Improved performance and consistency of deep learning 3D liver segmentation with heterogeneous cancer stages in magnetic resonance imaging.

PURPOSE: Accurate liver segmentation is key for volumetry assessment to guide treatment decisions. Moreover, it is an important pre-processing step for cancer detection algorithms. Liver segmentation can be especially challenging in patients with cancer-related tissue changes and shape deformation. The aim of this study was to assess the ability of state-of-the-art deep learning 3D liver segmentation algorithms to generalize across all different Barcelona Clinic Liver Cancer (BCLC) liver cancer stages. METHODS: This retrospective study, included patients from an institutional database that had arterial-phase T1-weighted magnetic resonance images with corresponding manual liver segmentations. The data was split into 70/15/15% for training/validation/testing each proportionally equal across BCLC stages. Two 3D convolutional neural networks were trained using identical U-net-derived architectures with equal sized training datasets: one spanning all BCLC stages ("All-Stage-Net": AS-Net), and one limited to early and intermediate BCLC stages ("Early-Intermediate-Stage-Net": EIS-Net). Segmentation accuracy was evaluated by the Dice Similarity Coefficient (DSC) on a dataset spanning all BCLC stages and a Wilcoxon signed-rank test was used for pairwise comparisons. RESULTS: 219 subjects met the inclusion criteria (170 males, 49 females, 62.8±9.1 years) from all BCLC stages. Both networks were trained using 129 subjects: AS-Net training comprised 19, 74, 18, 8, and 10 BCLC 0, A, B, C, and D patients, respectively; EIS-Net training comprised 21, 86, and 22 BCLC 0, A, and B patients, respectively. DSCs (mean±SD) were 0.954±0.018 and 0.946±0.032 for AS-Net and EIS-Net (p<0.001), respectively. The AS-Net 0.956±0.014 significantly outperformed the EIS-Net 0.941±0.038 on advanced BCLC stages (p<0.001) and yielded similarly good segmentation performance on early and intermediate stages (AS-Net: 0.952±0.021; EIS-Net: 0.949±0.027; p = 0.107). CONCLUSION: To ensure robust segmentation performance across cancer stages that is independent of liver shape deformation and tumor burden, it is critical to train deep learning models on heterogeneous imaging data spanning all BCLC stages.

MR Imaging of the Testicular and Extratesticular Tumors: When Do We Need?

Testicular ultrasound is typically the first-line imaging examination in evaluating scrotal pathology. However, MR imaging can often provide valuable additional information, especially when ultrasound and/or clinical examinations are inconclusive. This is particularly evident when encountering testicular or paratesticular lesions, where accurate localization and characterization are paramount for management and prognosis. After reviewing normal scrotal anatomy as seen on MR imaging and offering a sample imaging protocol, the article describes specific indications for scrotal MR imaging and highlights imaging findings unique to various benign and malignant causes.

Does MR enterography offer added value after a recent CT in the evaluation of abdominal pain in Crohn's disease patients?

PURPOSE: To determine if there is added value in performing MR enterography shortly after a contrast-enhanced CT of the abdomen and pelvis in patients with Crohn's disease presenting with acute abdominal pain. MATERIALS AND METHODS: A total of 45 consecutive patients who underwent MRE examination <7 days after a CT met our strict inclusion criteria. Independent and blinded review of both MRE and CT exams was performed by two abdominal radiologists. RESULTS: There were no significant differences in detection of various findings associated with Crohn's disease between modalities. These findings included abscess, fistula, bowel wall thickening, free fluid, stricture, and bowel obstruction. There was moderate interobserver agreement with CT (Kappa: 0.52, 95% CI: [0.4-0.6]) and fair with MRE (Kappa: 0.36, 95% CI:[0.3 = 0.5]). CONCLUSION: The routine use of MRE after a diagnostic CT should be avoided, as it provides no additional valuable information, at the expense of extra patient risks, discomfort, and higher health care costs.

Magnetic Resonance Imaging Features of Endometrial Polyps: Frequency of Occurrence and Interobserver Reliability.

OBJECTIVE: This study aimed to evaluate magnetic resonance imaging (MRI) features and interobserver agreement of endometrial polyps. METHODS: After institutional review board approval, our database was searched for women older than 18 years who underwent MRI pelvis and pelvic surgical intervention from 2012 to 2016. Seventy-two patients with polyps and 75 controls composed the study cohort. Two radiologists evaluated the MRIs retrospectively for polyps. Polyp characteristics and enhancement were assessed. RESULTS: Sensitivity and specificity of readers 1 and 2 were 59.7% and 88.0%, and 44.4 and 96.0%, respectively. There was moderate agreement for presence of polyps (κ = 0.556, P ≤ 0001), T2 fibrous core, and intratumoral cysts, with slight agreement for T2 signal and enhancement. Polyp size moderately correlated with pathology (κ = 0.465 [P = 0.025] for reader 1, κ = 0.562 [P = 0.029] for reader 2). The most common enhancement was same as myometrium. CONCLUSION: Magnetic resonance imaging is moderately sensitive for detecting endometrial polyps, demonstrating features that are not sensitive but can be specific, with moderate interobserver agreement.

