Reproducible Radiomics Features from Multi-MRI-Scanner Test-Retest-Study: Influence on Performance and Generalizability of Models

BACKGROUND

Radiomics models trained on data from one center typically show a decline of performance when applied to data from external centers, hindering their introduction into large-scale clinical practice. Current expert recommendations suggest to use only reproducible radiomics features isolated by multiscanner test-retest experiments, which might help to overcome the problem of limited generalizability to external data.

PURPOSE

To evaluate the influence of using only a subset of robust radiomics features, defined in a prior in vivo multi-MRI-scanner test-retest-study, on the performance and generalizability of radiomics models.

STUDY TYPE

Retrospective.

POPULATION

Patients with monoclonal plasma cell disorders. Training set (117 MRIs from center 1); internal test set (42 MRIs from center 1); external test set (143 MRIs from center 2-8).

FIELD STRENGTH/SEQUENCE

1.5T and 3.0T; T1-weighted turbo spin echo.

ASSESSMENT

The task for the radiomics models was to predict plasma cell infiltration, determined by bone marrow biopsy, noninvasively from MRI. Radiomics machine learning models, including linear regressor, support vector regressor (SVR), and random forest regressor (RFR), were trained on data from center 1, using either all radiomics features, or using only reproducible radiomics features. Models were tested on an internal (center 1) and a multicentric external data set (center 2-8).

STATISTICAL TESTS

Pearson correlation coefficient r and mean absolute error (MAE) between predicted and actual plasma cell infiltration. Fisher's z-transformation, Wilcoxon signed-rank test, Wilcoxon rank-sum test; significance level P < 0.05.

RESULTS

When using only reproducible features compared with all features, the performance of the SVR on the external test set significantly improved (r = 0.43 vs. r = 0.18 and MAE = 22.6 vs. MAE = 28.2). For the RFR, the performance on the external test set deteriorated when using only reproducible instead of all radiomics features (r = 0.33 vs. r = 0.44, P = 0.29 and MAE = 21.9 vs. MAE = 20.5, P = 0.10).

CONCLUSION

Using only reproducible radiomics features improves the external performance of some, but not all machine learning models, and did not automatically lead to an improvement of the external performance of the overall best radiomics model.

LEVEL OF EVIDENCE: 3

TECHNICAL EFFICACY

Stage 2.

Computed tomography hypoperfusion-hypodensity mismatch vs. automated perfusion mismatch to identify stroke patients eligible for thrombolysis

Background and purpose

Automated perfusion imaging can detect stroke patients with unknown time of symptom onset who are eligible for thrombolysis. However, the availability of this technique is limited. We, therefore, established the novel concept of computed tomography (CT) hypoperfusion-hypodensity mismatch, i.e., an ischemic core lesion visible on cerebral perfusion CT without visible hypodensity in the corresponding native cerebral CT. We compared both methods regarding their accuracy in identifying patients suitable for thrombolysis.

Methods

In a retrospective analysis of the MissPerfeCT observational cohort study, patients were classified as suitable or not for thrombolysis based on established time window and imaging criteria. We calculated predictive values for hypoperfusion-hypodensity mismatch and automated perfusion imaging to compare accuracy in the identification of patients suitable for thrombolysis.

Results

Of 247 patients, 219 (88.7%) were eligible for thrombolysis and 28 (11.3%) were not eligible for thrombolysis. Of 197 patients who were within 4.5 h of symptom onset, 190 (96.4%) were identified by hypoperfusion-hypodensity mismatch and 88 (44.7%) by automated perfusion mismatch (p < 0.001). Of 22 patients who were beyond 4.5 h of symptom onset but were eligible for thrombolysis, 5 patients (22.7%) were identified by hypoperfusion-hypodensity mismatch. Predictive values for the hypoperfusion-hypodensity mismatch vs. automated perfusion mismatch were as follows: sensitivity, 89.0% vs. 50.2%; specificity, 71.4% vs. 100.0%; positive predictive value, 96.1% vs. 100.0%; and negative predictive value, 45.5% vs. 20.4%.

Conclusion

The novel method of hypoperfusion-hypodensity mismatch can identify patients suitable for thrombolysis with higher sensitivity and lower specificity than established techniques. Using this simple method might therefore increase the proportion of patients treated with thrombolysis without the use of special automated software.The MissPerfeCT study is a retrospective observational multicenter cohort study and is registered with clinicaltrials.gov (NCT04277728).

Tumor classification of gastrointestinal liver metastases using CT-based radiomics and deep learning

OBJECTIVES

The goal of this study is to demonstrate the performance of radiomics and CNN-based classifiers in determining the primary origin of gastrointestinal liver metastases for visually indistinguishable lesions.

METHODS

In this retrospective, IRB-approved study, 31 pancreatic cancer patients with 861 lesions (median age [IQR]: 65.39 [56.87, 75.08], 48.4% male) and 47 colorectal cancer patients with 435 lesions (median age [IQR]: 65.79 [56.99, 74.62], 63.8% male) were enrolled. A pretrained nnU-Net performed automated segmentation of 1296 liver lesions. Radiomics features for each lesion were extracted using pyradiomics. The performance of several radiomics-based machine-learning classifiers was investigated for the lesions and compared to an image-based deep-learning approach using a DenseNet-121. The performance was evaluated by AUC/ROC analysis.

RESULTS

The radiomics-based K-nearest neighbor classifier showed the best performance on an independent test set with AUC values of 0.87 and an accuracy of 0.67. In comparison, the image-based DenseNet-121-classifier reached an AUC of 0.80 and an accuracy of 0.83.

CONCLUSIONS

CT-based radiomics and deep learning can distinguish the etiology of liver metastases from gastrointestinal primary tumors. Compared to deep learning, radiomics based models showed a varying generalizability in distinguishing liver metastases from colorectal cancer and pancreatic adenocarcinoma.

