strength in the range of ±100s(-1). CONCLUSION We observed a plug-like jet with a vortex ring in pulsatile stenotic flow by in vitro visualization using 4D flow MRI. In this plug-like jet, the local fastest flow region occurred at the post-systole phase in the post-stenotic region, which was distinguishable from a typical stenotic jet flow at systole phase.

Hemodynamic Measurement Using Four-Dimensional Phase-Contrast MRI: Quantification of Hemodynamic Parameters and Clinical Applications

Recent improvements have been made to the use of time-resolved, three-dimensional phase-contrast (PC) magnetic resonance imaging (MRI), which is also named four-dimensional (4D) PC-MRI or 4D flow MRI, in the investigation of spatial and temporal variations in hemodynamic features in cardiovascular blood flow. The present article reviews the principle and analytical procedures of 4D PC-MRI. Various fluid dynamic biomarkers for possible clinical usage are also described, including wall shear stress, turbulent kinetic energy, and relative pressure. Lastly, this article provides an overview of the clinical applications of 4D PC-MRI in various cardiovascular regions.

Validation of a CT-guided intervention robot for biopsy and radiofrequency ablation: experimental study with an abdominal phantom

PURPOSE We aimed to evaluate the accuracy of a needle-placement robot for biopsy and radiofrequency ablation on an abdominal phantom. METHODS A master-slave robotic system has been developed that includes a needle-path planning system and a needle-inserting robot arm with computed tomography (CT) and CT fluoroscopy guidance. For evaluation of its accuracy in needle placement, a commercially available abdominal phantom (Model 057A; CIRS Inc.) was used. The liver part of the phantom contains multiple spherical simulated tumors of three different size spheres. Various needle insertion trials were performed in the transverse plane and caudocranial plane two nodule sizes (10 mm and 20 mm in diameter) to test the reliability of this robot. To assess accuracy, a CT scan was performed after each trial with the needle in situ. RESULTS The overall error was 2 mm (0-2.6 mm), which was calculated as the distance from the planned trajectory before insertion to the actual needle trajectory after insertion. The standard deviations of the insertions on two nodules (10 mm and 20 mm in diameter) were 0.5 mm and 0.2 mm, respectively. CONCLUSION The CT-compatible needle placement robot for biopsy and radiofrequency ablation shows relatively acceptable accuracy and could be used for radiofrequency ablation of nodules ≥10 mm under CT fluoroscopy guidance.

The influence of the aortic valve angle on the hemodynamic features of the thoracic aorta

Since the first observation of a helical flow pattern in aortic blood flow, the existence of helical blood flow has been found to be associated with various pathological conditions such as bicuspid aortic valve, aortic stenosis, and aortic dilatation. However, an understanding of the development of helical blood flow and its clinical implications are still lacking. In our present study, we hypothesized that the direction and angle of aortic inflow can influence helical flow patterns and related hemodynamic features in the thoracic aorta. Therefore, we investigated the hemodynamic features in the thoracic aorta and various aortic inflow angles using patient-specific vascular phantoms that were generated using a 3D printer and time-resolved, 3D, phase-contrast magnetic resonance imaging (PC-MRI). The results show that the rotational direction and strength of helical blood flow in the thoracic aorta largely vary according to the inflow direction of the aorta, and a higher helical velocity results in higher wall shear stress distributions. In addition, right-handed rotational flow conditions with higher rotational velocities imply a larger total kinetic energy than left-handed rotational flow conditions with lower rotational velocities.

Estimation of turbulent kinetic energy using 4D phase-contrast MRI: Effect of scan parameters and target vessel size

Quantifying turbulence velocity fluctuation is important because it indicates the fluid energy dissipation of the blood flow, which is closely related to the pressure drop along the blood vessel. This study aims to evaluate the effects of scan parameters and the target vessel size of 4D phase-contrast (PC)-MRI on quantification of turbulent kinetic energy (TKE). Comprehensive 4D PC-MRI measurements with various velocity-encoding (VENC), echo time (TE), and voxel size values were carried out to estimate TKE distribution in stenotic flow. The total TKE (TKEsum), maximum TKE (TKEmax), and background noise level (TKEnoise) were compared for each scan parameter. The feasibility of TKE estimation in small vessels was also investigated. Results show that the optimum VENC for stenotic flow with a peak velocity of 125cm/s was 70cm/s. Higher VENC values overestimated the TKEsum by up to six-fold due to increased TKEnoise, whereas lower VENC values (30cm/s) underestimated it by 57.1%. TE and voxel size did not significantly influence the TKEsum and TKEnoise, although the TKEmax significantly increased as the voxel size increased. TKE quantification in small-sized vessels (3-5-mm diameter) was feasible unless high-velocity turbulence caused severe phase dispersion in the reference image.

Post-stenotic Recirculating Flow May Cause Hemodynamic Perforator Infarction

Background and Purpose The primary mechanism underlying paramedian pontine infarction (PPI) is atheroma obliterating the perforators. Here, we encountered a patient with PPI in the post-stenotic area of basilar artery (BA) without a plaque, shown by high-resolution magnetic resonance imaging (HR-MRI). We performed an experiment using a 3D-printed BA model and a particle image velocimetry (PIV) to explore the hemodynamic property of the post-stenotic area and the mechanism of PPI. Methods 3D-model of a BA stenosis was reconstructed with silicone compound using a 3D-printer based on the source image of HR-MRI. Working fluid seeded with fluorescence particles was used and the velocity of those particles was measured horizontally and vertically. Furthermore, microtubules were inserted into the posterior aspect of the model to measure the flow rates of perforators (pre-and post-stenotic areas). The flow rates were compared between the microtubules. Results A recirculating flow was observed from the post-stenotic area in both directions forming a spiral shape. The velocity of the flow in these regions of recirculation was about one-tenth that of the flow in other regions. The location of recirculating flow well corresponded with the area with low-signal intensity at the time-of-flight magnetic resonance angiography and the location of PPI. Finally, the flow rate through the microtubule inserted into the post-stenotic area was significantly decreased comparing to others (P<0.001). Conclusions Perforator infarction may be caused by a hemodynamic mechanism altered by stenosis that induces a recirculation flow. 3D-printed modeling and PIV are helpful understanding the hemodynamics of intracranial stenosis.