MRI-Ultrasound Fusion Targeted Biopsy of Prostate Imaging Reporting and Data System Version 2 Category 5 Lesions Found False-Positive at Multiparametric Prostate MRI.

OBJECTIVE: The purpose of this study was to determine imaging and clinical features associated with Prostate Imaging Reporting and Data System (PI-RADS) category 5 lesions identified prospectively at multiparametric MRI (mpMRI) that were found benign at MRI-ultrasound fusion targeted biopsy. MATERIALS AND METHODS: Between January 2015 and July 2016, 325 men underwent prostate mpMRI followed by MRI-ultrasound fusion targeted biopsy of 420 lesions prospectively identified and assessed with PI-RADS version 2. The frequency of clinically significant prostate cancer (defined as Gleason score ≥ 7) among PI-RADS 5 lesions was determined. Lesions with benign pathologic results were retrospectively reassessed by three abdominal radiologists and categorized as concordant or discordant between mpMRI and biopsy results. Multivariate logistic regression was used to identify factors associated with benign disease. Bonferroni correction was used. RESULTS: Of the 98 PI-RADS 5 lesions identified in 89 patients, 18% (18/98) were benign, 10% (10/98) were Gleason 6 disease, and 71% (70/98) were clinically significant prostate cancer. Factors associated with benign disease at multivariate analysis were lower prostate-specific antigen density (odds ratio [OR], 0.88; p < 0.001) and apex (OR, 3.54; p = 0.001) or base (OR, 7.11; p = 0.012) location. On secondary review of the 18 lesions with benign pathologic results, 39% (7/18) were scored as benign prostatic hyperplasia nodules, 28% (5/18) as inflammatory changes, 5% (1/18) as normal anatomic structures, and 28% (5/18) as discordant with imaging findings. CONCLUSION: PI-RADS 5 lesions identified during routine clinical interpretation are associated with a high risk of clinically significant prostate cancer. A benign pathologic result was significantly correlated with lower prostate-specific antigen density and apex or base location and most commonly attributed to a benign prostatic hyperplasia nodule. Integration of these clinical features may improve the interpretation of high-risk lesions identified with mpMRI.

Diagnosis of gastrogastric fistula on computed tomography: a quantitative approach.

OBJECTIVE: To determine if the attenuation of contrast material in the excluded stomach compared with the gastric pouch is helpful in diagnosing gastrogastric (GG) fistula. MATERIALS AND METHODS: In a retrospective study, 13 CT scans in 12 patients (age 43.2 ± 9.2, 10 females) who had undergone Roux-en-Y gastric bypass and who had oral contrast in both the gastric pouch and excluded stomach were qualitatively and quantitatively evaluated for GG fistula by two radiologists, using upper GI series (UGI) as the gold standard. Quantitative analysis was performed by computing the relative attenuation (RA) ratio (HU in excluded stomach/HU in gastric pouch). Statistical analysis was performed to determine if the RA ratio values correlated with the UGI findings of GG fistula. RESULTS: 46.2% (6/13) of UGI studies demonstrated a GG fistula. Statistical analysis demonstrated a significant difference in RA ratio (P < 0.05) between the fistula group (1.12 ± 0.29) and the reflux group (0.56 ± 0.19). A receiver operating characteristic analysis identified an RA ratio of 0.8 that maximized sensitivity (100%), at the expense of specificity (78.6%), for diagnosing GG fistula. In contrast, the initial qualitative evaluation for GG fistula yielded a lower sensitivity (45.8%) and a higher specificity (89.2%). After taking RA ratios into account, radiologists' final conclusions achieved higher sensitivity (58.3%) and specificity (100%). CONCLUSION: The relative attenuation ratio of oral contrast in the excluded stomach versus the gastric pouch can be a reliable tool in differentiating GG fistula from oral contrast reflux up the biliopancreatic limb on CT.

Risk of Clinically Significant Prostate Cancer Associated With Prostate Imaging Reporting and Data System Category 3 (Equivocal) Lesions Identified on Multiparametric Prostate MRI.