Prediction of Bone Marrow Biopsy Results From MRI in Multiple Myeloma Patients Using Deep Learning and Radiomics

OBJECTIVES

In multiple myeloma and its precursor stages, plasma cell infiltration (PCI) and cytogenetic aberrations are important for staging, risk stratification, and response assessment. However, invasive bone marrow (BM) biopsies cannot be performed frequently and multifocally to assess the spatially heterogenous tumor tissue. Therefore, the goal of this study was to establish an automated framework to predict local BM biopsy results from magnetic resonance imaging (MRI).

MATERIALS AND METHODS

This retrospective multicentric study used data from center 1 for algorithm training and internal testing, and data from center 2 to 8 for external testing. An nnU-Net was trained for automated segmentation of pelvic BM from T1-weighted whole-body MRI. Radiomics features were extracted from these segmentations, and random forest models were trained to predict PCI and the presence or absence of cytogenetic aberrations. Pearson correlation coefficient and the area under the receiver operating characteristic were used to evaluate the prediction performance for PCI and cytogenetic aberrations, respectively.

RESULTS

A total of 672 MRIs from 512 patients (median age, 61 years; interquartile range, 53-67 years; 307 men) from 8 centers and 370 corresponding BM biopsies were included. The predicted PCI from the best model was significantly correlated ( P ≤ 0.01) to the actual PCI from biopsy in all internal and external test sets (internal test set: r = 0.71 [0.51, 0.83]; center 2, high-quality test set: r = 0.45 [0.12, 0.69]; center 2, other test set: r = 0.30 [0.07, 0.49]; multicenter test set: r = 0.57 [0.30, 0.76]). The areas under the receiver operating characteristic of the prediction models for the different cytogenetic aberrations ranged from 0.57 to 0.76 for the internal test set, but no model generalized well to all 3 external test sets.

CONCLUSIONS

The automated image analysis framework established in this study allows for noninvasive prediction of a surrogate parameter for PCI, which is significantly correlated to the actual PCI from BM biopsy.

Periaortic adipose radiomics texture features associated with increased coronary calcium score-first results on a photon-counting-CT

BACKGROUND

Cardiovascular diseases remain the world's primary cause of death. The identification and treatment of patients at risk of cardiovascular events thus are as important as ever. Adipose tissue is a classic risk factor for cardiovascular diseases, has been linked to systemic inflammation, and is suspected to contribute to vascular calcification. To further investigate this issue, the use of texture analysis of adipose tissue using radiomics features could prove a feasible option.

METHODS

In this retrospective single-center study, 55 patients (mean age 56, 34 male, 21 female) were scanned on a first-generation photon-counting CT. On axial unenhanced images, periaortic adipose tissue surrounding the thoracic descending aorta was segmented manually. For feature extraction, patients were divided into three groups, depending on coronary artery calcification (Agatston Score 0, Agatston Score 1-99, Agatston Score ≥ 100). 106 features were extracted using pyradiomics. R statistics was used for statistical analysis, calculating mean and standard deviation with Pearson correlation coefficient for feature correlation. Random Forest classification was carried out for feature selection and Boxplots and heatmaps were used for visualization. Additionally, monovariable logistic regression predicting an Agatston Score > 0 was performed, selected features were tested for multicollinearity and a 10-fold cross-validation investigated the stability of the leading feature.

RESULTS

Two higher-order radiomics features, namely "glcm_ClusterProminence" and "glcm_ClusterTendency" were found to differ between patients without coronary artery calcification and those with coronary artery calcification (Agatston Score ≥ 100) through Random Forest classification. As the leading differentiating feature "glcm_ClusterProminence" was identified.

CONCLUSION

Changes in periaortic adipose tissue texture seem to correlate with coronary artery calcium score, supporting a possible influence of inflammatory or fibrotic activity in perivascular adipose tissue. Radiomics features may potentially aid as corresponding biomarkers in the future.

Phantom-based radiomics feature test-retest stability analysis on photon-counting detector CT

OBJECTIVES

Radiomics image data analysis offers promising approaches in research but has not been implemented in clinical practice yet, partly due to the instability of many parameters. The aim of this study is to evaluate the stability of radiomics analysis on phantom scans with photon-counting detector CT (PCCT).

METHODS

Photon-counting CT scans of organic phantoms consisting of 4 apples, kiwis, limes, and onions each were performed at 10 mAs, 50 mAs, and 100 mAs with 120-kV tube current. The phantoms were segmented semi-automatically and original radiomics parameters were extracted. This was followed by statistical analysis including concordance correlation coefficients (CCC), intraclass correlation coefficients (ICC), as well as random forest (RF) analysis, and cluster analysis to determine the stable and important parameters.

RESULTS

Seventy-three of the 104 (70%) extracted features showed excellent stability with a CCC value > 0.9 when compared in a test and retest analysis, and 68 features (65.4%) were stable compared to the original in a rescan after repositioning. Between the test scans with different mAs values, 78 (75%) features were rated with excellent stability. Eight radiomics features were identified that had an ICC value greater than 0.75 in at least 3 of 4 groups when comparing the different phantoms in a phantom group. In addition, the RF analysis identified many features that are important for distinguishing the phantom groups.

CONCLUSION

Radiomics analysis using PCCT data provides high feature stability on organic phantoms, which may facilitate the implementation of radiomics analysis likewise in clinical routine.

KEY POINTS

• Radiomics analysis using photon-counting computed tomography provides high feature stability. • Photon-counting computed tomography may pave the way for implementation of radiomics analysis in clinical routine.

In Vivo Repeatability and Multiscanner Reproducibility of MRI Radiomics Features in Patients With Monoclonal Plasma Cell Disorders: A Prospective Bi-institutional Study

OBJECTIVES

Despite the extensive number of publications in the field of radiomics, radiomics algorithms barely enter large-scale clinical application. Supposedly, the low external generalizability of radiomics models is one of the main reasons, which hinders the translation from research to clinical application. The objectives of this study were to investigate reproducibility of radiomics features (RFs) in vivo under variation of patient positioning, magnetic resonance imaging (MRI) sequence, and MRI scanners, and to identify a subgroup of RFs that shows acceptable reproducibility across all different acquisition scenarios.

