Anton S. Becker

Deep Learning in Mammography: Diagnostic Accuracy of a Multipurpose Image Analysis Software in the Detection of Breast Cancer

Objectives The aim of this study was to evaluate the diagnostic accuracy of a multipurpose image analysis software based on deep learning with artificial neural networks for the detection of breast cancer in an independent, dual-center mammography data set. Materials and Methods In this retrospective, Health Insurance Portability and Accountability Act-compliant study, all patients undergoing mammography in 2012 at our institution were reviewed (n = 3228). All of their prior and follow-up mammographies from a time span of 7 years (2008–2015) were considered as a reference for clinical diagnosis. After applying exclusion criteria (missing reference standard, prior procedures or therapies), patients with the first diagnosis of a malignoma or borderline lesion were selected (n = 143). Histology or clinical long-term follow-up served as reference standard. In a first step, a breast density-and age-matched control cohort was selected (n = 143) from the remaining patients with more than 2 years follow-up (n = 1003). The neural network was trained with this data set. From the publicly available Breast Cancer Digital Repository data set, patients with cancer and a matched control cohort were selected (n = 35 × 2). The performance of the trained neural network was also tested with this external data set. Three radiologists (3, 5, and 10 years of experience) evaluated the test data set. In a second step, the neural network was trained with all cases from January to September and tested with cases from October to December 2012 (screening-like cohort). The radiologists also evaluated this second test data set. The areas under the receiver operating characteristic curve between readers and the neural network were compared. A Bonferroni-corrected P value of less than 0.016 was considered statistically significant. Results Mean age of patients with lesion was 59.6 years (range, 35–88 years) and in controls, 59.1 years (35–83 years). Breast density distribution (A/B/C/D) was 21/59/42/21 and 22/60/41/20, respectively. Histologic diagnoses were invasive ductal carcinoma in 90, ductal in situ carcinoma in 13, invasive lobular carcinoma in 13, mucinous carcinoma in 3, and borderline lesion in 12 patients. In the first step, the area under the receiver operating characteristic curve of the trained neural network was 0.81 and comparable on the test cases 0.79 (P = 0.63). One of the radiologists showed almost equal performance (0.83, P = 0.17), whereas 2 were significantly better (0.91 and 0.94, P < 0.016). In the second step, performance of the neural network (0.82) was not significantly different from the human performance (0.77–0.87, P > 0.016); however, radiologists were consistently less sensitive and more specific than the neural network. Conclusions Current state-of-the-art artificial neural networks for general image analysis are able to detect cancer in mammographies with similar accuracy to radiologists, even in a screening-like cohort with low breast cancer prevalence.

Hybrid PET/MR Imaging: An Algorithm to Reduce Metal Artifacts from Dental Implants in Dixon-Based Attenuation Map Generation Using a Multiacquisition Variable-Resonance Image Combination Sequence

It was the aim of this study to implement an algorithm modifying Dixon-based MR imaging datasets for attenuation correction in hybrid PET/MR imaging with a multiacquisition variable resonance image combination (MAVRIC) sequence to reduce metal artifacts. Methods: After ethics approval, in 8 oncologic patients with dental implants data were acquired in a trimodality setup with PET/CT and MR imaging. The protocol included a whole-body 3-dimensional dual gradient-echo sequence (Dixon) used for MR imaging–based PET attenuation correction and a high-resolution MAVRIC sequence, applied in the oral area compromised by dental implants. An algorithm was implemented correcting the Dixon-based μ maps using the MAVRIC in areas of Dixon signal voids. The artifact size of the corrected μ maps was compared with the uncorrected MR imaging μ maps. Results: The algorithm was robust in all patients. There was a significant reduction in mean artifact size of 70.5% between uncorrected and corrected μ maps from 697 ± 589 mm2 to 202 ± 119 mm2 (P = 0.016). Conclusion: The proposed algorithm could improve MR imaging–based attenuation correction in critical areas, when standard attenuation correction is hampered by metal artifacts, using a MAVRIC.