OBJECTIVE: The objective of this study is to determine the frequency of clinically significant cancer (CSC) in Prostate Imaging Reporting and Data System (PI-RADS) category 3 (equivocal) lesions prospectively identified on multiparametric prostate MRI and to identify risk factors (RFs) for CSC that may aid in decision making. MATERIALS AND METHODS: Between January 2015 and July 2016, a total of 977 consecutively seen men underwent multiparametric prostate MRI, and 342 underwent MRI-ultrasound (US) fusion targeted biopsy. A total of 474 lesions were retrospectively reviewed, and 111 were scored as PI-RADS category 3 and were visualized using a 3-T MRI scanner. Multiparametric prostate MR images were prospectively interpreted by body subspecialty radiologists trained to use PI-RADS version 2. CSC was defined as a Gleason score of at least 7 on targeted biopsy. A multivariate logistic regression model was constructed to identify the RFs associated with CSC. RESULTS: Of the 111 PI-RADS category 3 lesions, 81 (73.0%) were benign, 11 (9.9%) were clinically insignificant (Gleason score, 6), and 19 (17.1%) were clinically significant. On multivariate analysis, three RFs were identified as significant predictors of CSC: older patient age (odds ratio [OR], 1.13; p = 0.002), smaller prostate volume (OR, 0.94; p = 0.008), and abnormal digital rectal examination (DRE) findings (OR, 3.92; p = 0.03). For PI-RADS category 3 lesions associated with zero, one, two, or three RFs, the risk of CSC was 4%, 16%, 62%, and 100%, respectively. PI-RADS category 3 lesions for which two or more RFs were noted (e.g., age ≥ 70 years, gland size ≤ 36 mL, or abnormal DRE findings) had a CSC detection rate of 67% with a sensitivity of 53%, a specificity of 95%, a positive predictive value of 67%, and a negative predictive value of 91%. CONCLUSION: Incorporating clinical parameters into risk stratification algorithms may improve the ability to detect clinically significant disease among PI-RADS category 3 lesions and may aid in the decision to perform biopsy.

Standards for MRI reporting-the evolution to PI-RADS v 2.0.

Prostate cancer (PCa) remains a leading cause of death in the United States, but the vast majority of men diagnosed with PCa will die from other causes. While historically the capability of assessing the risk of life-threatening versus indolent PCa has relied heavily on serum prostate-specific antigen (PSA) and transrectal ultrasound (TRUS), multiparametric magnetic resonance imaging (mpMRI) has emerged as the leading tool for detection and characterization of clinically significant PCa. However, wide variations and lack of standardization of mpMRI data acquisition, interpretation, and reporting have hampered its progress. The development of a set of consensus guidelines, initially called Prostate Imaging and Reporting and Data System (PI-RADS) and eventually updated to a document called PI-RADS v2 has attempted to solve these shortcomings. As it stands, PI-RADS v2 currently represents the most up-to-date information on how to acquire, interpret, and report mpMRI of the prostate.

Magnetic resonance imaging of the scrotum: pictorial review with ultrasound correlation.

The purpose of this review is to showcase the added value of scrotal magnetic resonance imaging (MRI) in the workup of neoplastic and non-neoplastic entities. While ultrasound (US) remains the first-line imaging modality for evaluating scrotal pathology, MRI may add valuable information, particularly when US findings are equivocal. The inherent soft tissue resolution characteristics of MRI, as well as the ability to detect subtle enhancement and provide wider field-of-view imaging, can prove useful in evaluating inconclusive US findings. The added value of MR in these instances is critical as it may have a significant impact on patient management.

Negative Multiparametric Magnetic Resonance Imaging of the Prostate Predicts Absence of Clinically Significant Prostate Cancer on 12-Core Template Prostate Biopsy.

OBJECTIVE: To determine the negative predictive value of multiparametric magnetic resonance imaging (mpMRI), we evaluated the frequency of prostate cancer detection by 12-core template mapping biopsy in men whose mpMRI showed no suspicious regions. METHODS: Six hundred seventy patients underwent mpMRI followed by transrectal ultrasound (TRUS)-guided systematic prostate biopsy from December 2012 to June 2016. Of this cohort, 100 patients had a negative mpMRI. mpMRI imaging sequences included T2-weighted and diffusion-weighted imaging, and dynamic contrast enhancement sequences. RESULTS: The mean age, prostate-specific antigen, and prostate volume of the 100 men included were 64.3 years, 7.2 ng/mL, and 71 mL, respectively. Overall cancer detection was 27% (27 of 100). Prostate cancer was detected in 26.3% (10 of 38) of patients who were biopsy-naïve, 12.1% (4 of 33) of patients who had a prior negative biopsy, and in 44.8% (13 of 29) of patients previously on active surveillance; Gleason grade ≥7 was detected in 3% of patients overall (3 of 100). The negative predictive value of a negative mpMRI was 73% for all prostate cancer and 97% for Gleason ≥7 prostate cancer. CONCLUSION: There is an approximately 3% chance of detecting clinically significant prostate cancer with systematic TRUS-guided biopsy in patients with no suspicious findings on mpMRI. This information should help guide recommendations to patients about undergoing systematic TRUS-guided biopsy when mpMRI is negative.

