Human AI Teaming for Coronary CT Angiography Assessment: Impact on Imaging Workflow and Diagnostic Accuracy.

As the number of coronary computed tomography angiography (CTA) examinations is expected to increase, technologies to optimize the imaging workflow are of great interest. The aim of this study was to investigate the potential of artificial intelligence (AI) to improve clinical workflow and diagnostic accuracy in high-volume cardiac imaging centers. A total of 120 patients (79 men; 62.4 (55.0-72.7) years; 26.7 (24.9-30.3) kg/m2) undergoing coronary CTA were randomly assigned to a standard or an AI-based (human AI) coronary analysis group. Severity of coronary artery disease was graded according to CAD-RADS. Initial reports were reviewed and changes were classified. Both groups were similar with regard to age, sex, body mass index, heart rate, Agatston score, and CAD-RADS. The time for coronary CTA assessment (142.5 (106.5-215.0) s vs. 195.0 (146.0-265.5) s; p < 0.002) and the total reporting time (274.0 (208.0-377.0) s vs. 350 (264.0-445.5) s; p < 0.02) were lower in the human AI than in the standard group. The number of cases with no, minor, or CAD-RADS relevant changes did not differ significantly between groups (52, 7, 1 vs. 50, 8, 2; p = 0.80). AI-based analysis significantly improves clinical workflow, even in a specialized high-volume setting, by reducing CTA analysis and overall reporting time without compromising diagnostic accuracy.

How scan parameter choice affects deep learning-based coronary artery disease assessment from computed tomography.

Recently, algorithms capable of assessing the severity of Coronary Artery Disease (CAD) in form of the Coronary Artery Disease-Reporting and Data System (CAD-RADS) grade from Coronary Computed Tomography Angiography (CCTA) scans using Deep Learning (DL) were proposed. Before considering to apply these algorithms in clinical practice, their robustness regarding different commonly used Computed Tomography (CT)-specific image formation parameters-including denoising strength, slab combination, and reconstruction kernel-needs to be evaluated. For this study, we reconstructed a data set of 500 patient CCTA scans under seven image formation parameter configurations. We select one default configuration and evaluate how varying individual parameters impacts the performance and stability of a typical algorithm for automated CAD assessment from CCTA. This algorithm consists of multiple preprocessing and a DL prediction step. We evaluate the influence of the parameter changes on the entire pipeline and additionally on only the DL step by propagating the centerline extraction results of the default configuration to all others. We consider the standard deviation of the CAD severity prediction grade difference between the default and variation configurations to assess the stability w.r.t. parameter changes. For the full pipeline we observe slight instability (± 0.226 CAD-RADS) for all variations. Predictions are more stable with centerlines propagated from the default to the variation configurations (± 0.122 CAD-RADS), especially for differing denoising strengths (± 0.046 CAD-RADS). However, stacking slabs with sharp boundaries instead of mixing slabs in overlapping regions (called true stack ± 0.313 CAD-RADS) and increasing the sharpness of the reconstruction kernel (± 0.150 CAD-RADS) leads to unstable predictions. Regarding the clinically relevant tasks of excluding CAD (called rule-out; AUC default 0.957, min 0.937) and excluding obstructive CAD (called hold-out; AUC default 0.971, min 0.964) the performance remains on a high level for all variations. Concluding, an influence of reconstruction parameters on the predictions is observed. Especially, scans reconstructed with the true stack parameter need to be treated with caution when using a DL-based method. Also, reconstruction kernels which are underrepresented in the training data increase the prediction uncertainty.

Computer-based liver volumetry in the liver perfusion simulator.

BACKGROUND: An exact preoperative liver volume calculation is important prior to liver surgery and living-related liver transplantation. However, CT or MRI assessment of preoperative liver volume is associated with an estimation error of 1.2% to 36%, and little data is available on its accuracy on the segmental level. The aim of this study was to validate arterial and portal venous flow rates and gain information on liver volumetry, including liver segments, in the liver perfusion simulator and compare it to in vivo measurements in a porcine model. MATERIAL AND METHODS: The arterial and portal venous flow rates and liver volumes of 10 pigs were measured in vivo and compared with the flow rates and volumes ex vivo. CT scans were performed and the volume of the liver and its lobes calculated by water displacement or computer-assistance based on the CT scans. The right lateral lobe was plasticized and reconstructed for the volume calculation. RESULTS: In the liver perfusion simulator, arterial and portal venous flow rates comparable to the in vivo rates were achieved. The liver volume had a mean difference of 10.3% between in vivo and ex vivo measurements. In the liver perfusion simulator, the mean deviation in liver volume between the computer calculation and water displacement was 2.8%. On the segmental level, the Heidelberg algorithm provided an accuracy of 97.7%. CONCLUSION: The liver perfusion simulator is an excellent device for studies in liver perfusion and volumetry. Furthermore, the simulator is applicable for teaching and performing interventions and surgeries in livers.