Adult neurogenesis and its anatomical context in the hippocampus of three mole-rat species

African mole-rats (family Bathyergidae) are small to medium sized, long-lived, and strictly subterranean rodents that became valuable animal models as a result of their longevity and diversity in social organization. The formation and integration of new hippocampal neurons in adult mammals (adult hippocampal neurogenesis, AHN) correlates negatively with age and positively with habitat complexity. Here we present quantitative data on AHN in wild-derived mole-rats of 1 year and older, and briefly describe its anatomical context including markers of neuronal function (calbindin and parvalbumin). Solitary Cape mole-rats (Georychus capensis), social highveld mole-rats (Cryptomys hottentotus pretoriae), and eusocial naked mole-rats (Heterocephalus glaber) were assessed. Compared to other rodents, the hippocampal formation in mole-rats is small, but shows a distinct cytoarchitecture in the dentate gyrus and CA1. Distributions of the calcium-binding proteins differ from those seen in rodents; e.g., calbindin in CA3 of naked mole-rats distributes similar to the pattern seen in early primate development, and calbindin staining extends into the stratum lacunosum-moleculare of Cape mole-rats. Proliferating cells and young neurons are found in low numbers in the hippocampus of all three mole-rat species. Resident granule cell numbers are low as well. Proliferating cells expressed as a percentage of resident granule cells are in the range of other rodents, while the percentage of young neurons is lower than that observed in surface dwelling rodents. Between mole-rat species, we observed no difference in the percentage of proliferating cells. The percentages of young neurons are high in social highveld and naked mole-rats, and low in solitary Cape mole-rats. The findings support that proliferation is regulated independently of average life expectancy and habitat. Instead, neuronal differentiation reflects species-specific demands, which appear lower in subterranean rodents.

The IVIM signal in the healthy cerebral gray matter: A play of spherical and non-spherical components

ABSTRACT The intra‐voxel incoherent motion (IVIM) model assumes that blood flowing in isotropically distributed capillary segments induces a phase dispersion of the MR signal, which increases the signal attenuation in diffusion‐weighted images. However, in most tissue types the capillary network has an anisotropic micro‐architecture. In this study, we investigated the possibility to indirectly infer the anisotropy of the capillary network in the healthy cerebral gray matter by evaluating the dependence of the IVIM signal from the direction of the diffusion‐encoding. Perfusion‐related indices and self‐diffusion were modelled as symmetric rank 2 tensors. The geometry of the tensors was quantified pixel‐wise by decomposing the tensor in sphere‐like, plane‐like, and line‐like components. Additionally, trace and fractional anisotropy of the tensors were computed. While the self‐diffusion tensor is dominated by a spherical geometry with a residual contribution of the non‐spherical components, both, fraction of perfusion and pseudo‐diffusion, present a substantial (in the order of 30%) contribution of planar and linear components to the tensor metrics. This study shows that the IVIM perfusion estimates in the cerebral gray matter present a detectable deviation from the spherical model. These non‐spherical components may reflect the direction‐dependent morphology of the microcirculation. Therefore, the tensor generalization of the IVIM model may provide a tool for the non‐invasive monitoring of cerebral capillary micro‐architecture during development, aging or in pathologies. Graphical abstract Figure. No Caption available. HighlightsTensor analysis reveals anisotropy of the IVIM signal in cerebral gray matter.Planar and linear components contribute to approx. 30% of the metric of the tensors.Fraction of perfusion anisotropy may reflect the morphometry of the microcirculation.The anisotropy of the pseudo‐diffusion may provide functional information.IVIM tensor imaging allows for quantitative characterization of microcirculation.