Differentiating Renal Neoplasms From Simple Cysts on Contrast-Enhanced CT on the Basis of Attenuation and Homogeneity.

OBJECTIVE: The purpose of this study is to compare the attenuation and homogeneity of renal neoplasms with those of cysts on contrast-enhanced CT. MATERIALS AND METHODS: A total of 129 renal neoplasms and 24 simple cysts were evaluated. Two readers determined whether each mass was qualitatively heterogeneous or homogeneous. Mean, minimum, and maximum attenuation values were measured. Statistical analysis was performed. RESULTS: A total of 116 heterogeneous renal cell carcinomas (RCCs) (99 clear cell, four papillary, four oncocytic, seven chromophobe, and two unclassified RCCs), 13 homogeneous RCCs (10 papillary, two oncocytic, and one chromophobe RCC), and 24 cysts (all of which were homogeneous) were evaluated. All homogeneous RCCs had mean attenuation values of more than 42 HU, whereas renal cysts had mean attenuation values of up to 30 HU (p < 0.001). Two readers qualitatively and identically categorized all RCCs as homogeneous or heterogeneous (κ = 1.0; p < 0.001). CONCLUSION: Homogeneous simple renal cysts can have mean attenuation values of up to 30 HU, as determined by contrast-enhanced CT, whereas homogeneous RCCs have mean attenuation values as low as 42 HU, with no overlap occurring between the two groups. These data suggest that further evaluation of a homogeneous renal mass with a mean attenuation value of 30 HU or less on a contrast-enhanced CT scan likely is unwarranted.

Pelvic Inflammatory Disease: Multimodality Imaging Approach with Clinical-Pathologic Correlation.

Pelvic inflammatory disease (PID) is a common medical problem, with almost 1 million cases diagnosed annually. Historically, PID has been a clinical diagnosis supplemented with the findings from ultrasonography (US) or magnetic resonance (MR) imaging. However, the diagnosis of PID can be challenging because the clinical manifestations may mimic those of other pelvic and abdominal processes. Given the nonspecific clinical manifestations, computed tomography (CT) is commonly the first imaging examination performed. General CT findings of early- and late-stage PID include thickening of the uterosacral ligaments, pelvic fat stranding with obscuration of fascial planes, reactive lymphadenopathy, and pelvic free fluid. Recognition of these findings, as well as those seen with cervicitis, endometritis, acute salpingitis, oophoritis, pyosalpinx, hydrosalpinx, tubo-ovarian abscess, and pyometra, is crucial in allowing prompt and accurate diagnosis. Late complications of PID include tubal damage resulting in infertility and ectopic pregnancy, peritonitis caused by uterine and/or tubo-ovarian abscess rupture, development of peritoneal adhesions resulting in bowel obstruction and/or hydroureteronephrosis, right upper abdominal inflammation (Fitz-Hugh-Curtis syndrome), and septic thrombophlebitis. Recognition of these late manifestations at CT can also aid in proper patient management. At CT, careful assessment of common PID mimics, such as endometriosis, adnexal torsion, ruptured hemorrhagic ovarian cyst, adnexal neoplasms, appendicitis, and diverticulitis, is important to avoid misinterpretation, delay in management, and unnecessary surgery. Correlation with the findings from complementary imaging examinations, such as US and MR imaging, is useful for establishing a definitive diagnosis. (©)RSNA, 2016.

Homogeneous T1 Hyperintense Renal Lesions with Smooth Borders: Is Contrast-enhanced MR Imaging Needed?