MATERIALS AND METHODS

Between November 30, 2020 and February 16, 2021, 55 patients with monoclonal plasma cell disorders were included in this prospective, bi-institutional, single-vendor study. Participants underwent one reference scan at a 1.5 T MRI scanner and several retest scans: once after simple repositioning, once with a second MRI protocol, once at another 1.5 T scanner, and once at a 3 T scanner. Radiomics feature from the bone marrow of the left hip bone were extracted, both from original scans and after different image normalizations. Intraclass correlation coefficient (ICC) was used to assess RF repeatability and reproducibility.

RESULTS

Fifty-five participants (mean age, 59 ± 7 years; 36 men) were enrolled. For T1-weighted images after muscle normalization, in the simple test-retest experiment, 110 (37%) of 295 RFs showed an ICC ≥0.8: 54 (61%) of 89 first-order features (FOFs), 35 (95%) of 37 volume and shape features, and 21 (12%) of 169 texture features (TFs). When the retest was performed with different technical settings, even after muscle normalization, the number of FOF/TF with an ICC ≥0.8 declined to 58/13 for the second protocol, 29/7 for the second 1.5 T scanner, and 49/7 for the 3 T scanner, respectively. Twenty-five (28%) of the 89 FOFs and 6 (4%) of the 169 TFs from muscle-normalized T1-weighted images showed an ICC ≥0.8 throughout all repeatability and reproducibility experiments.

CONCLUSIONS

In vivo, only few RFs are reproducible with different MRI sequences or different MRI scanners, even after application of a simple image normalization. Radiomics features selected by a repeatability experiment only are not necessarily suited to build radiomics models for multicenter clinical application. This study isolated a subset of RFs, which are robust to variations in MRI acquisition observed in scanners from 1 vendor, and therefore are candidates to build reproducible radiomics models for monoclonal plasma cell disorders for multicentric applications, at least when centers are equipped with scanners from this vendor.

Metal artifacts and artifact reduction of neurovascular coils in photon-counting detector CT versus energy-integrating detector CT - in vitro comparison of a standard brain imaging protocol

OBJECTIVES

Photon-counting detector computed tomography (PCD-CT) is a promising new technique for CT imaging. The aim of the present study was the in vitro comparison of coil-related artifacts in PCD-CT and conventional energy-integrating detector CT (EID-CT) using a comparable standard brain imaging protocol before and after metal artifact reduction (MAR).

METHODS

A nidus-shaped rubber latex, resembling an aneurysm of the cerebral arteries, was filled with neurovascular platinum coils and inserted into a brain imaging phantom. Image acquisition and reconstruction were repeatedly performed for PCD-CT and EID-CT (n = 10, respectively) using a standard brain imaging protocol. Moreover, linear interpolation MAR was performed for PCD-CT and EID-CT images. The degree of artifacts was analyzed quantitatively (standard deviation in a donut-shaped region of interest) and qualitatively (5-point scale analysis).

RESULTS

Quantitative and qualitative analysis demonstrated a lower degree of metal artifacts in the EID-CT images compared to the total-energy PCD-CT images (e.g., 82.99 ± 7.89 Hounsfield units (HU) versus 90.35 ± 6.28 HU; p < 0.001) with no qualitative difference between the high-energy bin PCD-CT images and the EID-CT images (4.18 ± 0.37 and 3.70 ± 0.64; p = 0.575). After MAR, artifacts were more profoundly reduced in the PCD-CT images compared to the EID-CT images in both analyses (e.g., 2.35 ± 0.43 and 3.18 ± 0.34; p < 0.001).

CONCLUSION

PCD-CT in combination with MAR have the potential to provide an improved option for reduction of coil-related artifacts in cerebral imaging in this in vitro study.

KEY POINTS

• Photon-counting detector CT produces more artifacts compared to energy-integrating detector CT without metal artifact reduction in cerebral in vitro imaging after neurovascular coil-embolization. • Spectral information of PCD-CT provides the potential for new post-processing techniques, since the coil-related artifacts were lower in PCD-CT images compared to EID-CT images after linear interpolation metal artifact reduction in this in vitro study.

Simplified Assessment of Lesion Water Uptake for Identification of Patients within 4.5 Hours of Stroke Onset: An Analysis of the MissPerfeCT Study

Background and Purpose Many patients with stroke cannot receive intravenous thrombolysis because the time of symptom onset is unknown. We tested whether a simple method of computed tomography (CT)-based quantification of water uptake in the ischemic tissue can identify patients with stroke onset within 4.5 hours. Methods This retrospective analysis of the MissPerfeCT study (August 2009 to November 2017) includes consecutive patients with known onset of symptoms from seven tertiary stroke centers. We developed a simplified algorithm based on region of interest (ROI) measurements to quantify water uptake of the ischemic lesion and thereby quantify time of symptom onset within and beyond 4.5 hours. Perfusion CT was used to identify ischemic brain tissue, and its density was measured in non-contrast CT and related to the density of the corresponding area of the contralateral hemisphere to quantify lesion water uptake. Results Of 263 patients, 204 (77.6%) had CT within 4.5 hours. Water uptake was significantly lower in patients with stroke onset within (6.7%; 95% confidence interval [CI], 6.0% to 7.4%) compared to beyond 4.5 hours (12.7%; 95% CI, 10.7% to 14.7%). The area under the curve for distinguishing these patient groups according to percentage water uptake was 0.744 with an optimal cut-off value of 9.5%. According to this cut-off the positive predictive value was 88.8%, sensitivity was 73.5%, specificity 67.8%, negative predictive value was 42.6%. Conclusions Ischemic stroke patients with unknown time of symptom onset can be identified as being within a timeframe of 4.5 hours using a ROI-based method to assess water uptake on admission non-contrast head CT.

Hybrid imaging with [68Ga]PSMA-11 PET-CT and PET-MRI in biochemically recurrent prostate cancer

AIM

To compare [⁶⁸Ga]PSMA-11 PET-CT, [⁶⁸Ga]PSMA-11 PET-MRI and MRI in a cohort of prostate cancer (PCa) patients in biochemical recurrence after initial curative therapy.