Dose Reduction in Tomosynthesis of the Wrist

OBJECTIVE The purpose of this study was to quantitatively and qualitatively determine the impact of radiation dose reduction on the image noise and quality of tomosynthesis studies of the wrist. MATERIALS AND METHODS Imaging of six cadaver wrists was performed with tomosynthesis in anteroposterior position at a tube voltage of 60 kV and tube current of 80 mA and subsequently at 60 or 50 kV with different tube currents of 80, 40, or 32 mA. Dose-area products (DAP) were obtained from the electronically logged protocol. Image noise was measured with an ROI. Two independent and blinded readers evaluated all images. Interreader agreement was measured with a Cohen kappa. Readers assessed overall quality and delineation of soft tissue, cortical bone, and trabecular bone on a 4-point Likert scale. RESULTS The highest DAP (3.892 ± 0.432 Gy · cm2) was recorded for images obtained with 60 kV and 80 mA; the lowest (0.857 ± 0.178 Gy · cm2) was recorded for images obtained with 50 kV and 32 mA. Noise was highest when a combination of 50 kV and 32 mA (389 ± 26.6) was used and lowest when a combination of 60 kV and 80 mA (218 ± 12.3) was used. The amount of noise on images acquired using 60 kV and 80 mA was statistically significantly different from the amount measured on all other images (p < 0.0001). Interreader agreement was excellent (κ = 0.93). Delineation of anatomy and overall quality were scored best on images obtained with 60 kV and 80 mA and worst on images obtained with 50 kV and 32 mA. The difference in delineation and quality on images obtained using 50 kV and 40 mA was not statistically significantly different compared with images obtained using 60 kV and 80 mA. CONCLUSION Significant dose reduction for tomosynthesis of the wrist is possible while image quality and delineation of anatomic structures remain preserved.

Anatomical Grading for Metabolic Activity of Brown Adipose Tissue

Background Recent advances in obesity research suggest that BAT activity, or absence thereof, may be an important factor in the growing epidemic of obesity and its manifold complications. It is thus important to assess larger populations for BAT-activating and deactivating factors. 18FDG-PET/CT is the standard method to detect and quantify metabolic BAT activity, however, the manual measurement is not suitable for large studies due to its time-consuming nature and poor reproducibility across different software and devices. Methodology/Main Findings In a retrospective study, 1060 consecutive scans of 1031 patients receiving a diagnostic 18FDG-PET/CT were examined for the presence of active BAT. Patients were classified according to a 3-tier system (supraclavicular, mediastinal, infradiaphragmatic) depending on the anatomical location of their active BAT depots, with the most caudal location being the decisive factor. The metabolic parameters (maximum activity, total volume and total glycolysis) were measured on a standard PET/CT workstation. Mean age of the population was 60±14.6y. 41.61% of patients were female. Metabolically active BAT was found in 53 patients (5.1%). Female, younger and leaner patients tended to have more active BAT, higher metabolic activity and more caudally active BAT. In total, 15 patients showed only supraclavicular, 27 additional mediastinal, and 11 infradiaphragmal activity. Interestingly, the activation of BAT always followed a cranio-caudal gradient. This anatomical pattern correlated with age and BMI as well as with all metabolic parameters, including maximum and total glycolysis (p<0.001). Conclusion Based on our data we propose a simple method to grade or quantify the degree of BAT amount/activity in patients based on the most caudally activated depot. As new modalities for BAT visualization may arise in the future, this system would allow direct comparability with other modalities, in contrary to the PET-metrics, which are restricted to 18FDG-PET/CT.

Diffusion-weighted imaging of the abdomen: Impact of b-values on texture analysis features.

The purpose of this work was to systematically assess the impact of the b‐value on texture analysis in MR diffusion‐weighted imaging (DWI) of the abdomen.

