Bayesian Networks in Radiology.

A Bayesian network is a graphical model that uses probability theory to represent relationships among its variables. The model is a directed acyclic graph whose nodes represent variables, such as the presence of a disease or an imaging finding. Connections between nodes express causal influences between variables as probability values. Bayesian networks can learn their structure (nodes and connections) and/or conditional probability values from data. Bayesian networks offer several advantages: (a) they can efficiently perform complex inferences, (b) reason from cause to effect or vice versa, (c) assess counterfactual data, (d) integrate observations with canonical ("textbook") knowledge, and (e) explain their reasoning. Bayesian networks have been employed in a wide variety of applications in radiology, including diagnosis and treatment planning. Unlike deep learning approaches, Bayesian networks have not been applied to computer vision. However, hybrid artificial intelligence systems have combined deep learning models with Bayesian networks, where the deep learning model identifies findings in medical images and the Bayesian network formulates and explains a diagnosis from those findings. One can apply a Bayesian network's probabilistic knowledge to integrate clinical and imaging findings to support diagnosis, treatment planning, and clinical decision-making. This article reviews the fundamental principles of Bayesian networks and summarizes their applications in radiology. Keywords: Bayesian Network, Machine Learning, Abdominal Imaging, Musculoskeletal Imaging, Breast Imaging, Neurologic Imaging, Radiology Education Supplemental material is available for this article. © RSNA, 2023.

Three-dimensional printing of patient-specific computed tomography lung phantoms: a reader study.

In modern clinical decision-support algorithms, heterogeneity in image characteristics due to variations in imaging systems and protocols hinders the development of reproducible quantitative measures including for feature extraction pipelines. With the help of a reader study, we investigate the ability to provide consistent ground-truth targets by using patient-specific 3D-printed lung phantoms. PixelPrint was developed for 3D-printing lifelike computed tomography (CT) lung phantoms by directly translating clinical images into printer instructions that control density on a voxel-by-voxel basis. Data sets of three COVID-19 patients served as input for 3D-printing lung phantoms. Five radiologists rated patient and phantom images for imaging characteristics and diagnostic confidence in a blinded reader study. Effect sizes of evaluating phantom as opposed to patient images were assessed using linear mixed models. Finally, PixelPrint's production reproducibility was evaluated. Images of patients and phantoms had little variation in the estimated mean (0.03-0.29, using a 1-5 scale). When comparing phantom images to patient images, effect size analysis revealed that the difference was within one-third of the inter- and intrareader variabilities. High correspondence between the four phantoms created using the same patient images was demonstrated by PixelPrint's production repeatability tests, with greater similarity scores between high-dose acquisitions of the phantoms than between clinical-dose acquisitions of a single phantom. We demonstrated PixelPrint's ability to produce lifelike CT lung phantoms reliably. These phantoms have the potential to provide ground-truth targets for validating the generalizability of inference-based decision-support algorithms between different health centers and imaging protocols and for optimizing examination protocols with realistic patient-based phantoms. Classification: CT lung phantoms, reader study.

A CT Algorithm Can Elevate the Differential Diagnosis of Interstitial Lung Disease by Non-specialists to Equal That of Specialist Thoracic Radiologists.

BACKGROUND: Diagnosis of diffuse parenchymal lung diseases (DPLD) on high resolution CT (HRCT) is difficult for non-expert radiologists due to varied presentation for any single disease and overlap in presentation between diseases. RATIONALE AND OBJECTIVES: To evaluate whether a pattern-based training algorithm can improve the ability of non-experts to diagnosis of DPLD. MATERIALS AND METHODS: Five experts (cardiothoracic-trained radiologists), and 25 non-experts (non-cardiothoracic-trained radiologists, radiology residents, and pulmonologists) were each assigned a semi-random subset of cases from a compiled database of DPLD HRCTs. Each reader was asked to create a top three differential for each case. The non-experts were then given a pattern-based training algorithm for identifying DPLDs. Following training, the non-experts were again asked to create a top three differential for each case that they had previously evaluated. Accuracy between groups was compared using Chi-Square analysis. RESULTS: A total of 400 and 1450 studies were read by experts and non-experts, respectively. Experts correctly placed the diagnosis as the first item on the differential versus having the correct diagnosis as one of their top three diagnoses at an overall rate of 48 and 64.3%, respectively. Pre-training, non-experts achieved a correct diagnosis/top three of 32.5 and 49.7%, respectively. Post-training, non-experts demonstrated a correct diagnosis/top three of 41.2 and 65%, a statistically significant increase (p < 0.0001). In addition, post training, there was no difference between non-experts and experts in placing the correct diagnosis within their top three differential. CONCLUSION: The diagnosis of DPLDs by HRCT imaging alone is relatively poor. However, use of a pattern-based teaching algorithm can improve non-expert interpretation and enable non-experts to include the correct diagnosis within their differential diagnoses at a rate comparable to expert cardiothoracic trained radiologists.

Microbubble-mediated intravascular ultrasound imaging and drug delivery.

