Comparing Diagnostic Accuracy of Radiologists versus GPT-4V and Gemini Pro Vision Using Image Inputs from Diagnosis Please Cases.

Background The diagnostic abilities of multimodal large language models (LLMs) using direct image inputs and the impact of the temperature parameter of LLMs remain unexplored. Purpose To investigate the ability of GPT-4V and Gemini Pro Vision in generating differential diagnoses at different temperatures compared with radiologists using Radiology Diagnosis Please cases. Materials and Methods This retrospective study included Diagnosis Please cases published from January 2008 to October 2023. Input images included original images and captures of the textual patient history and figure legends (without imaging findings) from PDF files of each case. The LLMs were tasked with providing three differential diagnoses, repeated five times at temperatures 0, 0.5, and 1. Eight subspecialty-trained radiologists solved cases. An experienced radiologist compared generated and final diagnoses, considering the result correct if the generated diagnoses included the final diagnosis after five repetitions. Accuracy was assessed across models, temperatures, and radiology subspecialties, with statistical significance set at P < .007 after Bonferroni correction for multiple comparisons across the LLMs at the three temperatures and with radiologists. Results A total of 190 cases were included in neuroradiology (n = 53), multisystem (n = 27), gastrointestinal (n = 25), genitourinary (n = 23), musculoskeletal (n = 17), chest (n = 16), cardiovascular (n = 12), pediatric (n = 12), and breast (n = 5) subspecialties. Overall accuracy improved with increasing temperature settings (0, 0.5, 1) for both GPT-4V (41% [78 of 190 cases], 45% [86 of 190 cases], 49% [93 of 190 cases], respectively) and Gemini Pro Vision (29% [55 of 190 cases], 36% [69 of 190 cases], 39% [74 of 190 cases], respectively), although there was no evidence of a statistically significant difference after Bonferroni adjustment (GPT-4V, P = .12; Gemini Pro Vision, P = .04). The overall accuracy of radiologists (61% [115 of 190 cases]) was higher than that of Gemini Pro Vision at temperature 1 (T1) (P < .001), while no statistically significant difference was observed between radiologists and GPT-4V at T1 after Bonferroni adjustment (P = .02). Radiologists (range, 45%-88%) outperformed the LLMs at T1 (range, 24%-75%) in most subspecialties. Conclusion Using direct radiologic image inputs, GPT-4V and Gemini Pro Vision showed improved diagnostic accuracy with increasing temperature settings. Although GPT-4V slightly underperformed compared with radiologists, it nonetheless demonstrated promising potential as a supportive tool in diagnostic decision-making. © RSNA, 2024 See also the editorial by Nishino and Ballard in this issue.

N-(2,3-dihydroxypropyl) carbamoyl side chain: a potentially significant factor for recurrent iodinated contrast medium-related adverse drug reactions.

PURPOSE: To determine whether switching to contrast media based on the sharing of N-(2,3-dihydroxypropyl) carbamoyl side chain reduces the recurrence of iodinated contrast media (ICM)-associated adverse drug reactions (ADRs). MATERIALS AND METHODS: This single-center retrospective study included 2133 consecutive patients (mean age ± SD, 56.1 ± 11.4 years; male, 1052 [49.3%]) who had a history of ICM-associated ADRs and underwent contrast-enhanced CT examinations. The per-patient and per-exam-based recurrence ADR rates were compared between cases of switching and non-switching the ICM from ICMs that caused the previous ADRs, and between cases that used ICMs with common and different carbamoyl side chains from ICMs that caused the previous ADRs. Downgrade rates (no recurrence or the occurrence of ADR less severe than index ADRs) were also compared. Propensity score matching (PSM) and inverse probability of treatment weighting (IPTW) analysis were additionally performed. RESULTS: In per-patient analysis, switching of ICM showed a lower recurrence rate (switching, 10.4% [100/965] vs. non-switching, 28.4% [332/1168]), with the adjusted odds ratio (OR) of 0.27 (95% CI: 0.21, 0.34; p < 0.001). The result was consistent in PSM (OR, 0.29 [95% CI: 0.22, 0.39]; p < 0.001), IPTW (OR, 0.28 [95% CI: 0.22, 0.36]; p < 0.001), and in per-exam analysis (5.5% vs. 13.8%; OR, 0.32 [95% CI: 0.27, 0.37]; p < 0.001). There was lower per-exam recurrence (5.0% [195/3938] vs. 7.8% [79/1017]; OR, 0.63 [95% CI: 0.47, 0.83]; p = 0.001) and higher downgrade rates (95.6% [3764/3938] vs. 93.3% [949/1017]; OR, 1.51 [95% CI: 1.12, 2.03]; p = 0.006) when using different side chain groups. CONCLUSION: Switching to an ICM with a different carbamoyl side chain reduced the recurrent ADRs and their severity during subsequent examinations. CLINICAL RELEVANCE STATEMENT: Switching to an iodinated contrast media with a different carbamoyl side chain reduced the recurrent adverse drug reactions and their severity during subsequent examinations.

Focal nodular enhancement on DCE MRI of the prostatectomy bed: radiologic-pathologic correlations and prognostic value.