MATERIALS AND METHODS

Fifty-three patients with biochemically recurrent PCa underwent whole-body [⁶⁸Ga]PSMA-11 PET-CT 1 hour post-injection (p.i.) followed by [⁶⁸Ga]PSMA-11 PET-MRI 2.5 hours p.i., including a multiparametric MRI pelvic protocol examination. Imaging data analysis consisted of visual (qualitative) evaluation of the PET-CT, PET-MRI and MRI scans, as well as semi-quantitative and quantitative analyses of the PET and MRI data, including calculation of the parameters standardized uptake value (SUV) and apparent diffusion coefficient (ADC) derived from the PCa lesions. Association analysis was performed between imaging and clinical data, including PSA level and Gleason score. The results were considered significant for p-values less than 0.05 (p < 0.05).

RESULTS

The hybrid imaging modalities [⁶⁸Ga]PSMA-11 PET-CT and PET-MRI were positive in more patients than MRI alone. In particular, PET-CT detected lesions suggestive of PCa relapse in 34/53 (64.2%), PET-MRI in 36/53 (67.9%) and MRI in 23/53 patients (43.4%). While no significant differences in lesion detection rate were observed between PET-CT and PET-MRI, the latter was particularly efficient in detection of local recurrences in the prostate bed mainly due to the contribution of the MRI part of the modality. Association analysis revealed a statistically significant increase in the probability of a positive scan with increasing PSA levels for all imaging modalities. Accordingly, there was no significant association between scan positivity rate and Gleason score for any imaging modality. No significant correlation was observed between SUV and ADC values in lymph node metastases.

CONCLUSION

[⁶⁸Ga]PSMA-11 PET-CT and PET-MRI provide equally good detection rates for PCa recurrence, both outperforming stand-alone MRI.

Influence of diffusion on transverse relaxation rates and phases of an ensemble of magnetic spheres

In quantitative susceptibility mapping, the tissue susceptibility is determined from the magnitude and phase of the gradient echo signal, which is influenced by the interplay of complex susceptibility and diffusion effect. Herein, we analytically analyze the influence of diffusion on magnitude and phase images generated by randomly arranged magnetic spheres as a model of intracerebral iron depositions. We demonstrate that both gradient and spin echo relaxation rate constants have a strong and nonlinear dependence on diffusion strength and give empirical formulas for magnitude and phase. This may be used in the future to improve QSM processing methods. In addition, we show that, in theory, combined acquisitions of gradient and spin echo can be used to determine the dimension of the magnetic spheres and the diffusion strength.

Ultrahigh resolution whole body photon counting computed tomography as a novel versatile tool for translational research from mouse to man

X-ray computed tomography (CT) is a cardinal tool in clinical practice. It provides cross-sectional images within seconds. The recent introduction of clinical photon-counting CT allowed for an increase in spatial resolution by more than a factor of two resulting in a pixel size in the center of rotation of about 150 µm. This level of spatial resolution is in the order of dedicated preclinical micro-CT systems. However so far, the need for different dedicated clinical and preclinical systems often hinders the rapid translation of early research results to applications in men. This drawback might be overcome by ultra-high resolution (UHR) clinical photon-counting CT unifying preclinical and clinical research capabilities in a single machine. Herein, the prototype of a clinical UHR PCD CT (SOMATOM CounT, Siemens Healthineers, Forchheim, Germany) was used. The system comprises a conventional energy-integrating detector (EID) and a novel photon-counting detector (PCD). While the EID provides a pixel size of 0.6 mm in the centre of rotation, the PCD provides a pixel size of 0.25 mm. Additionally, it provides a quantification of photon energies by sorting them into up to four distinct energy bins. This acquisition of multi-energy data allows for a multitude of applications, e.g. pseudo-monochromatic imaging. In particular, we examine the relation between spatial resolution, image noise and administered radiation dose for a multitude of use-cases. These cases include ultra-high resolution and multi-energy acquisitions of mice administered with a prototype bismuth-based contrast agent (nanoPET Pharma, Berlin, Germany) as well as larger animals and actual patients. The clinical EID provides a spatial resolution of about 9 lp/cm (modulation transfer function at 10%, MTF_10%) while UHR allows for the acquisition of images with up to 16 lp/cm allowing for the visualization of all relevant anatomical structures in preclinical and clinical specimen. The spectral capabilities of the system enable a variety of applications previously not available in preclinical research such as pseudo-monochromatic images. Clinical ultra-high resolution photon-counting CT has the potential to unify preclinical and clinical research on a single system enabling versatile imaging of specimens and individuals ranging from mice to man.

Myocardial Radiomics Texture Features Associated with Increased Coronary Calcium Score—First Results of a Photon-Counting CT

The coronary artery calcium score is an independent risk factor of the development of adverse cardiac events. The severity of coronary artery calcification may influence the myocardial texture. Due to higher spatial resolution and signal-to-noise ratio, new CT technologies such as PCCT may improve the detection of texture alterations depending on the severity of coronary artery calcification. In this retrospective, single-center, IRB-approved study, left ventricular myocardium was segmented and radiomics features were extracted using pyradiomics. The mean and standard deviation with the Pearson correlation coefficient for correlations of features were calculated and visualized as boxplots and heatmaps. Random forest feature selection was performed. Thirty patients (26.7% women, median age 58 years) were enrolled in the study. Patients were divided into two subgroups depending on the severity of coronary artery calcification (Agatston score 0 and Agatston score ≥ 100). Through random forest feature selection, a set of four higher-order features could be defined to discriminate myocardial texture between the two groups. When including the additional Agatston 1–99 groups as a validation, a severity-associated change in feature intensity was detected. A subset of radiomics features texture alterations of the left ventricular myocardium was associated with the severity of coronary artery calcification estimated by the Agatston score.

Influence of flow and susceptibility effects on spin dephasing in lung tissue

PURPOSE

Magnetic resonance imaging (MRI) of the lung can be used for diagnosis and monitoring of interstitial lung disease. Biophysical models of alveolar lung tissue are needed to understand the complex interplay of susceptibility, diffusion, and flow effects, and their influence on magnetic resonance (MR) spin dephasing.

