Faculty Bibliography

PURPOSE/OBJECTIVE:To evaluate the accuracy of a deep learning software (DLS) in the discrimination between phyllodes tumors (PT) and fibroadenomas (FA). METHODS:In this IRB-approved, retrospective, single-center study, we collected all ultrasound images of histologically secured PT (n = 11, 36 images) and a random control group with FA (n = 15, 50 images). The images were analyzed with a DLS designed for industrial grade image analysis, with 33 images withheld from training for validation purposes. The lesions were also interpreted by four radiologists. Diagnostic performance was assessed by the area under the receiver operating characteristic curve (AUC). Sensitivity, specificity, negative and positive predictive values were calculated at the optimal cut-off (Youden Index). RESULTS:The DLS was able to differentiate between PT and FA with good diagnostic accuracy (AUC = 0.73) and high negative predictive value (NPV = 100%). Radiologists showed comparable accuracy (AUC 0.60-0.77) at lower NPV (64-80%). When performing the readout together with the DLS recommendation, the radiologist's accuracy showed a non-significant tendency to improve (AUC 0.75-0.87, p = 0.07). CONCLUSION/CONCLUSIONS:Deep learning based image analysis may be able to exclude PT with a high negative predictive value. Integration into the clinical workflow may enable radiologists to more confidently exclude PT, thereby reducing the number of unnecessary biopsies.

OBJECTIVES/OBJECTIVE:To develop age-, gender-, and regional-specific normative values for texture analysis (TA) of spinal computed tomography (CT) in subjects with normal bone mineral density (BMD), as defined by dual X-ray absorptiometry (DXA), and to determine age-, gender-, and regional-specific differences. MATERIALS AND METHODS/METHODS:In this retrospective, IRB-approved study, TA was performed on sagittal CT bone images of the thoracic and lumbar spine using dedicated software (MaZda) in 141 individuals with normal DXA BMD findings. Numbers of female and male subjects were balanced in each of six age decades. Three hundred and five TA features were analyzed in thoracic and lumbar vertebrae using free-hand regions-of-interest. Intraclass correlation (ICC) coefficients were calculated for determining intra- and inter-observer agreement of each feature. Further dimension reduction was performed with correlation analyses. RESULTS:The TA features with an ICC < 0.81 indicating compromised intra- and inter-observer agreement and with Pearson correlation scores r > 0.8 with other features were excluded from further analysis for dimension reduction. From the remaining 31 texture features, a significant correlation with age was found for the features mean (r = -0.489, p < 0.001), variance (r = -0.681, p < 0.001), kurtosis (r = 0.273, p < 0.001), and WavEnLL_s4 (r = 0.273, p < 0.001). Significant differences were found between genders for various higher-level texture features (p < 0.001). Regional differences among the thoracic spine, thoracic-lumbar junction, and lumbar spine were found for most TA features (p < 0.021). CONCLUSION/CONCLUSIONS:This study established normative values of TA features on CT images of the spine and showed age-, gender-, and regional-specific differences in individuals with normal BMD as defined by DXA.

BACKGROUND:Texture analysis in oncological magnetic resonance imaging (MRI) may yield surrogate markers for tumor differentiation and staging, both of which are important factors in the treatment planning for cervical cancer. PURPOSE/OBJECTIVE:To identify texture features which may predict tumor differentiation and nodal status in diffusion-weighted imaging (DWI) of cervical carcinoma. MATERIAL AND METHODS/METHODS:value < 0.05 was considered statistically significant. RESULTS: = 0.02), were higher in patients with positive nodal status. Cluster analysis revealed several co-correlations. CONCLUSION/CONCLUSIONS:We identified potentially predictive GLM features for histological tumor differentiation and histogram features for nodal cancer stage.

PURPOSE/OBJECTIVE:to simultaneously evaluate interreader agreement and diagnostic accuracy in the of PI-RADS v2 and compare it to v1. METHODS:). A p-value <0.05 was considered statistically significant. RESULTS:=0.78-0.88). CONCLUSION/CONCLUSIONS: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.

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.

PURPOSE/OBJECTIVE:To evaluate the interreader agreement of a three-tier craniocaudal grading system for brown fat activation and investigate the accuracy of the distinction between the three grades. MATERIALS AND METHODS/METHODS:After IRB approval, 340 cases were retrospectively selected from patients undergoing (18)FDG-PET/CT between 2007 and 2015 at our institution, with 85 cases in each grade and 85 controls with no active brown fat. Three readers evaluated all cases independently. Furthermore standardized uptake values (SUV) measurements were performed by two readers in a subset of 53 cases. Agreement between the readers was assessed with Cohen's Kappa (k), the concordance correlation coefficient (CCC) and the intraclass correlation coefficient (ICC). Accuracy was assessed with Bland-Altman and receiver operating characteristics (ROC) analysis. A Bonferroni-corrected two-tailed p<0.016 was considered statistically significant. RESULTS:Agreement for BAT grade was excellent by all three metrics with k=0.83-0.89, CCC=0.83-0.89 and ICC=0.91-0.94. Bland-Altman analysis revealed only slight average over- or underestimation (-0.01-0.14) with the majority of disagreements within one grade. ROC analysis yielded slightly less accurate classification between higher vs. lower grades (Area under the ROC curves 0.78-0.84 vs. 0.88-0.92) but no significant differences between readers. Agreement was also excellent for the maximum SUV and the total brown fat volume (k=0.90 and 0.94, CCC=0.93 and 0.99, ICC=0.96 and 0.99), but Bland-Altman plots revealed a tendency to underestimate activity by one of the readers. CONCLUSION/CONCLUSIONS:Grading the activation of brown fat by assessment of the most caudally activated depots results in excellent interreader agreement, comparable to SUV measurements.

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.