OBJECTIVES: To determine the concordance of dynamic contrast-enhanced (DCE) imaging findings with clinico-pathologic characteristics and their prognostic impact for predicting biochemical recurrence (BCR) in patients who underwent radical prostatectomy (RP) for prostate cancer. METHODS: This retrospective study included patients who underwent MRI within 1 year after RP between November 2019 and October 2020. DCE findings and their concordance with the presence and location of positive surgical margin (PSM) were assessed using RP specimens. Kaplan-Meier and logistic regression analyses were used to evaluate the prognostic impact of DCE findings for BCR. RESULTS: Among the 272 men (mean age ± standard deviation, 66.6 ± 7.4 years), focal nodular enhancement was more frequently observed in those with PSM compared to those with negative margin (85.4% versus 14.6%; p < 0.001). The sites of focal nodular enhancement were 72.9% (35/48) concordant with the PSM locations. Focal nodular enhancement was associated with a higher Gleason score, higher preoperative PSA (≥ 10 ng/mL), higher Gleason grade at the surgical margin, and non-limited margin involvement (p = 0.002, 0.006, 0.032, and 0.001, respectively). In patients without BCR at the time of MRI, focal nodular enhancement was associated with a shorter time to BCR (p < 0.001) and a significant factor predicting 1-year BCR in both univariate (odds ratio = 8.4 [95% CI: 2.5-28.3]; p = 0.001) and multivariate (odds ratio = 5.49 [1.56-19.3]; p = 0.008) analyses. CONCLUSIONS: Focal nodular enhancement on post-prostatectomy MRI was associated with adverse clinico-pathologic characteristics of high risk for recurrence and can be a predictor for 1-year BCR in patients undergoing RP. KEY POINTS: • Focal nodular enhancement (PI-RR DCE score ≥ 4) was 72.9% (35/48) concordant with the site of positive resection margin by radiologic-histologic correlation. • Focal nodular enhancement (PI-RR DCE score ≥ 4) was associated with higher Gleason score ( ≥ 8), preoperative PSA ( > 10 ng/mL), and Gleason grade 4 or 5 at the surgical margin and non-limited margin involvement (p ≤ 0.032). • In patients without BCR at the time of MRI, focal nodular enhancement was a significant factor predicting 1-year BCR (odds ratio = 5.49; 95% CI: 1.56-19.3; p = 0.008).

Proportion of malignancy in Bosniak classification of cystic renal masses version 2019 (v2019) classes: systematic review and meta-analysis.

OBJECTIVES: Determine the proportion of malignancy within Bosniak v2019 classes. METHODS: MEDLINE and EMBASE were searched. Eligible studies contained patients with cystic renal masses undergoing CT or MRI renal protocol examinations with pathology confirmation, applying Bosniak v2019. Proportion of malignancy was estimated within Bosniak v2019 class. Risk of bias was assessed using QUADAS-2. RESULTS: We included 471 patients with 480 cystic renal masses. No class I malignant masses were observed. Pooled proportion of malignancy were class II, 12% (6/51, 95% CI 5-24%); class IIF, 46% (37/85, 95% CI 28-66%); class III, 79% (138/173, 95% CI 68-88%); and class IV, 84% (114/135, 95% CI 77-90%). Proportion of malignancy differed between Bosniak v2019 II-IV classes (p = 0.004). Four studies reported the proportion of malignancy by wall/septa feature. The pooled proportion of malignancy with 95% CI were class III thick smooth wall/septa, 77% (41/56, 95% CI 53-91%); class III obtuse protrusion ≤ 3 mm (irregularity), 83% (97/117, 95% CI 75-89%); and class IV nodule with acute angulation, 86% (50/58, 95% CI 75-93%) or obtuse angulation ≥ 4 mm, 83%, (64/77, 95% CI 73-90%). Subgroup analysis by wall/septa feature was limited by sample size; however, no differences were found comparing class III masses with irregularity to class IV masses (p = 0.74) or between class IV masses by acute versus obtuse angles (p = 0.62). CONCLUSION: Preliminary data suggest Bosniak v2019 class IIF masses have higher proportion of malignancy compared to the original classification, controlling for pathologic reference standard. There are no differences in proportion of malignancy comparing class III masses with irregularities to class IV masses with acute or obtuse nodules. KEY POINTS: • The proportion of malignancy in Bosniak v2019 class IIF cystic masses is 46% (37 malignant/85 total IIF masses, 95% confidence intervals (CI) 28-66%). • The proportion of malignancy in Bosniak v2019 class III cystic masses is 79% (138/173, 95% CI 68-88%) and in Bosniak v2019 class IV cystic masses is 84% (114/135, 95% CI 77-90%). • Class III cystic masses with irregularities had similar proportion of malignancy (83%, 97/117, 95% CI 75-89%) compared to Bosniak class IV masses (84%, 114/135, 95% CI 77-90%) overall (p = 0.74) with no difference within class IV masses by acute versus obtuse angulation (p = 0.62).

Fast T2-Weighted Imaging With Deep Learning-Based Reconstruction: Evaluation of Image Quality and Diagnostic Performance in Patients Undergoing Radical Prostatectomy.