Direct comparison of PI-RADS version 2 and version 1 regarding interreader agreement and diagnostic accuracy for the detection of clinically significant prostate cancer

PURPOSE to simultaneously evaluate interreader agreement and diagnostic accuracy in the of PI-RADS v2 and compare it to v1. METHODS A total of 67 patients (median age 65.3 y, range 51.2-78.2 y; PSA 6.8μg/L, 0.2-33μg/L) undergoing MRI of the prostate and subsequent transperineal template biopsy within ≤6 months from MRI were included. Four readers from two institutions evaluated the likelihood of prostate cancer using PI-RADS v1 and v2 in two separate reading sessions ≥3 months apart. Interreader agreement was assessed for each pulse-sequence and for total PI-RADS scores using the intraclass correlation coefficient (ICC). Differences were considered significant for non-overlapping 95%-confidence intervals. Diagnostic accuracy was assessed with the area under the receiver operating characteristic curve (AZ). A p-value <0.05 was considered statistically significant. RESULTS Interreader agreement for DCE-scores was good in v2 (ICC2=0.70; 95% CI: 0.66-0.74) and slightly lower in v1 (ICC1=0.64, 0.59-0.69). Agreement for DWI scores (ICC1=0.77, ICC2=0.76) as well as final PI-RADS scores per quadrant were nearly identical (ICC1=ICC2=0.71). Diagnostic accuracy showed no significant differences (p=0.09-0.93) between v1 and v2 in any of the readers (range: AZ=0.78-0.88). CONCLUSION PI-RADS scores show similar interreader agreement in v2 and v1 at comparable diagnostic performance. The simplification of the DCE interpretation in v2 might slightly improve agreement while not negatively affecting diagnostic performance.

Assessment of Cervical Cancer with a Parameter-Free Intravoxel Incoherent Motion Imaging Algorithm

Objective To evaluate the feasibility of a parameter-free intravoxel incoherent motion (IVIM) approach in cervical cancer, to assess the optimal b-value threshold, and to preliminarily examine differences in the derived perfusion and diffusion parameters for different histological cancer types. Materials and Methods After Institutional Review Board approval, 19 female patients (mean age, 54 years; age range, 37–78 years) gave consent and were enrolled in this prospective magnetic resonance imaging study. Clinical staging and biopsy results were obtained. Echo-planar diffusion weighted sequences at 13 b-values were acquired at 3 tesla field strength. Single-sliced region-of-interest IVIM analysis with adaptive b-value thresholds was applied to each tumor, yielding the optimal fit and the optimal parameters for pseudodiffusion (D*), perfusion fraction (Fp) and diffusion coefficient (D). Monoexponential apparent diffusion coefficient (ADC) was calculated for comparison with D. Results Biopsy revealed squamous cell carcinoma in 10 patients and adenocarcinoma in 9. The b-value threshold (median [interquartile range]) depended on the histological type and was 35 (22.5–50) s/mm2 in squamous cell carcinoma and 150 (100–150) s/mm2 in adenocarcinoma (p < 0.05). Comparing squamous cell vs. adenocarcinoma, D* (45.1 [25.1–60.4] × 10−3 mm2/s vs. 12.4 [10.5–21.2] × 10−3 mm2/s) and Fp (7.5% [7.0–9.0%] vs. 9.9% [9.0–11.4%]) differed significantly between the subtypes (p < 0.02), whereas D did not (0.89 [0.75–0.94] × 10−3 mm2/s vs. 0.90 [0.82–0.97] × 10−3 mm2/s, p = 0.27). The residuals did not differ (0.74 [0.60–0.92] vs. 0.94 [0.67–1.01], p = 0.32). The ADC systematically underestimated the magnitude of diffusion restriction compared to D (p < 0.001). Conclusion The parameter-free IVIM approach is feasible in cervical cancer. The b-value threshold and perfusion-related parameters depend on the tumor histology type.

Intravoxel incoherent motion imaging measurement of perfusion changes in the parotid gland provoked by gustatory stimulation: A pilot study: Dynamic Parotid IVIM

To demonstrate the feasibility of intravoxel incoherent motion imaging (IVIM) for quantification of perfusion changes in the parotid gland after gustatory stimulation.