Michael Perkuhn

Automatic, three-segment, MR-based attenuation correction for whole-body PET/MR data

PurposeThe combination of positron emission tomography (PET) and magnetic resonance (MR) tomography in a single device is anticipated to be the next step following PET/CT for future molecular imaging application. Compared to CT, the main advantages of MR are versatile soft tissue contrast and its capability to acquire functional information without ionizing radiation. However, MR is not capable of measuring a physical quantity that would allow a direct derivation of the attenuation values for high-energy photons.MethodsTo overcome this problem, we propose a fully automated approach that uses a dedicated T1-weighted MR sequence in combination with a customized image processing technique to derive attenuation maps for whole-body PET. The algorithm automatically identifies the outer contour of the body and the lungs using region-growing techniques in combination with an intensity analysis for automatic threshold estimation. No user interaction is required to generate the attenuation map.ResultsThe accuracy of the proposed MR-based attenuation correction (AC) approach was evaluated in a clinical study using whole-body PET/CT and MR images of the same patients (n = 15). The segmentation of the body and lung contour (L-R directions) was evaluated via a four-point scale in comparison to the original MR image (mean values >3.8). PET images were reconstructed using elastically registered MR-based and CT-based (segmented and non-segmented) attenuation maps. The MR-based AC showed similar behaviour as CT-based AC and similar accuracy as offered by segmented CT-based AC. Standardized uptake value (SUV) comparisons with reference to CT-based AC using predefined attenuation coefficients showed the largest difference for bone lesions (mean value ± standard variation of SUVmax: −3.0% ± 3.9% for MR; −6.5% ± 4.1% for segmented CT). A blind comparison of PET images corrected with segmented MR-based, CT-based and segmented CT-based AC afforded identical lesion detectability, but slight differences in image quality were found.ConclusionOur MR‐based attenuation correction method offers similar correction accuracy as offered by segmented CT. According to the specialists involved in the blind study, these differences do not affect the diagnostic value of the PET images.

Wearable approach for continuous ECG - and activity patient-monitoring

The paper describes an approach to monitor a persons ECG and activity continuously with functional clothing. A belt with integrated electronics has been developed and has proven long-term robustness of all electrical components. A low-power module measures the ECG signal as well as the acceleration (2-axis) and stores data continuously up to two days. A user test has been performed to evaluate the belt according to system performance at different daily-life activities like sleeping, walking and so on. System parameters are ECG-signal quality, system up-time, and ECG-signal coverage during a day.

MR compatibility aspects of a silicon photomultiplier-based PET/RF insert with integrated digitisation

The combination of Positron Emission Tomography (PET) and Magnetic Resonance Imaging (MRI) into a single device is being considered a promising tool for molecular imaging as it combines the high sensitivity of PET with the functional and anatomical images of MRI. For highest performance, a scalable, MR compatible detector architecture with a small form factor is needed, targeting at excellent PET signal-to-noise ratios and time-of-flight information. Therefore it is desirable to use silicon photo multipliers and to digitize their signals directly in the detector modules inside the MRI bore. A preclinical PET/RF insert for clinical MRI scanner was built to demonstrate a new architecture and to study the interactions between the two modalities.The disturbance of the MRIs static magnetic field stays below 2 ppm peak-to-peak within a diameter of 56 mm (90 mm using standard automatic volume shimming). MRI SNR is decreased by 14%, RF artefacts (dotted lines) are only visible in sequences with very low SNR. Ghosting artefacts are visible to the eye in about 26% of the EPI images, severe ghosting only in 7.6%. Eddy-current related heating effects during long EPI sequences are noticeable but with low influence of 2% on the coincidences count rate. The time resolution of 2.5 ns, the energy resolution of 29.7% and the volumetric spatial resolution of 1.8 mm(3) in the PET isocentre stay unaffected during MRI operation. Phantom studies show no signs of other artefacts or distortion in both modalities. A living rat was simultaneously imaged after the injection with (18)F-Fluorodeoxyglucose (FDG) proving the in vivo capabilities of the system.

Diffusion-weighted MRI does not reflect kidney fibrosis in a rat model of fibrosis.

To assess the apparent diffusion coefficient (ADC) derived from diffusion‐weighted (DW) magnetic resonance imaging (MRI) as a specific marker of renal fibrosis in rats with unilateral ureteral obstruction (UUO).

Intraindividual quantitative and qualitative comparison of gadopentetate dimeglumine and gadobutrol in time-resolved contrast-enhanced 4-dimensional magnetic resonance angiography in minipigs.