BACKGROUND: Deep learning-based reconstruction (DLR) can potentially improve image quality by reduction of noise, thereby enabling fast acquisition of magnetic resonance imaging (MRI). However, a systematic evaluation of image quality and diagnostic performance of MRI using short acquisition time with DLR has rarely been investigated in men with prostate cancer. PURPOSE: To assess the image quality and diagnostic performance of MRI using short acquisition time with DLR for the evaluation of extraprostatic extension (EPE). STUDY TYPE: Retrospective. POPULATION: One hundred and nine men. FIELD STRENGTH/SEQUENCE: 3 T; turbo spin echo T2-weighted images (T2WI), echo-planar diffusion-weighted, and spoiled gradient echo dynamic contrast-enhanced images. ASSESSMENT: To compare image quality, signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) and subjective analysis using Likert scales on three T2WIs (MRI using conventional acquisition time, MRI using short acquisition time [fast MRI], and fast MRI with DLR) were performed. The diagnostic performance for EPE was evaluated by three independent readers. STATISTICAL TESTS: SNR, CNR, and image quality scores across the three imaging protocols were compared using Friedman tests. The diagnostic performance for EPE was assessed using the area under receiver operating characteristic curves (AUCs). P < 0.05 was considered statistically significant. RESULTS: Fast MRI with DLR demonstrated significantly higher SNR (mean ± SD, 14.7 ± 6.8 vs. 8.8 ± 4.9) and CNR (mean ± SD, 6.5 ± 6.3 vs. 3.4 ± 3.6) values and higher image quality scores (median, 4.0 vs. 3.0 for three readers) than fast MRI. The AUCs for EPE were significantly higher with the use of DLR (0.86 vs. 0.75 for reader 2 and 0.82 vs. 0.73 for reader 3) compared with fast MRI, whereas differences were not significant for reader 1 (0.81 vs. 0.74; P = 0.09). DATA CONCLUSION: DLR may be useful in reducing the acquisition time of prostate MRI without compromising image quality or diagnostic performance. LEVEL OF EVIDENCE: 4 TECHNICAL EFFICACY: Stage 3.

Ancillary imaging and clinical features for the characterization of prostate lesions: A proposed approach to reduce false positives.

The relatively low specificity and positive predictive value of the Prostate Imaging-Reporting and Data System (PI-RADS) can lead to considerable false-positive results and unnecessary biopsies. The aim of this study was to propose ancillary features (AFs) indicating clinically significant prostate cancer (csPCa) or benign tissues in PI-RADS category ≥3 lesions and determine the usefulness of these AFs in reducing false-positive assessments of suspicious lesions in men at csPCa risk. This was a retrospective study, which included 199 men. A 3T, including turbo spin echo T2 -weighted, echo-planar diffusion-weighted, and spoiled gradient echo dynamic contrast-enhanced (DCE) images, was used. Five AFs (prostate-specific antigen density ≥0.15 ng/mL2 ; size ≥10 mm; heterogeneous T2 signal intensity; circumscribed nodule in the junction of peripheral and transition zone; and DCE time curves) indicating csPCa or non-csPCa were evaluated by three independent readers. The sensitivity and specificity of each AF were calculated. Inter-reader agreement was evaluated using κ statistics. Univariate and multivariate logistic regression analyses were conducted to determine significant AFs. The reduction in positive call rates and csPCa detection rates with combined AF use were calculated and compared with the findings obtained with PI-RADS use alone. The sensitivities and specificities of the AFs indicating csPCa were 72.1%-96.5% and 27.4%-75.2% for reader 1, 66.3%-96.5% and 23.9%-62.0% for reader 2, and 67.4%-96.5% and 34.5%-78.8% for reader 3, with moderate to substantial inter-reader agreement (Fleiss κ, 0.551-0.643). The combined use of two or more AFs for assessing PI-RADS ≥3 lesions resulted in a 19.6%-30.7% reduction in positive calls (p < .05) compared to PI-RADS use alone while preserving the csPCa detection rates (p ≥ .06) for three readers. The use of AFs in combination with PI-RADS can reduce positive calls and false positives without csPCa under-detection.

Performance of Prostate Imaging Reporting and Data System Version 2.1 for Diagnosis of Prostate Cancer: A Systematic Review and Meta-Analysis.

BACKGROUND: The Prostate Imaging Reporting and Data System (PI-RADS) was introduced in 2012 and updated to version 2.1 (v2.1) in early 2019 to improve diagnostic performance and interreader reliability. PURPOSE: To evaluate the diagnostic performance of PI-RADS v2.1 in comparison with v2. METHODS: A systematic review and meta-analysis of the literature was performed using MEDLINE, EMBASE, and Cochrane databases to identify studies evaluating the diagnostic performance of PI-RADS v2.1 for diagnosing clinically significant prostate cancer (csPCa). STUDY TYPE: Systematic review and meta-analysis. SUBJECT: One thousand two hundred forty-eight patients with 1406 lesions from 10 eligible articles. FIELD STRENGTH/SEQUENCE: Conventional MR sequences at 1.5 T and 3 T. ASSESSMENT: Two reviewers independently identified and reviewed the original articles reporting diagnostic performance of PI-RADS v2.1. STATISTICAL TESTS: Meta-analytic summary sensitivity and specificity were calculated using a bivariate random effects model. Meta-analytic sensitivity and specificity between PI-RADS v2 and v2.1 were compared. RESULTS: The pooled sensitivity and specificity of PI-RADS v2.1 were 87% (95% confidence intervals, 82-91%) and 74% (63-82%), respectively. In five studies available for a head-to-head comparison between PI-RADS v2.1 and v2, there were no significant differences in either sensitivity (90% [86-94%] vs. 88% [83-93%], respectively) or specificity (76% [59-93%] vs. 61% [39-83%], respectively; P = 0.37). The sensitivity and specificity were 81% (73-87%) and 82% (68-91%), respectively, for a PI-RADS score cutoff of ≥4, and 94% (88-97%) and 56% (35-97%) for ≥3. Regarding the zonal location, the sensitivity and specificity for the transitional zone only were 90% (84-96%) and 76% (62-90%) respectively, whereas for the whole gland they were 85% (79-91%) and 71% (57-85%). DATA CONCLUSION: PI-RADS v2.1 demonstrated good overall performance for the diagnosis of csPCa. PI-RADS v2.1 tended to show higher specificity than v2, but the difference lacked statistical significance. LEVEL OF EVIDENCE: 3 TECHNICAL EFFICACY STAGE: 3.