Intravascular ultrasound (IVUS) provides radiation-free, real-time imaging and assessment of atherosclerotic disease in terms of anatomical, functional, and molecular composition. The primary clinical applications of IVUS imaging include assessment of luminal plaque volume and real-time image guidance for stent placement. When paired with microbubble contrast agents, IVUS technology may be extended to provide nonlinear imaging, molecular imaging, and therapeutic delivery modes. In this review, we discuss the development of emerging imaging and therapeutic applications that are enabled by the combination of IVUS imaging technology and microbubble contrast agents.

Reducing Neointima Formation in a Swine Model with IVUS and Sirolimus Microbubbles.

Potent therapeutic compounds with dose dependent side effects require more efficient and selective drug delivery to reduce systemic drug doses. Here, we demonstrate a new platform that combines intravascular ultrasound (IVUS) and drug-loaded microbubbles to enhance and localize drug delivery, while enabling versatility of drug type and dosing. Localization and degree of delivery with IVUS and microbubbles was assessed using fluorophore-loaded microbubbles and different IVUS parameters in ex vivo swine arteries. Using a swine model of neointimal hyperplasia, reduction of neointima formation following balloon injury was evaluated when using the combination of IVUS and sirolimus-loaded microbubbles. IVUS and microbubble enhanced fluorophore delivery was greatest when applying low amplitude pulses in the ex vivo model. In the in vivo model, neointima formation was reduced by 50% after treatment with IVUS and the sirolimus-loaded microbubbles. This reduction was achieved with a sirolimus whole blood concentration comparable to a commercial drug-eluting stent (0.999 ng/mL). We anticipate this therapy will find clinical use localizing drug delivery for numerous other diseases in addition to serving as an adjunct to stents in treating atherosclerosis.

In vivo imaging of microfluidic-produced microbubbles.

Microfluidics-based production of stable microbubbles for ultrasound contrast enhancement or drug/gene delivery allows for precise control over microbubble diameter but at the cost of a low production rate. In situ microfluidic production of microbubbles directly in the vasculature may eliminate the necessity for high microbubble production rates, long stability, or small diameters. Towards this goal, we investigated whether microfluidic-produced microbubbles directly administered into a mouse tail vein could provide sufficient ultrasound contrast. Microbubbles composed of nitrogen gas and stabilized with 3 % bovine serum albumin and 10 % dextrose were injected for 10 seconds into wild type C57BL/6 mice, via a tail-vein catheter. Short-axis images of the right and left ventricle were acquired at 12.5 MHz and image intensity over time was analyzed. Microbubbles were produced on the order of 10(5) microbubbles/s and were observed in both the right and left ventricles. The median rise time, duration, and decay time within the right ventricle were 2.9, 21.3, and 14.3 s, respectively. All mice survived the procedure with no observable respiratory or heart rate distress despite microbubble diameters as large as 19 µm.

Synthesis and characterization of transiently stable albumin-coated microbubbles via a flow-focusing microfluidic device.

We describe a method for synthesizing albumin-shelled, large-diameter (>10 µm), transiently stable microbubbles using a flow-focusing microfluidic device (FFMD). The microfluidic device enables microbubbles to be produced immediately before insonation, thus relaxing the requirements for stability. Both reconstituted fractionated bovine serum albumin (BSA) and fresh bovine blood plasma were investigated as shell stabilizers. Microbubble coalescence was inhibited by the addition of either dextrose or glycerol and propylene glycol. Microbubbles were observed to have an acoustic half-life of approximately 6 s. Microbubbles generated directly within a vessel phantom containing flowing blood produced a 6.5-dB increase in acoustic signal within the lumen. Microbubbles generated in real time upstream of in vitro rat aortic smooth muscle cells under physiologic flow conditions successfully permeabilized 58% of the cells on insonation at a peak negative pressure of 200 kPa. These results indicate that transiently stable microbubbles produced via flow-focusing microfluidic devices are capable of image enhancement and drug delivery. In addition, successful microbubble production with blood plasma suggests the potential to use blood as a stabilizing shell.

Enhanced intracellular delivery of a model drug using microbubbles produced by a microfluidic device.

Focal drug delivery to a vessel wall facilitated by intravascular ultrasound and microbubbles holds promise as a potential therapy for atherosclerosis. Conventional methods of microbubble administration result in rapid clearance from the bloodstream and significant drug loss. To address these limitations, we evaluated whether drug delivery could be achieved with transiently stable microbubbles produced in real time and in close proximity to the therapeutic site. Rat aortic smooth muscle cells were placed in a flow chamber designed to simulate physiological flow conditions. A flow-focusing microfluidic device produced 8 µm diameter monodisperse microbubbles within the flow chamber, and ultrasound was applied to enhance uptake of a surrogate drug (calcein). Acoustic pressures up to 300 kPa and flow rates up to 18 mL/s were investigated. Microbubbles generated by the flow-focusing microfluidic device were stabilized with a polyethylene glycol-40 stearate shell and had either a perfluorobutane (PFB) or nitrogen gas core. The gas core composition affected stability, with PFB and nitrogen microbubbles exhibiting half-lives of 40.7 and 18.2 s, respectively. Calcein uptake was observed at lower acoustic pressures with nitrogen microbubbles (100 kPa) than with PFB microbubbles (200 kPa) (p < 0.05, n > 3). In addition, delivery was observed at all flow rates, with maximal delivery (>70% of cells) occurring at a flow rate of 9 mL/s. These results demonstrate the potential of transiently stable microbubbles produced in real time and in close proximity to the intended therapeutic site for enhancing localized drug delivery.