METHODS

In this work, we present a method for modeling the signal decay of lung tissue by utilizing a two-compartment model, which considers the different spin dephasing mechanisms in the alveolar vasculature and interstitial tissue. This allows calculating the magnetization dynamics and the MR lineshape, which can be measured noninvasively using clinical MR scanners.

RESULTS

The accuracy of the method was evaluated using finite element simulations and the experimentally measured lineshapes of a healthy volunteer. In this comparison, the model performs well, indicating that the relevant spin dephasing mechanisms are correctly taken into account.

CONCLUSIONS

The proposed method can be used to estimate the influence of blood flow and alveolar geometry on the MR lineshape of lung tissue.

Comparison Study of Myocardial Radiomics Feature Properties on Energy-Integrating and Photon-Counting Detector CT

The implementation of radiomics-based, quantitative imaging parameters is hampered by a lack of stability and standardization. Photon-counting computed tomography (PCCT), compared to energy-integrating computed tomography (EICT), does rely on a novel detector technology, promising better spatial resolution and contrast-to-noise ratio. However, its effect on radiomics feature properties is unknown. This work investigates this topic in myocardial imaging. In this retrospective, single-center IRB-approved study, the left ventricular myocardium was segmented on CT, and the radiomics features were extracted using pyradiomics. To compare features between scanners, a t-test for non-paired samples and F-test was performed, with a threshold of 0.05 set as a benchmark for significance. Feature correlations were calculated by the Pearson correlation coefficient, and visualization was performed with heatmaps. A total of 50 patients (56% male, mean age 56) were enrolled in this study, with equal proportions of PCCT and EICT. First-order features were, nearly, comparable between both groups. However, higher-order features showed a partially significant difference between PCCT and EICT. While first-order radiomics features of left ventricular myocardium show comparability between PCCT and EICT, detected differences of higher-order features may indicate a possible impact of improved spatial resolution, better detection of lower-energy photons, and a better signal-to-noise ratio on texture analysis on PCCT.

Quantitative Analysis of DCE and DSC-MRI: From Kinetic Modeling to Deep Learning

BACKGROUND

 Perfusion MRI is a well-established imaging modality with a multitude of applications in oncological and cardiovascular imaging. Clinically used processing methods, while stable and robust, have remained largely unchanged in recent years. Despite promising results from novel methods, their relatively minimal improvement compared to established methods did not generally warrant significant changes to clinical perfusion processing.

RESULTS AND CONCLUSION

 Machine learning in general and deep learning in particular, which are currently revolutionizing computer-aided diagnosis, may carry the potential to change this situation and truly capture the potential of perfusion imaging. Recent advances in the training of recurrent neural networks make it possible to predict and classify time series data with high accuracy. Combining physics-based tissue models and deep learning, using either physics-informed neural networks or universal differential equations, simplifies the training process and increases the interpretability of the resulting models. Due to their versatility, these methods will potentially be useful in bridging the gap between microvascular architecture and perfusion parameters, akin to MR fingerprinting in structural MR imaging. Still, further research is urgently needed before these methods may be used in clinical practice.

KEY POINTS

· Machine learning offers promising methods for processing of perfusion data.. · Recurrent neural networks can classify time series with high accuracy.. · Data augmentation is essentially especially when using small datasets..

CITATION FORMAT

· Rotkopf LT, Zhang KS, Tavakoli AA et al. Quantitative Analysis of DCE and DSC-MRI: From Kinetic Modeling to Deep Learning. Fortschr Röntgenstr 2022; DOI: 10.1055/a-1762-5854.

Influence of local aortic calcification on periaortic adipose tissue radiomics texture features-a primary analysis on PCCT

Perivascular adipose tissue is known to be metabolically active. Volume and density of periaortic adipose tissue are associated with aortic calcification as well as aortic diameter indicating a possible influence of periaortic adipose tissue on the development of aortic calcification. Due to better spatial resolution and signal-to-noise ratio, new CT technologies such as photon-counting computed tomography may allow the detection of texture alterations of periaortic adipose tissue depending on the existence of local aortic calcification possibly outlining a biomarker for the development of arteriosclerosis. In this retrospective, single-center, IRB-approved study, periaortic adipose tissue was segmented semiautomatically and radiomics features were extracted using pyradiomics. Statistical analysis was performed in R statistics calculating mean and standard deviation with Pearson correlation coefficient for feature correlation. For feature selection Random Forest classification was performed. A two-tailed unpaired t test was applied to the final feature set. Results were visualized as boxplots and heatmaps. A total of 30 patients (66.6% female, median age 57 years) were enrolled in this study. Patients were divided into two subgroups depending on the presence of local aortic calcification. By Random Forest feature selection a set of seven higher-order features could be defined to discriminate periaortic adipose tissue texture between these two groups. The t test showed a statistic significant discrimination for all features (p < 0.05). Texture changes of periaortic adipose tissue associated with the existence of local aortic calcification may lay the foundation for finding a biomarker for development of arteriosclerosis.

Spin echo formation in muscle tissue

Determination of the spin echo signal evolution and of transverse relaxation rates is of high importance for microstructural modeling of muscle tissue in magnetic resonance imaging. So far, numerically exact solutions for the NMR signal dynamics in muscle tissue models have been reported only for the gradient echo free induction decay, with spin echo problems usually solved by approximate methods. In this work, we modeled the spin echo signal numerically exact by discretizing the radial dimension of the Bloch-Torrey equation and expanding the angular dependency in terms of even Chebyshev polynomials. This allows us to express the time dependence of the local magnetization as a closed-form matrix expression. Using this method, we were able to accurately capture the spin echo local and total magnetization dynamics. The obtained transverse relaxation rates showed a high concordance with random walker and finite-element simulations. We could demonstrate that in cases of smaller diffusion coefficients, the commonly used strong collision approximation significantly underestimates the true value considerably. Instead, the limiting behavior in this regime is correctly described either by the full solution or by the slow diffusion approximation. Experimentally measured transverse relaxation rates of a mouse limb muscle showed an angular dependence in accordance with the theoretical prediction.