Preoperative MRI-based estimation of risk for positive resection margin after radical prostatectomy in patients with prostate cancer: development and validation of a simple scoring system.

OBJECTIVES: To develop a simplified MRI-based model to predict the risk for positive surgical margins (PSMs) after radical prostatectomy (RP) in patients with prostate cancer (PCa). METHODS: Consecutive patients who underwent RP for PCa were retrospectively identified from a tertiary referral hospital. Patients who underwent RP between January 2014 and June 2014 were assigned as derivation cohort (n = 330) and those between January 2018 and February 2018 were assigned as validation cohort (n = 100). MRI-based predictors associated with PSM were assessed: tumor size, tumor-capsule contact length, the Prostate Imaging Reporting and Data System (PI-RADS) category, tumor location (tumor contact to the apex or posterolateral side near the neurovascular bundle), apical depth, and prostate volume. A prediction model was developed by using multivariable logistic regression, and then it was transformed into a scoring system. The prediction and calibration performance of this scoring system was evaluated using the C statistics and Hosmer-Lemeshow goodness-of-fit test. RESULTS: A total of 121 (36.7%) and 32 (32.0%) of patients in the derivation and validation cohorts had PSMs after RP. The scoring system consisted of the following variables: tumor-capsule contact length, PI-RADS category, tumor located at the apex and/or posterolateral side. This scoring system provided good prediction performance for PSM in the derivation (C statistics, 0.80 [95% CI: 0.76, 0.85]) and validation (C statistics, 0.77 [95% CI: 0.68, 0.87]) cohorts, and also showed good calibration in both cohorts (p = 0.83 and 0.86, respectively). CONCLUSIONS: An MRI-based scoring system can help estimate the risk of PSM after RP. KEY POINTS: • An MRI-based scoring system served as a tool to estimate the risk of positive surgical margin (C statistics, 0.80 and 0.77 in the derivation and validation cohorts, respectively) after radical prostatectomy. • Tumor with contact to the apex or posterolateral aspect, the tumor contact length to capsule, and higher PI-RADS category were independent predictors for the presence of positive resection margins after radical prostatectomy in men with prostate cancer. • High-risk patients as determined by the scoring system demonstrated adverse post-surgical outcomes compared with low- or intermediate-risk patients, in regard to longer length (mean length, 13.0 mm versus 3.9 mm in low risk or 6.2 mm in intermediate risk; p ≤ 0.001) and higher Gleason grade at the margin (grades 4 and 5 in 69.4% and 20.4% versus 16.7% and 16.7% in low risk or 46.7% and 5.4% in intermediate risk; p < 0.001).

Bosniak Classification of Cystic Renal Masses Version 2019: Comparison With Version 2005 for Class Distribution, Diagnostic Performance, and Interreader Agreement Using CT and MRI.

BACKGROUND. The Bosniak classification system for cystic renal masses (CRMs) was updated in 2019, requiring further investigation. OBJECTIVE. The purpose of this study was to compare versions 2005 and 2019 of the Bosniak classification system in terms of class distribution, diagnostic performance, interreader agreement, and intermodality agreement between CT and MRI. METHODS. This retrospective study included 100 patients (mean age, 52.4 ± 11.6 years; 68 men, 32 women) with 104 CRMs (74 malignant) who underwent CT, MRI, and resection between 2010 and 2019. Two radiologists independently evaluated CRMs in separate sessions for each combination of version and modality and assigned a Bosniak class. Diagnostic performance was compared using McNemar tests. Interreader and intermodality agreement were analyzed using weighted kappa coefficients. RESULTS. Across readers and modalities, the proportion of class IIF CRMs was higher for version 2019 than version 2005 (reader 1: 28.8-30.8% vs 6.7-12.5%; reader 2: 26.0-28.8% vs 8.7-19.2%), although 95% CIs overlapped for reader 2 on CT. The proportion of class III CRMs was lower for version 2019 than version 2005 (reader 1: 33.7-35.6% vs 49.0-51.9%; reader 2: 31.7-40.4% vs 37.5-52.9%), although 95% CIs overlapped for all comparisons. Version 2019 showed lower sensitivity for malignancy than version 2005 across readers and modalities (all p < .05); for example, using CT, sensitivity was 75.7% for both readers with version 2019 versus 85.1-87.8% with version 2005. However, version 2019 showed higher specificity than version 2005, which was significant (all p < .05) for reader 1. For example, using CT, specificity was 73.3% (reader 1) and 70.0% (reader 2) with version 2019 versus 50.0% (reader 1) and 56.7% (reader 2) with version 2005. Diagnostic accuracy was not different between versions (version 2005: 76.9-85.6%; version 2019: 74.0-78.8%). Interreader and intermodality agreement were substantial for version 2005 (κ = 0.676-0.782 and 0.711-0.723, respectively) and version 2019 (κ = 0.756-0.804 and 0.704-0.781, respectively). CONCLUSION. Use of version 2019 versus version 2005 of the Bosniak classification system results in a shift in CRM assignment from class III to class IIF. Version 2019 results in lower sensitivity, higher specificity, and similar accuracy versus version 2005. Interreader and intermodality agreement are similar between versions. CLINICAL IMPACT. Version 2019 facilitates recommending imaging surveillance for more CRMs.

Extraprostatic Tumor Extension: Comparison of Preoperative Multiparametric MRI Criteria and Histopathologic Correlation after Radical Prostatectomy.