How many CT detector rows are necessary to perform adequate three dimensional visualization?

INTRODUCTION: The technical development of computer tomography (CT) imaging has experienced great progress. As consequence, CT data to be used for 3D visualization is not only based on 4 row CTs and 16 row CTs but also on 64 row CTs, respectively. The main goal of this study was to examine whether the increased amount of CT detector rows is correlated with improved quality of the 3D images. MATERIAL AND METHODS: All CTs were acquired during routinely performed preoperative evaluation. Overall, there were 12 data sets based on 4 detector row CT, 12 data sets based on 16 detector row CT, and 10 data sets based on 64 detector row CT. Imaging data sets were transferred to the DKFZ Heidelberg using the CHILI teleradiology system. For the analysis all CT scans were examined in a blinded fashion, i.e. both the name of the patient as well as the name of the CT brand were erased. For analysis, the time for segmentation of liver, both portal and hepatic veins as well as the branching depth of portal veins and hepatic veins was recorded automatically. In addition, all results were validated in a blinded fashion based on given quality index. RESULTS: Segmentation of the liver was performed in significantly shorter time (p<0.01, Kruskal-Wallis test) in the 16 row CT (median 479 s) compared to 4 row CT (median 611 s), and 64 row CT (median 670 s), respectively. The branching depth of the portal vein did not differ significantly among the 3 different data sets (p=0.37, Kruskal-Wallis test). However, the branching depth of the hepatic veins was significantly better (p=0.028, Kruskal-Wallis test) in the 4 row CT and 16 row CT compared to 64 row CT. The grading of the quality index was not statistically different for portal veins and hepatic veins (p=0.80, Kruskal-Wallis test). Even though the total quality index was better for the vessel tree based on 64 row CT data sets (mean scale 2.6) compared to 4 CT row data (mean scale 3.25) and 16 row CT data (mean scale 3.0), these differences did not reach statistical difference (p=0.53, Kruskal-Wallis test). CONCLUSION: Even though 3D visualization is useful in operation planning, the quality of the 3D images appears to be not dependent of the number of CT detector rows.

Value of three-dimensional reconstructions in pancreatic carcinoma using multidetector CT: initial results.

AIM: To evaluate the use of three-dimensional imaging of pancreatic carcinoma using multidetector computed tomography (CT) in a prospective study. METHODS: Ten patients with suspected pancreatic tumors were examined prospectively using multidetector CT (Somatom Sensation 16, Siemens, Erlangen, Germany). The images were evaluated for the presence of a pancreatic carcinoma and invasion of the peripancreatic vessels and surrounding organs. Using the isotropic CT data sets, a three-dimensional image was created with automatic vascular analysis and semi-automatic segmentation of the organs and pancreatic tumor by a radiologist. The CT examinations and the three-dimensional images were presented to the surgeon directly before and during the patient's operation using the Medical Imaging Interaction Toolkit-based software "ReLiver". Immediately after surgery, the value of the two images was judged by the surgeon. The operation and the histological results served as the gold standard. RESULTS: Nine patients had a pancreatic carcinoma (all pT3), and one patient had a serous cystadenoma. One tumor infiltrated the superior mesenteric vein. The infiltration was correctly evaluated. All carcinomas were resectable. In comparison to the CT image with axial and coronal reconstructions, the three-dimensional image was judged by the surgeons as better for operation planning and consistently described as useful. CONCLUSION: A 3D-image of the pancreas represents an invaluable aid to the surgeon. However, the 3D-software must be further developed in order to be integrated into daily clinical routine.

Assessment of reproducibility and stability of different breath-hold maneuvres by dynamic MRI: comparison between healthy adults and patients with pulmonary hypertension.

To assess the stability and reproducibility of different breath-hold levels in healthy volunteers and patients using dynamic MRI (dMRI). In ten healthy volunteers and ten patients with pulmonary hypertension (PH) and normal lung function craniocaudal intrathoracic distances (CCD) were measured during inspiratory and expiratory breath-hold (15 s) (in healthy volunteers additionally at a self-chosen mid-inspiratory breath-hold) using dMRI (trueFISP, three images/s). To evaluate stability and intraobserver reproducibility of the different breath-hold levels, CCDs, time-distance curves, confidence intervals (CIs), Mann-Witney U test and regression equations were calculated. In healthy volunteers there was a substantial decrease of the CCD during the inspiratory breath-hold in contrast to the expiratory breath-hold. The CI at inspiration was 2.84+/-1.28 in the right and 2.1+/-0.68 in the left hemithorax. At expiration the CI was 2.54+/-1.18 and 2.8+/-1.48. Patients were significantly less able to hold their breath at inspiration than controls (P<0.05). In patients CI was 4.53+/-4.06 and 3.46+/-2.21 at inspiration and 4.45+/-4.23 and 4.76+/-3.73 at expiration. Intraobserver variability showed no significant differences either in patients or in healthy subjects. Reproducibility was significantly lower at a self-chosen breath-hold level of the healthy volunteers. DMRI is able to differentiate stability and reproducibility of different breath-hold levels. Expiratory breath-hold proved to be more stable than inspiratory breath-hold in healthy volunteers and patients.