Threshold-dependent iodine imaging and spectral separation in a whole-body photon-counting CT system

OBJECTIVE

To evaluate the dual-energy (DE) performance and spectral separation with respect to iodine imaging in a photon-counting CT (PCCT) and compare it to dual-source CT (DSCT) DE imaging.

METHODS

A semi-anthropomorphic phantom extendable with fat rings equipped with iodine vials is measured in an experimental PCCT. The system comprises a PC detector with two energy bins (20 keV, T) and (T, eU) with threshold T and tube voltage U. Measurements using the PCCT are performed at all available tube voltages (80 to 140 kV) and threshold settings (50-90 keV). Further measurements are performed using a conventional energy-integrating DSCT. Spectral separation is quantified as the relative contrast media ratio R between the energy bins and low/high images. Image noise and dose-normalized contrast-to-noise ratio (CNRD) are evaluated in resulting iodine images. All results are validated in a post-mortem angiography study.

RESULTS

R of the PC detector varies between 1.2 and 2.6 and increases with higher thresholds and higher tube voltage. Reference R of the EI DSCT is found as 2.20 on average overall phantoms. Maximum CNRD in iodine images is found for T = 60/65/70/70 keV for 80/100/120/140 kV. The highest CNRD of the PCCT is obtained using 140 kV and is decreasing with decreasing tube voltage. All results could be confirmed in the post-mortem angiography study.

CONCLUSION

Intrinsically acquired DE data are able to provide iodine images similar to conventional DSCT. However, PCCT thresholds should be chosen with respect to tube voltage to maximize image quality in retrospectively derived image sets.

KEY POINTS

• Photon-counting CT allows for the computation of iodine images with similar quality compared to conventional dual-source dual-energy CT. • Thresholds should be chosen as a function of the tube voltage to maximize iodine contrast-to-noise ratio in derived image sets. • Image quality of retrospectively computed image sets can be maximized using optimized threshold settings.

Dependence of the frequency distribution around a sphere on the voxel orientation

Microscopically small magnetic field inhomogeneities within an external static magnetic field cause a free induction decay in magnetic resonance imaging that generally exhibits two transverse components that are usually summarized to a complex entity. The Fourier transform of the complex-valued free induction decay is the purely real and positive-valued frequency distribution which allows an easy interpretation of the underlying dephasing mechanism. Typically, the frequency distribution inside a cubic voxel as caused by a spherical magnetic field inhomogeneity is determined by a histogram technique in terms of subdivision of the whole voxel into smaller subvoxels. A faster and more accurate computation is achieved by analytical expressions for the frequency distribution that are derived in this work. In contrast to the usually assumed simplified case of a spherical voxel, we also consider the tilt angles of the cubic voxel to the external magnetic field. The typical asymmetric form of the frequency distribution is reproduced and analyzed for the more realistic case of a cubic voxel. We observe a splitting of frequency distribution peaks for increasing tilt of the cubic voxel against the direction of the external magnetic field in analogy to the case for dephasing around cylindrical, vessel-like objects inside cubic voxels. These results are of value, e.g., for the analysis of susceptibility-weighted images or in quantitative susceptibility imaging since the reconstruction of these images is performed in cubic-shaped voxels.

Photon-counting detectors in computed tomography: from quantum physics to clinical practice

Over the last decade, a fundamentally new type of computed tomography (CT) detectors has proved its superior capabilities in both physical and preclinical evaluations and is now approaching the stage of clinical practice. These detectors are able to discriminate single photons and quantify their energy and are hence called photon-counting detectors. Among the promising benefits of this technology are improved radiation dose efficiency, increased contrast-to-noise ratio, reduced metal artifacts, improved spatial resolution, simultaneous multi-energy acquisitions, and the prospect of multi-phase imaging within a single acquisition using multiple contrast agents. Taking the conventional energy-integrating detectors as a reference, the authors demonstrate the technical principles of this new technology and provide phantom and patient images acquired by a whole-body photon-counting CT. These images serve as a basis for discussing the potential future of clinical CT.

Potential of ultra-high-resolution photon-counting CT of bone metastases: initial experiences in breast cancer patients

Conventional CT scanners use energy-integrating detectors (EIDs). Photon-counting detector (PCD) computed tomography (CT) utilizes a CT detector technology based on smaller detector pixels capable of counting single photons and in addition discriminating their energy. Goal of this study was to explore the potential of higher spatial resolution for imaging of bone metastases. Four female patients with histologically confirmed breast cancer and bone metastases were included between July and October 2019. All patients underwent conventional EID CT scans followed by a high resolution non-contrast experimental PCD CT scan. Ultra-high resolution (UHR) reconstruction kernels were used to reconstruct axial slices with voxel sizes of 0.3 mm × 0.3 mm (inplane) × 1 mm (z-direction). Four radiologists blinded for patient identity assessed the images and compared the quality to conventional CT using a qualitative Likert scale. In this case series, we present images of bone metastases in breast cancer patients using an experimental PCD CT scanner and ultra-high-resolution kernels. A tendency to both a smaller inter-reader variability in the structural assessment of lesion sizes and in the readers' opinion to an improved visualization of lesion margins and content was observed. In conclusion, while further studies are warranted, PCD CT has a high potential for therapy monitoring in breast cancer.

Ischemic Core Volume Combined with the Relative Perfusion Ratio for Stroke Outcome Prediction after Endovascular Thrombectomy

BACKGROUND AND PURPOSE

Imaging-based selection of stroke patients for endovascular thrombectomy (EVT) remains an ongoing challenge. Our aim was to determine the value of a combined parameter of ischemic core volume (ICV) and the relative degree of cerebral blood flow in the penumbra for morphologic and clinical outcome prediction.

METHODS

In this Institutional Review Board (IRB)-approved prospective observational study, 221 consecutive patients with large vessel occlusion anterior circulation stroke within 6 hours of symptom onset and subsequent EVT were included between June 2015 and August 2017. Admission computed tomography perfusion was analyzed using automated threshold-based algorithms. Perfusion-weighted ICV (pw-ICV) was calculated by multiplying ICV with the relative cerebral blood flow reduction within the penumbra. Functional outcome was assessed by standardized assessment of the modified Rankin scale (mRS) after 3 months.