Background There are no standardized and well-validated criteria for assessing the risk of extraprostatic extension (EPE) of prostate cancer at preoperative multiparametric MRI. Purpose To compare diagnostic performance, intra- and interreader agreement, and correlations of MRI-based criteria for assessment of EPE after radical prostatectomy, including EPE grade, European Society of Urogenital Radiology (ESUR) score, Likert scale, and capsular contact length (CCL). Materials and Methods This retrospective study included consecutive men who underwent MRI and radical prostatectomy between July 2016 and March 2017. Two genitourinary radiologists independently estimated the probability of EPE by using four MRI-based scoring methods. The diagnostic accuracies and intra- and interobserver agreement were evaluated with area under the receiver operating characteristic curve (AUC) and κ statistics, respectively. Correlations between MRI-based score and histologic extent of EPE were analyzed by using the Spearman correlation coefficient (ρ). Results A total of 301 men (mean age ± standard deviation, 65 years ± 7) were evaluated. A total of 129 (42.9%) men had EPE. The AUC ranges of EPE grade, ESUR score, Likert scale, and CCL for assessment of EPE were 0.77-0.81, 0.79-0.81, 0.78-0.79, and 0.78-0.85, respectively, for the two readers. The Likert scale showed lower sensitivity (68.2% [88 of 129] for reader 1, 72.1% [93 of 129] for reader 2) than did EPE grade (77.5% [100 of 129] for reader 1, 79.8% [103 of 129] for reader 2; P ≤ .04). Intra- and interreader agreement were substantial (κ range, 0.61-0.74) for the four methods, with ESUR score showing the lowest values (κ = 0.61 and κ = 0.63, respectively). EPE grade showed highest correlation with histologic extent of EPE (ρ = 0.53 and ρ = 0.55 for circumferential length; ρ = 0.42 and ρ = 0.55 for radial length for readers 1 and 2, respectively; P < .001). Conclusion Extraprostatic extension (EPE) grade, European Society of Urogenital Radiology score, Likert scale, and capsular contact length showed good overall diagnostic performance; however, the EPE grade resulted in more reliable performance and had the highest correlation with histologic EPE extent. © RSNA, 2020 Online supplemental material is available for this article. See also the editorial by Padhani and Petralia in this issue.

Interreader Agreement with Prostate Imaging Reporting and Data System Version 2 for Prostate Cancer Detection: A Systematic Review and Meta-Analysis.

PURPOSE: We evaluated interreader agreement with PI-RADS® (Prostate Imaging Reporting and Data System) version 2 for detection of prostate cancer. MATERIALS AND METHODS: We searched MEDLINE®, Embase® and the Cochrane Library between 2015 and June 3, 2019 to identify original research reporting interreader agreement in the use of PI-RADS version 2. Quality of the retrieved studies was assessed by 2 independent reviewers using the Guidelines for Reporting Reliability and Agreement Studies. Pooled κ for PI-RADS version 2 was calculated, and a head-to-head comparison with version 1 was performed for the available studies. Subgroup analysis was performed according to zonal anatomy (peripheral or transitional zone), cutoff value (4 or higher, or 3 or higher) and specific imaging sequences (T2-weighted, diffusion-weighted and dynamic contrast enhanced). Meta-regression analysis was performed to assess the cause of study heterogeneity. RESULTS: A total of 30 studies (4,095 patients) were included. Pooled κ of PI-RADS version 2 was 0.61 (95% CI 0.55-0.67). In 4 studies evaluating head-to-head comparisons PI-RADS versions 1 and 2 showed similar pooled κ values (0.61, 95% CI 0.33-0.90 vs 0.68, 95% CI 0.57-0.79; p=0.61). Substantial interreader agreement was noted with a cutoff of 4 or higher (κ=0.61) and moderate agreement was observed with a cutoff of 3 or higher (κ=0.57), peripheral zone (κ=0.64), transitional zone (κ=0.49) and the 3 magnetic resonance imaging sequences (κ 0.42-0.58). Difference in reader experience was the single significant factor affecting study heterogeneity (p=0.01). CONCLUSIONS: PI-RADS version 2 provides substantial interreader agreement in overall scoring in patients with suspicious prostate cancer, with a similar level of agreement to version 1.

Risk Stratification of Prostate Cancer According to PI-RADS® Version 2 Categories: Meta-Analysis for Prospective Studies.

PURPOSE: The Prostate Imaging Reporting and Data System (PI-RADS) version 2 (v2) categorizes the likelihood of clinically significant prostate cancer on magnetic resonance imaging and determines the diagnostic pathway. We determined clinically significant prostate cancer and all prostate cancer detection rates in each PI-RADS v2 category. MATERIALS AND METHODS: MEDLINE®, EMBASE® and Cochrane databases were searched for prospective studies reporting the detection rates of clinically significant prostate cancer or all prostate cancer. Random effects models were used to determine pooled detection rates of clinically significant prostate cancer and all prostate cancer for each PI-RADS category. The risk of bias was assessed with the Quality Assessment of Diagnostic Accuracy Studies-2. Meta-regression analysis was performed to identify factors affecting study heterogeneity. RESULTS: Thirteen prospective studies including 4,265 men who underwent magnetic resonance imaging targeted biopsy and/or systematic biopsy for a PI-RADS v2 category 3 or greater, or systematic biopsy for PI-RADS 1-2 were included. The pooled detection rates of clinically significant prostate cancer monotonically increased for each PI-RADS v2 category, ie 4% (95% CI 2-8) for category 1-2, 17% (95% CI 13-21) for category 3, 46% (95% CI 38-55) for category 4 and 75% (95% CI 73-78) for category 5. Substantial study heterogeneity was noted in clinically significant prostate cancer detection rates for categories 1-2 and 4, which were significantly affected by study subject selection (biopsy naïve patients only or not) and studies with a high risk of bias. CONCLUSIONS: PI-RADS v2 can be useful for the stratification of the risk of clinically significant prostate cancer in patients at risk for prostate cancer but the limitations in category 4 still remain.