Production rate and diameter analysis of spherical monodisperse microbubbles from two-dimensional, expanding-nozzle flow-focusing microfluidic devices.

Flow-focusing microfluidic devices (FFMDs) can produce microbubbles (MBs) with precisely controlled diameters and a narrow size distribution. In this paper, poly-dimethyl-siloxane based, rectangular-nozzle, two-dimensional (2-D) planar, expanding-nozzle FFMDs were characterized using a high speed camera to determine the production rate and diameter of Tween 20 (2% v/v) stabilized MBs. The effect of gas pressure and liquid flow rate on MB production rate and diameter was analyzed in order to develop a relationship between FFMD input parameters and MB production. MB generation was observed to transition through five regimes at a constant gas pressure and increasing liquid flow rate. Each MB generation event (i.e., break-off to break-off) was further separated into two characteristic phases: bubbling and waiting. The duration of the bubbling phase was linearly related to the liquid flow rate, while the duration of the waiting phase was related to both liquid flow rate and gas pressure. The MB production rate was found to be inversely proportional to the sum of the bubbling and waiting times, while the diameter was found to be proportional to the product of the gas pressure and bubbling time.

Real-time targeted molecular imaging using singular value spectra properties to isolate the adherent microbubble signal.

Ultrasound-based real-time molecular imaging in large blood vessels holds promise for early detection and diagnosis of various important and significant diseases, such as stroke, atherosclerosis, and cancer. Central to the success of this imaging technique is the isolation of ligand-receptor bound adherent microbubbles from free microbubbles and tissue structures. In this paper, we present a new approach, termed singular spectrum-based targeted molecular (SiSTM) imaging, which separates signal components using singular value spectra content over local regions of complex echo data. Simulations were performed to illustrate the effects of acoustic target motion and harmonic energy on SiSTM imaging-derived measurements of statistical dimensionality. In vitro flow phantom experiments were performed under physiologically realistic conditions (2.7 cm s⁻¹ flow velocity and 4 mm diameter) with targeted and non-targeted phantom channels. Both simulation and experimental results demonstrated that the relative motion and harmonic characteristics of adherent microbubbles (i.e. low motion and large harmonics) yields echo data with a dimensionality that is distinct from free microbubbles (i.e. large motion and large harmonics) and tissue (i.e. low motion and low harmonics). Experimental SiSTM images produced the expected trend of a greater adherent microbubble signal in targeted versus non-targeted microbubble experiments (P < 0.05, n = 4). The location of adherent microbubbles was qualitatively confirmed via optical imaging of the fluorescent DiI signal along the phantom channel walls after SiSTM imaging. In comparison with two frequency-based real-time molecular imaging strategies, SiSTM imaging provided significantly higher image contrast (P < 0.001, n = 4) and a larger area under the receiver operating characteristic curve (P < 0.05, n = 4).

Assessing and improving acoustic radiation force image quality using a 1.5-D transducer design.

A 1.5-D transducer array was proposed to improve acoustic radiation force impulse (ARFI) imaging signal-to-noise ratio (SNRARFI) and image contrast relative to a conventional 1-D array. To predict performance gains from the proposed 1.5-D transducer array, an analytical model for SNRARFI upper bound was derived. The analytical model and 1.5-D ARFI array were validated using a finite element modelbased numerical simulation framework. The analytical model demonstrated good agreement with numerical results (correlation coefficient = 0.995), and simulated lesion images yielded a significant (2.92 dB; p < 0.001) improvement in contrast-tonoise ratio when rendered using the 1.5-D ARFI array.

Focused ultrasound-mediated drug delivery from microbubbles reduces drug dose necessary for therapeutic effect on neointima formation--brief report.

OBJECTIVE: We hypothesized that (1) neointima formation in a rat carotid balloon injury model could be reduced in vivo following targeted ultrasound delivery of rapamycin microbubbles (RMBs), and (2) the addition of dual-mode ultrasound decreases the total amount of drug needed to reduce neointima formation. METHODS AND RESULTS: Balloon injury was performed in rat carotids to induce neointima formation. High or low doses of RMBs were injected intravenously and ruptured at the site of injury with ultrasound. Compared with nontreated injured arteries, neointima formation was reduced by 0% and 35.9% with 10(8) RMBs and by 28.7% and 34.9% in arteries treated with 10(9) RMBs with and without ultrasound, respectively. CONCLUSIONS: Without ultrasound, 10-fold higher concentrations of RMBs were needed to reduce neointima formation by at least 28%, whereas 10(8) RMBs combined with ultrasound were sufficient to achieve the same therapeutic effect, demonstrating that this technology may have promise for localized potent drug therapy.