Purpose To retrospectively determine if homogeneous high T1 signal intensity (SI) masses with smooth borders on unenhanced magnetic resonance (MR) images can be characterized as benign. Materials and Methods Institutional review board approval was obtained for this HIPAA-compliant retrospective study, with waiver of informed consent. MR images in 84 patients with hemorrhagic or proteinaceous cysts and 50 patients with renal cell carcinoma (RCC) were evaluated. Sixty-three cysts and 49 RCCs underwent unenhanced computed tomography (CT). SI ratio and CT attenuation were determined. Two radiologists evaluated lesions as follows: score 1, homogeneous with smooth borders; score 2, mildly heterogeneous with mildly lobulated borders; score 3, moderately heterogeneous and irregular borders; and score 4, markedly heterogeneous with markedly irregular borders. Statistical analysis was performed by using multivariable logistic regression, Welch t test, Z test, Fisher-exact test, Shapiro-Wilk test, and receiver operating characteristic curve analysis. A diagnostic criterion was formulated by using classification and regression tree analysis. Results SI ratio and attenuation of hemorrhagic or proteinaceous cysts were significantly higher than those of RCCs (SI ratio: cyst 2.4 ± 0.8, RCC 1.5 ± 0.3; attenuation: cyst 51.9 ± 21.5, RCC: 34.8 ± 10.0). Reader 1 scored morphology of 68 (81%) hemorrhagic or proteinaceous cysts as score 1 on MR images and as score 45 (71%) on CT scans. Reader 2 scored morphology of 59 (70%) hemorrhagic or proteinaceous cysts as score 1 on MR images and as score 43 (68%) on CT scans. Two-step classification tree suggested that homogeneous high T1 SI lesions with smooth borders and SI ratio of greater than 1.6 predict the lesion as benign cysts. Similar algorithm for CT suggested threshold of 51 HU. Increasing threshold to 2.5 for SI ratio and 66 for Hounsfield units resulted in 99.9% confidence for characterizing benign cysts. Conclusion The retrospective assessment shows that morphologic assessment and SI quantification on unenhanced T1-weighted MR images can be used to differentiate benign hemorrhagic or proteinaceous cysts from RCC, although prospective assessment will be needed to confirm these results. (©) RSNA, 2016.

Utility of Magnetic Resonance Imaging in the Evaluation of Intraoperatively Confirmed Pelvic Adhesions.

OBJECTIVE: The aim of this study was to evaluate the ability of magnetic resonance imaging (MRI) to identify pelvic adhesions. DESIGN: This was an institutional review board-approved retrospective analysis. METHODS: Ninety-nine patients met inclusion criteria and constituted our study population. INCLUSION CRITERIA: patients who underwent MRI and subsequent gynecologic abdominal surgery within 6 months after MRI. All imaging and operative reports were reviewed for the presence of pelvic adhesions by independent and blinded specialists. The findings were compared to calculate MRI sensitivity, specificity, positive predictive value, negative predictive value, and accuracy in the evaluation of pelvic adhesions. RESULTS: The specificity of MRI in the detection of pelvic adhesions was greater than 90% in all locations with the exception of the posterior cul-de-sac. Accuracy was highest in the anterior cul-de-sac at 88%. The positive predictive value was greater than 85% for prediction of non-location-specific adhesions. CONCLUSIONS: Magnetic resonance imaging is very specific in the evaluation of pelvic adhesions.

Gastric pneumatosis: Laboratory and imaging findings associated with mortality in adults.

AIM: To describe laboratory and imaging findings associated with mortality in patients with gastric pneumatosis. MATERIALS AND METHODS: Institution review board approval was obtained for this retrospective study. Using radiology report databases, all patients with "gastric pneumatosis" or "emphysematous gastritis" in their CT reports were identified from two institutions during 12 or 9 year periods. Clinical parameters and laboratory values [lactic acid, white blood cell (WBC) count, and serum creatinine] were obtained from medical records and images were reviewed in consensus by two readers. Bivariate associations between continuous variables were tested by Mann-Whitney tests. Fisher's exact test was used to evaluate bivariate associations between categorical variables. RESULTS: Of the 24 patients identified, there were five (21%) deaths. Median serum lactic acid and creatinine levels were significantly higher in patients who died compared to surviving patients [median (interquartile range, IQR): 1.95 (1.45-4.15) versus 1.5 (1.3-2.6), p = 0.001; 1.2 (1-2.8) versus 1 (0.8-1.4), p = 0.005, respectively). There was no significant difference in WBC levels between the groups. Coexistent small bowel pneumatosis and colonic pneumatosis were significantly more common in patients who died compared to surviving patients (80% versus 0%, p < 0.001; 40% versus 0%, p = 0.04, respectively). There was no significant difference for portal or mesenteric venous gas, free intraperitoneal gas, or dilated bowel. CONCLUSIONS: When the imaging finding of gastric pneumatosis was associated with elevated serum lactic acid, elevated serum creatinine, or concomitant small bowel or colonic pneumatosis, an association with mortality was observed. These findings suggest that more aggressive treatment may be warranted in patients with these laboratory or imaging abnormalities.