RESULTS

In multivariate analyses, pw-ICV was significantly associated with final infarction volume (FIV) (β = .38, P < .001) after adjustment for penumbra volume, age, sex and time from symptom onset. In separate multivariate analysis with either pw-ICV or ICV, pw-ICV outperformed ICV for the prediction of FIV (Akaike's information criterion: 1,072 vs. 1,089; conditional variable importance: 1,494 vs. 955). There was also a highly significant association between FIV and clinical outcome as measured by an mRS score of 2 or less (odds ratio per 10 mL = .78, P < .001). Both pw-ICV and ICV were significantly associated with NIHSS improvement (both P<.05).

CONCLUSION

In EVT-treated stroke patients, pw-ICV outperforms the more commonly used ICV in the prediction of morphological and functional outcome.

The wavelet power spectrum of perfusion weighted MRI correlates with tumor vascularity in biopsy-proven glioblastoma samples

Background

Wavelet transformed reconstructions of dynamic susceptibility contrast (DSC) MR perfusion (wavelet-MRP) are a new and elegant way of visualizing vascularization. Wavelet-MRP maps yield a clear depiction of hypervascular tumor regions, as recently shown.

Objective

The aim of this study was to elucidate a possible connection of the wavelet-MRP power spectrum in glioblastoma (GBM) with local vascularity and cell proliferation.

Methods

For this IRB-approved study 12 patients (63.0+/-14.9y; 7m) with histologically confirmed IDH-wildtype GBM were included. Target regions for biopsies were prospectively marked on tumor regions as seen on preoperative 3T MRI. During subsequent neurosurgical tumor resection 43 targeted biopsies were taken from these target regions, of which all 27 matching samples were analyzed. All specimens were immunohistochemically analyzed for endothelial cell marker CD31 and proliferation marker Ki67 and correlated to the wavelet-MRP power spectrum as derived from DSC perfusion weighted imaging.

Results

There was a strong correlation between wavelet-MRP power spectrum (median = 4.41) and conventional relative cerebral blood volume (median = 5.97 ml/100g) in Spearman's rank-order correlation (κ = .83, p < .05). In a logistic regression model, the wavelet-MRP power spectrum showed a significant correlation to CD31 dichotomized to no or present staining (p = .04), while rCBV did not show a significant correlation to CD31 (p = .30). No significant association between Ki67 and rCBV or wavelet-MRP was found (p = .62 and p = .70, respectively).

Conclusion

The wavelet-MRP power spectrum derived from existing DSC-MRI data might be a promising new surrogate for tumor vascularity in GBM.

Drip and ship for mechanical thrombectomy within the Neurovascular Network of Southwest Bavaria

Objective

To determine feasibility and safety of stroke care organization within our Neurovascular Network of Southwest Bavaria (NEVAS) in a rural area with distances of up to 100 kilometers, we compared patients who underwent mechanical thrombectomy (MT) in large vessel occlusion admitted directly to our center (direct to center [DTC]) to patients who were transferred for MT via NEVAS (drip and ship [DS]).

Methods

This is a retrospective analysis of prospectively collected data of all MT patients between January 2015 and May 2018. Successful recanalization was defined as a thrombolysis in cerebral infarction score of 2b-3. Symptomatic intracerebral hemorrhage (sICH) was defined according to European Cooperative Acute Stroke Study 3. Modified Rankin Scale (mRS) score of 0–2 at 3 months indicated good outcome.

Results

MT was performed in 410 patients: 221 DTC and 189 DS. Median NIH Stroke Scale (NIHSS) score was 16 and premorbid mRS score was 0. Thrombolysis was applied in 62.2% with the same time from symptom onset in both groups (94.5 vs 95 minutes). Successful recanalization (79.3% vs 77.8%) and NIHSS score reduction from admission to discharge (16–7 vs 17–6) were comparable. Time delay from onset to revascularization was 96 minutes in DS (212 vs 308 minutes, p = 0.001). At follow-up, DTC patients had a trend to better outcome (33.5% vs 24.3%, p = 0.056). Neither sICH (6.3% vs 5.9%, p = 0.840) nor mortality (31.2% vs 34.4%, p = 0.387) differed between the groups.

Conclusion

DS patients benefit from MT without relevant safety concerns, but with a trend to unfavorable outcome compared to DTC patients. These results suggest that DS is suitable to provide MT in rural areas where DTC is not possible.

Incremental Value of Computed Tomography Perfusion for Final Infarct Prediction in Acute Ischemic Cerebellar Stroke

Background

The diagnosis of ischemic cerebellar stroke is challenging because of nonspecific symptoms and very limited accuracy of commonly applied computed tomography (CT) imaging. Advances in CT perfusion imaging provide increasing value in the detection of posterior circulation stroke, but the prognostic value remains unclear. We aimed to identify imaging parameters that predict morphologic outcome in cerebellar stroke patients using advanced CT including whole‐brain CT perfusion (WB‐CTP).

Methods and Results

We selected all subjects with cerebellar WB‐CTP perfusion deficits and follow‐up‐confirmed cerebellar infarction from a consecutive cohort with suspected stroke who underwent WB‐CTP. Posterior‐circulation‐Acute‐Stroke‐Prognosis‐Early‐CT‐Score (pc‐ASPECTS) was determined on noncontrast CT, CT angiography source images, and on parametric WB‐CTP maps. Cerebellar perfusion deficit volumes on all maps and the final infarction volume on follow‐up imaging were quantified. Uni‐ and multivariate regression analyses were performed. Sixty patients fulfilled the inclusion criteria. pc‐ASPECTS on CT angiography source images (ß, −9.239; 95% CI, −14.220 to −4.259; P<0.001) and cerebral blood flow deficit volume (ß, 0.886; 95% CI, 0.684 to 1.089; P<0.001) were significantly associated with final infarction volume in univariate linear regression analysis. The association of cerebral blood flow deficit volume (ß, 0.830; 95% CI, 0.605–1.055; P<0.001) was confirmed in a multivariate linear regression model adjusted for age, sex, pc‐ASPECTS on noncontrast CT, and CT angiography source images and the National Institutes of Health Stroke Scale score on admission. No other clinical or imaging parameters were associated with cerebellar stroke final infarction volume (P>0.05).