Direct Comparison of PI-RADS Version 2 and 2.1 in Transition Zone Lesions for Detection of Prostate Cancer: Preliminary Experience.

BACKGROUND: There appears to be less agreement in the identification of cancers in the transition zone (TZ), which is not as reliable as those in peripheral zone when using the Prostate Imaging Reporting and Data System (PI-RADS) version 2 (v2). In response to such shortcomings, the updated version 2.1 was introduced, which incorporated diffusion-weighted imaging (DWI) into category 2 and clarified lexicons. PURPOSE: To compare the diagnostic performance for the detection of clinically significant TZ prostate cancers (csPCa) and interreader agreement between PI-RADS v2.1 and v2. STUDY TYPE: Retrospective study. POPULATION: In all, 142 patients, 201 TZ lesions. FIELD STRENGTH/SEQUENCE: 3.0T; T2 -weighted image and DWI. ASSESSMENT: Lesions were scored by three independent readers using PI-RADS v2 and v2.1. STATISTICAL TESTS: The sensitivity and specificity at category ≥3 were compared between v2 and v2.1 using the generalized estimating equation model. Detection rates for csPCa of upgraded and downgraded lesions in the use of PI-RADS v2.1 from v2 were assessed. Interreader agreement was assessed using κ statistics. RESULTS: PI-RADS v2.1 showed a higher sensitivity and specificity (94.5% and 60.9%) than v2 (91.8% and 56.3%) for category ≥3 lesions in the detection of csPCa, although not significantly. Of eight upgraded lesions from category 2 to 3 (2 + 1) with an incorporated DWI, 50% (4/8) were csPCa. This was significantly higher than category 2 lesions (4.4%; P = 0.003). No csPCa was detected among the 22.8% (46/201) downgraded lesions. There was a moderate interreader agreement for scores ≥3 (κ = 0.565) in v2.1, which was slightly higher than that for v2 (κ = 0.534), although not significantly. DATA CONCLUSION: PI-RADS v2.1 provides moderate and comparable interreader agreement at category ≥3 than v2 in the TZ lesions. Upgraded lesions from category 2 to 3 demonstrated a higher detection rate of csPCa than category 2 lesions in v2.1. LEVEL OF EVIDENCE: 4 TECHNICAL EFFICACY STAGE: 2 J. Magn. Reson. Imaging 2020;52:577-586.

Radiomics-based prediction model for outcomes of PD-1/PD-L1 immunotherapy in metastatic urothelial carcinoma.

OBJECTIVES: To evaluate the usefulness of a radiomics-based prediction model for predicting response and survival outcomes of patients with metastatic urothelial carcinoma treated with immunotherapy targeting programmed cell death 1 (PD-1) and its ligand (PD-L1). METHODS: Sixty-two patients who underwent immunotherapy were divided into training (n = 41) and validation sets (n = 21). A total of 224 measurable lesions were identified on contrast-enhanced CT. A radiomics signature was constructed with features selected using a least absolute shrinkage and selection operator algorithm in the training set. A radiomics-based model was built based on a radiomics signature consisting of five reliable RFs and the presence of visceral organ involvement using multivariate logistic regression. According to a cutoff determined on the training set, patients in the validation set were assigned to either high- or low-risk groups. Kaplan-Meier analysis was performed to compare progression-free and overall survival between high- and low-risk groups. RESULTS: For predicting objective response and disease control, the areas under the receiver operating characteristic curves of the radiomics-based model were 0.87 (95% CI, 0.65-0.97) and 0.88 (95% CI, 0.67-0.98) for the validation set, providing larger net benefit determined by decision curve analysis than without radiomics-based model. The high-risk group in the validation set showed shorter progression-free and overall survival than the low-risk group (log-rank p = 0.044 and p = 0.035). CONCLUSIONS: The radiomics-based model may predict the response and survival outcome in patients treated with PD-1/PD-L1 immunotherapy for metastatic urothelial carcinoma. This approach may provide important and decision tool for planning immunotherapy. KEY POINTS: • A radiomics-based model was built based on radiomics features and the presence of visceral organ involvement for prediction of outcomes in metastatic urothelial carcinoma treated with immunotherapy. • This prediction model demonstrated good prediction of treatment response and higher net benefit than no model in the independent validation set. • This radiomics-based model demonstrated significant associations with progression-free and overall survival between low-risk and high-risk groups.

Sonographic Features of Small (< 4 cm) Renal Tumors With Low Signal Intensity on T2-Weighted MR Images: Differentiating Minimal-Fat Angiomyolipoma From Renal Cell Carcinoma.