The medical imaging interaction toolkit.

Thoroughly designed, open-source toolkits emerge to boost progress in medical imaging. The Insight Toolkit (ITK) provides this for the algorithmic scope of medical imaging, especially for segmentation and registration. But medical imaging algorithms have to be clinically applied to be useful, which additionally requires visualization and interaction. The Visualization Toolkit (VTK) has powerful visualization capabilities, but only low-level support for interaction. In this paper, we present the Medical Imaging Interaction Toolkit (MITK). The goal of MITK is to significantly reduce the effort required to construct specifically tailored, interactive applications for medical image analysis. MITK allows an easy combination of algorithms developed by ITK with visualizations created by VTK and extends these two toolkits with those features, which are outside the scope of both. MITK adds support for complex interactions with multiple states as well as undo-capabilities, a very important prerequisite for convenient user interfaces. Furthermore, MITK facilitates the realization of multiple, different views of the same data (as a multiplanar reconstruction and a 3D rendering) and supports the visualization of 3D+t data, whereas VTK is only designed to create one kind of view of 2D or 3D data. MITK reuses virtually everything from ITK and VTK. Thus, it is not at all a competitor to ITK or VTK, but an extension, which eases the combination of both and adds the features required for interactive, convenient to use medical imaging software. MITK is an open-source project (www.mitk.org).

Influence of different breathing maneuvers on internal and external organ motion: use of fiducial markers in dynamic MRI.

PURPOSE: To investigate, with dynamic magnetic resonance imaging (dMRI) and a fiducial marker, the influence of different breathing maneuvers on internal organ and external chest wall motion. METHODS AND MATERIALS: Lung and chest wall motion of 16 healthy subjects (13 male, 3 female) were examined with real-time trueFISP (true fast imaging with steady-state precession) dMRI and a small inductively coupled marker coil on either the abdomen or thorax. Three different breathing maneuvers were performed (predominantly "abdominal breathing," "thoracic breathing," and unspecific "normal breathing"). The craniocaudal (CC), anteroposterior (AP), and mediolateral (ML) lung distances were correlated (linear regression coefficient) with marker coil position during forced and quiet breathing. RESULTS: Differences of the CC distance between maximum forced inspiration and expiration were significant between abdominal and thoracic breathing (p < 0.05). The correlation between CC distance and coil position was best for forced abdominal breathing and a marker coil in the abdominal position (r = 0.89 +/- 0.04); for AP and ML distance, forced thoracic breathing and a coil in the thoracic position was best (r = 0.84 +/- 0.03 and 0.82 +/- 0.03, respectively). In quiet breathing, a lower correlation was found. CONCLUSION: A fiducial marker coil external to the thorax in combination with dMRI is a new technique to yield quantitative information on the correlation of internal organ and external chest wall motion. Correlations are highly dependent on the breathing maneuver.

The segments of the hepatic veins-is there a spatial correlation to the Couinaud liver segments?

PURPOSE: To investigate and describe the volume, position and shape of venous segments within the human liver and define their spatial correlation to the Couinaud segments (CS) and to the portal vein segments (PVS). MATERIAL AND METHODS: This study was based on 64 routinely acquired CT scans of patients undergoing hepatic surgery. The final analysis included 19 patients. All 19 CT data sets were transformed into 3D liver models. Three venous segments were postulated reflecting the left, middle, and right hepatic vein. Each venous segment was furthermore divided in two venous subsegments. Volume, position and shape of these venous segments/subsegments were calculated and, finally, compared with the volume, position and shape of the Couinaud segments and the portal vein segments. RESULTS: The right hepatic vein covers with 539.8 +/- 119.5 ml (47.1%) the largest part of total liver volume followed by the middle hepatic vein 372.7 +/- 151.1 ml (32.5%) and the left hepatic vein 248 +/- 75.9 ml (20.4%). The Couinaud liver segments and portal vein segments 2, 3, 5, 7, and 8 have consistent positional assignments within the three venous segments. Only the CS 4a, 4b, and 6 showed significantly different positions compared to the PVS 4a, 4b, and 6 (P < 0.03). The venous subsegments have a broad volumetric distribution reaching from 79 to 337 ml. There is no positional correlation of venous subsegments compared to Couinaud segments or portal vein segments at all (kappa < 0.75). In contrast, the venous segments/subsegments which can be assigned to either liver halve and either liver lobe have an identical volume, shape and position compared to the corresponding Couinaud liver segments (kappa > 0.75). CONCLUSION: The venous segments distinguish liver areas divided by the left and middle hepatic vein in exactly the same pattern as Couinaud segments and portal vein segments do. However, the comparison of shape and position of venous subsegments showed no correlation with both liver segmental approaches.