ObjectivesThe concentration and relaxivities of contrast agents affect quantitative and qualitative image quality in contrast-enhanced time-resolved 4-dimensional magnetic resonance angiography (4D-MRA). Gadobutrol has a high relaxivity and is the only gadolinium (Gd)-based contrast agent approved for clinical use at a 1 M concentration. This promises to confer superior bolus characteristics by generating a steeper and shorter bolus with a higher peak Gd concentration. The purpose of this study was to quantitatively examine bolus characteristics of 1 M gadobutrol compared with 0.5 M gadopentetate dimeglumine and to evaluate image quality in thoracoabdominal 4D-MRA. Materials and MethodsA total of 7 Goettingen minipigs received dynamic computed tomography (CT) on a clinical 64-slice CT (transverse slices, 80 kV, 20 seconds, 0.3 s/dynamic frame) and 4D-MRA (time-resolved imaging with stochastic trajectories; 1. transverse slices, 30 seconds, 0.49 s/frame; 2. coronal slices, 70 seconds, 1.3 s/frame) on a 1.5-T clinical whole-body magnetic resonance imaging under general anesthesia using gadopentetate dimeglumine and gadobutrol in an intraindividual comparative study. Computed tomography attenuations were converted into Gd concentrations on the basis of previous phantom experiments. Quantitative analysis included measurements of the full width at half maximum, time-to-peak intervals, and peak of each bolus in dynamic CT and transverse 4D-MRA. These studies were carried out at equivalent contrast agent flow rates of 1 mL/s. Quantitative analysis (7 arteries and veins) and qualitative image analysis were performed on coronal thoracoabdominal 4D-MRA studies carried out at flow rates of 1 mL/s and, in the case of gadopentetate dimeglumine, also at molarity-adjusted flow rates of 2 mL/s. ResultsThe bolus in both transverse 4D-MRA and dynamic CT was significantly narrower (full width at half maximum), earlier (time to peak), and higher (signal intensity enhancement in 4D-MRA, Gd concentration in dynamic CT) when using gadobutrol instead of gadopentetate dimeglumine at a flow rate of 1.0 mL/s (P = 0.008−< 0.0001). In thoracoabdominal 4D-MRA, the signal intensity level and overall image quality were highest in examinations with gadobutrol, followed by examinations with gadopentetate dimeglumine at flow rates of 2 mL/s, and lowest in examinations with gadopentetate dimeglumine at flow rates of 1 mL/s. ConclusionsA more compact bolus shape was observed after administration of gadobutrol compared with gadopentetate dimeglumine in minipigs. This was demonstrated both in 4D-MRA, where Gd concentration, relaxivity, and the image-acquisition technique play a role, and in CT, where the signal intensity depends only on the Gd concentration. The overall image quality was rated higher in examinations with 1.0 M gadobutrol than with 0.5 M gadopentetate dimeglumine.

Combining Image and Non-image Clinical Data: An Infrastructure that Allows Machine Learning Studies in a Hospital Environment

Over the past years Machine Learning and Deep Learning techniques are showing their huge potential in medical research. However, this research is mainly done by using public or private datasets that were created for study purposes. Despite ensuring reproducibility, these datasets need to be constantly updated.

Fully automated detection and segmentation of meningiomas using deep learning on routine multiparametric MRI

ObjectivesMagnetic resonance imaging (MRI) is the method of choice for imaging meningiomas. Volumetric assessment of meningiomas is highly relevant for therapy planning and monitoring. We used a multiparametric deep-learning model (DLM) on routine MRI data including images from diverse referring institutions to investigate DLM performance in automated detection and segmentation of meningiomas in comparison to manual segmentations.MethodsWe included 56 of 136 consecutive preoperative MRI datasets [T1/T2-weighted, T1-weighted contrast-enhanced (T1CE), FLAIR] of meningiomas that were treated surgically at the University Hospital Cologne and graded histologically as tumour grade I (n = 38) or grade II (n = 18). The DLM was trained on an independent dataset of 249 glioma cases and segmented different tumour classes as defined in the brain tumour image segmentation benchmark (BRATS benchmark). The DLM was based on the DeepMedic architecture. Results were compared to manual segmentations by two radiologists in a consensus reading in FLAIR and T1CE.ResultsThe DLM detected meningiomas in 55 of 56 cases. Further, automated segmentations correlated strongly with manual segmentations: average Dice coefficients were 0.81 ± 0.10 (range, 0.46-0.93) for the total tumour volume (union of tumour volume in FLAIR and T1CE) and 0.78 ± 0.19 (range, 0.27-0.95) for contrast-enhancing tumour volume in T1CE.ConclusionsThe DLM yielded accurate automated detection and segmentation of meningioma tissue despite diverse scanner data and thereby may improve and facilitate therapy planning as well as monitoring of this highly frequent tumour entity.Key Points• Deep learning allows for accurate meningioma detection and segmentation• Deep learning helps clinicians to assess patients with meningiomas• Meningioma monitoring and treatment planning can be improved