OBJECTIVE: The purpose of this study is to characterize and assess the diagnostic utility of sonographic features of minimal-fat angiomyolipoma (AML) and renal cell carcinoma (RCC) with regard to small (< 4 cm) renal masses with a predominantly low signal intensity (SI) on T2-weighted MR images. MATERIALS AND METHODS: Fifty small renal masses with a predominantly low SI on T2-weighted MR images and no macroscopic fat, all of which had US images available, were assessed. MRI variables (T2 ratio, signal intensity index [SII], and tumor-to-spleen ratio on chemical-shift images), CT features (enhancement patterns and attenuations values on unenhanced images and images obtained in the corticomedullary and nephrographic phases), and sonographic features (echogenicity, heterogeneity, and the presence of acoustic shadowing, a hypoechoic rim, or an intratumoral cyst) were recorded in a blinded manner. Echo-genicity was classified as hypo-, iso-, or hyperechoic compared with the renal parenchyma or markedly hyperchoic when equivalent to that of the renal sinus fat. RESULTS: Minimal-fat AML and RCC were confirmed in 22 and 28 patients, respectively. T2 ratios were significantly lower for minimal-fat AML versus RCCs (p = 0.044). Minimal-fat AMLs exhibited echogenicities that were considered hypoechoic (31.8%), isoechoic (4.5%), hyperechoic (18.2%), or markedly hyperechoic (45.5%). No RCC showed marked hyperechogenicity. CT attenuation values were significantly higher for the minimal-fat AMLs seen in all imaging phases. When the combination of the T2 ratio, nephrographic phase attenuation, and echogenicity was assessed, the AUC value was 0.93 (95% CI, 0.81-0.98), which was a significant increase over the AUC value of 0.83 (95% CI, 0.69-0.92) for noted the combination of the T2 ratio and nephrographic phase attenuation. CONCLUSION: Additional reviews of the echogenicity of small renal masses with low SI on T2-weighted MR images may aid the diagnosis of minimal-fat AML.

Gastric-Type Adenocarcinoma of the Uterine Cervix: Magnetic Resonance Imaging Features, Clinical Outcomes, and Prognostic Factors.

OBJECTIVE: Gastric-type mucinous adenocarcinoma (GAS) of uterine cervix has been suggested as distinct clinicopathologic behavior. Our study aimed to investigate the distinguishing imaging features and clinical outcomes of GAS. METHODS: Fifteen cases of GAS and 45 cases of 3:1 age-matched usual-type endocervical carcinoma (UEA) were assessed. Clinical parameters (age, tumor size, clinical stage, and time to recurrence [TTR] or death) and pathologic results confirmed by surgery were recorded. Imaging features (tumor shape, margin, presence of vaginal involvement, or cyst) on magnetic resonance T2-weighed images were evaluated by consensus reading. Parametrial invasion was assessed by 2 independent readers on a 4-point scale system. RESULTS: Gastric-type mucinous adenocarcinomas were associated with a higher stage at presentation and higher rates of deep cervical stromal invasion, parametrial invasion, and lymphovascular invasion. On magnetic resonance images, GASs also demonstrated a more infiltrative shape (93.3% vs 22.2%; P < 0.001) with/without cysts, ill-defined margin (93.3% vs 31.8%; P < 0.001), endocervical location (46.7%) or extension to the entire cervix (46.7%), higher rates of vaginal involvement (40.0% vs 4.4%; P = 0.002), and higher grade of parametrial invasion by both readers (66.7% vs 22.2%; P = 0.003) compared with UEAs. Patients with GAS had a significantly shorter TTR compared with UEA cases (P < 0.001) by log-rank test. A multivariate Cox proportional hazards model revealed that an infiltrative tumor shape (hazard ratio, 5.92; P = 0.004) and stage II or greater (hazard ratio, 4.19; P = 0.011) were prognostic indicators of a shorter TTR. CONCLUSIONS: Gastric-type mucinous adenocarcinoma may have characteristic imaging features and poorer outcomes compared with UEA.

Pre- and Posttreatment Glioma: Comparison of Amide Proton Transfer Imaging with MR Spectroscopy for Biomarkers of Tumor Proliferation.

PURPOSE: To correlate and compare diagnostic performance with amide proton transfer (APT) imaging as a tumor proliferation index with that with magnetic resonance (MR) spectroscopy in subgroups of patients with pre- and posttreatment glioma. MATERIALS AND METHODS: This retrospective study was approved by the institutional review board. In 40 patients with pretreatment glioma and 25 patients with posttreatment glioma, correlation between APT asymmetry and the choline-to-creatine and choline-to-N-acetylaspartate ratios in corresponding voxels of interest was determined, and the 90% histogram cutoff of APT asymmetry values (APT90) for the entire solid portion of gliomas was calculated for diagnostic performance. Area under the receiver operating characteristic curve (AUC), leave-one-out cross validation, and intraclass correlation coefficients were analyzed. RESULTS: The APT asymmetry values showed a moderate correlation (r = 0.49, P < .001) with the choline-to-creatine ratios and a mild correlation with the choline-to-N-acetyl-aspartate ratios (r = 0.32, P = .011) in the corresponding lesions. The APT90 showed comparable diagnostic accuracy for grading of gliomas (AUC, 0.81-0.84 vs 0.86; P = .582-.864) and superior accuracy for differentiation of tumor progression from treatment-related change (AUC, 0.89-0.90 vs 0.60; P = .031-.046) compared with those with MR spectroscopy. The cross-validated area under the curve and accuracy of the APT90 in posttreatment gliomas were 0.89-0.90 and 72%, respectively. The interreader agreement for APT90 was excellent in both pretreatment and posttreatment gliomas (intraclass correlation coefficient, 0.95 and 0.96, respectively). CONCLUSION: APT imaging used as a tumor proliferation index showed moderate correlation with MR spectroscopic values and is a superior imaging method to MR spectroscopy, particularly for assessment of posttreatment gliomas.

Added value of amide proton transfer imaging to conventional and perfusion MR imaging for evaluating the treatment response of newly diagnosed glioblastoma.