Conclusions

In contrast to noncontrast CT and CT angiography, WB‐CTP imaging contains prognostic information for morphologic outcome in patients with acute cerebellar stroke.

Clinical and Imaging Parameters Associated With Hyperacute Infarction Growth in Large Vessel Occlusion Stroke

Background and Purpose—

Large vessel occlusion stroke leads to highly variable hyperacute infarction growth. Our aim was to identify clinical and imaging parameters associated with hyperacute infarction growth in patients with an large vessel occlusion stroke of the anterior circulation.

Methods—

Seven hundred twenty-two consecutive patients with acute stroke were prospectively included in our monocentric stroke registry between 2009 and 2017. We selected all patients with a large vessel occlusion stroke of the anterior circulation, documented times from symptom onset, and CT perfusion on admission for our analysis (N=178). Ischemic core volume was determined with CT perfusion using automated thresholds. Hyperacute infarction growth was defined as ischemic core volume divided by times from symptom onset, assuming linear progression during times from symptom onset to imaging on admission. For collateral assessment, the regional leptomeningeal collateral score (rLMC) was used. Clinical data included the National Institutes of Health Stroke Scale score on admission and cardiovascular risk factors. Regression analysis was performed to adjust for confounders.

Results—

Median ischemic core volume was 34.4 mL, and median hyperacute infarction growth was 0.27 mL/min. In regression analysis including age, sex, National Institutes of Health Stroke Scale, clot burden score, diabetes mellitus, smoking, hypercholesteremia, hypertension, Alberta Stroke Program Early CT Score, and rLMC scores, only the rLMC score had a significant, independent association with hyperacute infarction growth (adjusted β=−0.35; P <0.001). Trichotomizing patients by rLMC scores yielded 65 patients with good (rLMC >15), 67 with intermediate (rLMC 11–15) and 46 with poor collaterals (rLMC <11) with an infarction growth of 0.17 mL/min, 0.26 mL/min, and 0.41 mL/min, respectively.

Conclusions—

Hyperacute infarction growth strongly depends on collaterals. In primary stroke centers, hyperacute infarction growth may be extrapolated to estimate the stroke progression during transfer times to thrombectomy centers and to support decisions on which patients to transfer.

Differential Benefit of Collaterals for Stroke Patients Treated with Thrombolysis or Supportive Care : A Propensity Score Matched Analysis

PURPOSE

Leptomeningeal collaterals can slow down infarction growth; however, despite good collaterals in the DAWN and DEFUSE 3 trials, outcomes were devastating if reperfusion was not attempted. The aim of this study was to compare the influence of collaterals on morphological and functional outcome in patients with acute middle cerebral artery (MCA) stroke undergoing intravenous thrombolysis (IVT) vs. supportive care (non-IVT).

METHODS

Out of 1639 consecutive patients examined with multiparametric computed tomography (CT) for suspected ischemic stroke, all patients with confirmed MCA stroke who did not undergo endovascular thrombectomy were selected. Propensity score matching (PSM) was used to match IVT and non-IVT treated patients for potential confounders including age, sex, National Institutes of Health Stroke Scale (NIHSS) score on admission, Alberta Stroke Program Early CT Score (ASPECTS), and occlusion site. Regression analysis after PSM was performed to identify independent associations.

RESULTS

After PSM, 90 IVT patients were matched with 90 non-IVT patients. In multivariable regression analysis, a high regional leptomeningeal collateral (rLMC) score was independently associated with lower final infarction volume (FIV) in the IVT group (b = -0.472, p < 0.001) but not in the non-IVT group (b = -0.116, p = 0.327). The trichotomized rLMC scores predicted functional outcome in IVT treated patients (adjusted odds ratio, aOR = 4.57, 95% confidence interval, CI, 1.03-20.32, p = 0.046) but showed no independent association with outcome in the non-IVT group (aOR = 0.69, 95% CI 0.07-6.80, p = 0.753).

CONCLUSION

Good collaterals favored smaller FIV and good functional outcome in IVT treated patients but not in non-IVT treated patients. Good collateral flow may have limited prognostic value if IVT is not administered to attempt reperfusion.

Effect of stroke thrombolysis predicted by distal vessel occlusion detection

OBJECTIVE

Among ischemic stroke patients with negative CT angiography (CTA), we aimed to determine the predictive value of enhanced distal vessel occlusion detection using CT perfusion postprocessing (waveletCTA) for the treatment effect of IV thrombolysis (IVT).

METHODS

Patients were selected from 1,851 consecutive patients who had undergone CT perfusion. Inclusion criteria were (1) significant cerebral blood flow (CBF) deficit, (2) no occlusion on CTA, and (3) infarction confirmed on follow-up. Favorable morphologic response was defined as smaller values of final infarction volume divided by initial CBF deficit volume (FIV/CBF). Favorable functional outcome was defined as modified Rankin Scale score of ≤2 after 90 days and decrease in NIH Stroke Scale score of ≥3 from admission to 24 hours (∆NIHSS).

RESULTS

Among patients with negative CTA (n = 107), 58 (54%) showed a distal occlusion on waveletCTA. There was no difference between patients receiving IVT (n = 57) vs supportive care (n = 50) regarding symptom onset, early ischemic changes, perfusion mismatch, or admission NIHSS score (all p > 0.05). In IVT-treated patients, the presence of an occlusion was an independent predictor of a favorable morphologic response (FIV/CBF: β -1.43; 95% confidence interval [CI] -1.96, -0.83; p = 0.001) and functional outcome (90-day modified Rankin Scale: odds ratio 7.68; 95% CI 4.33-11.51; p = 0.039; ∆NIHSS: odds ratio 5.76; 95% CI 3.98-8.27; p = 0.013), while it did not predict outcome in patients receiving supportive care (all p > 0.05).

CONCLUSION

In stroke patients with negative CTA, distal vessel occlusions as detected by waveletCTA are an independent predictor of a favorable response to IVT.