OBJECTIVES: To determine the added value of amide proton transfer (APT) imaging to conventional and perfusion MRI for differentiating tumour progression (TP) from the treatment-related effect (TE) in patients with post-treatment glioblastomas. METHODS: Sixty-five consecutive patients with enlarging contrast-enhancing lesions following concurrent chemoradiotherapy were assessed using contrast-enhanced T1-weighted MRI (CE-T1WI), 90th percentile histogram parameters of normalized cerebral blood volume (nCBV90) and APT asymmetry value (APT90). Diagnostic performance was determined using the area under the receiver operating characteristic curve (AUC) and cross validations. RESULTS: There were statistically significant differences in the mean APT90 between the TP and the TE groups (3.87-4.01 % vs. 1.38-1.41 %; P < .001). Compared with CE-T1WI alone, the addition of APT90 to CE-T1WI significantly improved cross-validated AUC from 0.58-0.74 to 0.89-0.91 for differentiating TP from TE. The combination of CE-T1WI, nCBV90 and APT90 resulted in greater diagnostic accuracy for differentiating TP from TE than the combination of CE-T1WI and nCBV90 (cross-validated AUC, 0.95-0.97 vs. 0.84-0.91). The inter-reader agreement between the expert and trainee was excellent for the measurements of APT90 (intraclass correlation coefficient, 0.94). CONCLUSION: Adding APT imaging to conventional and perfusion MRI improves the diagnostic performance for differentiating TP from TE. KEY POINTS: • APT imaging could provide a reliable distinction between TP and TE • Adding APT imaging to CE-T1WI improved the diagnostic accuracy versus CE-T1WI alone • Multimodal imaging using CE-T1WI, perfusion and APT imaging led to accurate diagnosis • The inter-reader agreement of APT histogram parameters was excellent.

Multi-Contrast High-Resolution Magnetic Resonance Findings of Spontaneous and Unruptured Intracranial Vertebral Artery Dissection: Qualitative and Quantitative Analysis According to Stages.

BACKGROUND: Although high-resolution magnetic resonance imaging (HR-MRI) has been used as a strong imaging method for diagnosing intracranial vertebral artery dissection (IVAD), the diagnosis is sometimes challenging because a dissection has geometric changes in the natural course. The radiologic features may change or disappear over time, which makes the diagnosis confusing. Our study was to present radiological findings according to the stages in spontaneous and unruptured, IVAD on 3T HR-MRI and to guide the age estimation of IVAD with the distinguishing findings according to the stages. METHODS: From January 2011 to July 2014, the 41 vertebral arteries (M:F = 18:12; age range 32-67 years) were retrospectively enrolled. Spontaneous, unruptured IVAD was diagnosed if it had a clear onset based on clinical and radiological findings. The stages were classified as acute (0-3 days), early subacute (3-10 days), late subacute (10-60 days) and chronic stage (>60 days; recovery and non-recovery groups) according to the time intervals from symptom onset, based on the prior published studies. HR-MR findings were assessed and compared in the intimal flap, double lumen, aneurysmal dilatation (maximal outer diameter, maximal wall thickness, wall thickness index and remodeling index), intramural hematoma (relative signal intensity) and vessel wall enhancement according to the stages with qualitative and quantitative methods. Two radiologists analyzed the HR-MR findings with consensus reading. RESULTS: IVAD was classified into acute (n = 6), early subacute (n = 8), late subacute (n = 16) and chronic (n = 11) stages. HR-MR dissection findings such as intimal flap, double lumen, aneurysmal dilatation and intramural hematoma significantly decreased from the earlier stages to the chronic stage (p < 0.05). The quantitative indices in aneurysmal dilatation and the relative signal intensity of intramural hematoma showed significant higher values in the earlier stages followed by a significant decrease in the chronic stage recovery group (p < 0.05). The degree of vessel wall enhancement was higher in the earlier stage and decreased in the chronic stage (p < 0.05), but mild vessel wall enhancement was identified 900 days after symptom onset. CONCLUSION: The 3T HR-MRI reveals the vessel wall characteristics and provides distinguishing findings between earlier stages and the chronic stage in spontaneous and unruptured IVAD. Characterization of these radiological findings according to stages may assist with the age estimation of the dissection and may help to understand IVAD as a whole.

Virtual stenting of intracranial aneurysms: application of hemodynamic modification analysis.

BACKGROUND: Practical application of hemodynamic modification analysis based on computational fluid dynamics (CFD) in intracranial aneurysms is still under study. PURPOSE: To determine the clinical applicability of virtual stenting of aneurysms by comparing the simulated results with clinical outcome of real stenting. MATERIAL AND METHODS: Three-dimensional (3D) digital subtraction angiography (DSA) images were imported to a dedicated integrated prototypic CFD platform (Siemens Healthcare GmbH) which allows all necessary steps of 3D models for CFD analysis. The results of CFD simulation with virtual implantation of a stent can be visualized in the same platform for qualitative comparisons on a color-coded volume visualization window. Five small intracranial aneurysms with and without virtual stenting were analyzed and assessed on a qualitative level. Expert rating were performed for evaluating the simulated results, and comparing those to the long-term follow-up outcomes of real stenting. RESULTS: CFD simulation after virtual stenting was feasible in five differently located aneurysms and corresponded to the long-term changes of stented aneurysms by showing alteration in flow pattern. There was no significant difference (P = 0.5) between the simulated hemodynamic changes after virtual stenting and the angiographic changes after stenting in four aneurysms except one. There was good agreement regarding the assessment of the changes by two raters (kappa = 0.657). CONCLUSION: CFD analysis using patient-specific virtual stenting of the CFD platform may be used as a simple and less time-consuming test tool predicting the involution of aneurysms after stent placement by analyzing the vector visualization of the flow changes.