Automated Breast Density Assessment in MRI Using Deep Learning and Radiomics: Strategies for Reducing Inter-Observer Variability

BACKGROUND

Accurate breast density evaluation allows for more precise risk estimation but suffers from high inter-observer variability.

PURPOSE

To evaluate the feasibility of reducing inter-observer variability of breast density assessment through artificial intelligence (AI) assisted interpretation.

STUDY TYPE

Retrospective.

POPULATION

Six hundred and twenty-one patients without breast prosthesis or reconstructions were randomly divided into training (N = 377), validation (N = 98), and independent test (N = 146) datasets.

FIELD STRENGTH/SEQUENCE

1.5 T and 3.0 T; T1-weighted spectral attenuated inversion recovery.

ASSESSMENT

Five radiologists independently assessed each scan in the independent test set to establish the inter-observer variability baseline and to reach a reference standard. Deep learning and three radiomics models were developed for three classification tasks: (i) four Breast Imaging-Reporting and Data System (BI-RADS) breast composition categories (A-D), (ii) dense (categories C, D) vs. non-dense (categories A, B), and (iii) extremely dense (category D) vs. moderately dense (categories A-C). The models were tested against the reference standard on the independent test set. AI-assisted interpretation was performed by majority voting between the models and each radiologist's assessment.

STATISTICAL TESTS

Inter-observer variability was assessed using linear-weighted kappa (κ) statistics. Kappa statistics, accuracy, and area under the receiver operating characteristic curve (AUC) were used to assess models against reference standard.

RESULTS

In the independent test set, five readers showed an overall substantial agreement on tasks (i) and (ii), but moderate agreement for task (iii). The best-performing model showed substantial agreement with reference standard for tasks (i) and (ii), but moderate agreement for task (iii). With the assistance of the AI models, almost perfect inter-observer variability was obtained for tasks (i) (mean κ = 0.86), (ii) (mean κ = 0.94), and (iii) (mean κ = 0.94).

DATA CONCLUSION

Deep learning and radiomics models have the potential to help reduce inter-observer variability of breast density assessment.

LEVEL OF EVIDENCE

3 TECHNICAL EFFICACY: Stage 1.

Study protocol for a multicenter randomized controlled trial to compare the efficacy of end-ischemic dual hypothermic oxygenated machine perfusion with static cold storage in preventing non-anastomotic biliary strictures after transplantation of liver grafts donated after circulatory death: DHOPE-DCD trial

BACKGROUND

The major concern in liver transplantation of grafts from donation after circulatory death (DCD) donors remains the high incidence of non-anastomotic biliary strictures (NAS). Machine perfusion has been proposed as an alternative strategy for organ preservation which reduces ischemia-reperfusion injury (IRI). Experimental studies have shown that dual hypothermic oxygenated machine perfusion (DHOPE) is associated with less IRI, improved hepatocellular function, and better preserved mitochondrial and endothelial function compared to conventional static cold storage (SCS). Moreover, DHOPE was safely applied with promising results in a recently performed phase-1 study. The aim of the current study is to determine the efficacy of DHOPE in reducing the incidence of NAS after DCD liver transplantation.

METHODS

This is an international multicenter randomized controlled trial. Adult patients (≥18 yrs. old) undergoing transplantation of a DCD donor liver (Maastricht category III) will be randomized between the intervention and control group. In the intervention group, livers will be subjected to two hours of end-ischemic DHOPE after SCS and before implantation. In the control group, livers will be subjected to care as usual with conventional SCS only. Primary outcome is the incidence of symptomatic NAS diagnosed by a blinded adjudication committee. In all patients, magnetic resonance cholangiography will be obtained at six months after transplantation.

DISCUSSION

DHOPE is associated with reduced IRI of the bile ducts. Whether reduced IRI of the bile ducts leads to lower incidence of NAS after DCD liver transplantation can only be examined in a randomized controlled trial.

TRIAL REGISTRATION

The trial was registered in Clinicaltrials.gov in September 2015 with the identifier NCT02584283 .

False positives in PIRADS (V2) 3, 4, and 5 lesions: relationship with reader experience and zonal location

PURPOSE

To investigate the effect of reader experience and zonal location on the occurrence of false positives (FPs) in PIRADS (V2) 3, 4, and 5 lesions on multiparametric (MP)-MRI of the prostate.

MATERIALS AND METHODS

This retrospective study included 139 patients who had consecutively undergone an MP-MRI of the prostate in combination with a transrectal ultrasound MRI fusion-guided biopsy between 2014 and 2017. MRI exams were prospectively read by a group of inexperienced radiologists (cohort 1; 54 patients) and an experienced radiologist (cohort 2; 85 patients). Multivariable logistic regression analysis was performed to determine the association of experience of the radiologist and zonal location with a FP reading. FP rates were compared between readings by inexperienced and experienced radiologists according to zonal location, using Chi-square (χ²) tests.

RESULTS

A total of 168 lesions in 139 patients were detected. Median patient age was 68 years (Interquartile range (IQR) 62.5-73), and median PSA was 10.9 ng/mL (IQR 7.6-15.9) for the entire patient cohort. According to multivariable logistic regression, inexperience of the radiologist was significantly (P = 0.044, odds ratio 1.927, 95% confidence interval [CI] 1.017-3.651) and independently associated with a FP reading, while zonal location was not (P = 0.202, odds ratio 1.444, 95% CI 0.820-2.539). In the transition zone (TZ), the FP rate of the inexperienced radiologists 59% (17/29) was significantly higher (χ²P = 0.033) than that of the experienced radiologist 33% (13/40).

CONCLUSION

Inexperience of the radiologist is significantly and independently associated with a FP reading, while zonal location is not. Inexperienced radiologists have a significantly higher FP rate in the TZ.

89Zr-labeled Bispecific T-cell Engager AMG 211 PET Shows AMG 211 Accumulation in CD3-rich Tissues and Clear, Heterogeneous Tumor Uptake

PURPOSE

Biodistribution of bispecific antibodies in patients is largely unknown. We therefore performed a feasibility study in 9 patients with advanced gastrointestinal adenocarcinomas to explore AMG 211 biodistribution (also known as MEDI-565), an approximately 55 kDa bispecific T-cell engager (BiTE®) directed against carcinoembryonic antigen (CEA) on tumor cells and cluster of differentiation 3 (CD3) on T-cells.

EXPERIMENTAL DESIGN

⁸⁹Zr-labeled AMG 211 as tracer was administered alone or with cold AMG 211, for PET imaging before and/or during AMG 211 treatment.

RESULTS

Before AMG 211 treatment, the optimal imaging dose was 200-μg ⁸⁹Zr-AMG 211 + 1,800-μg cold AMG 211. At 3 hours, the highest blood pool standardized uptake value (SUV)_mean was 4.0, and tracer serum half-life was 3.3 hours. CD3-mediated uptake was clearly observed in CD3-rich lymphoid tissues including spleen and bone marrow (SUV_mean 3.2 and 1.8, respectively), and the SUV_mean decreased more slowly than in other healthy tissues. ⁸⁹Zr-AMG 211 remained intact in plasma and was excreted predominantly via the kidneys in degraded forms. Of 43 visible tumor lesions, 37 were PET quantifiable, with a SUV_max of 4.0 [interquartile range (IQR) 2.7-4.4] at 3 hours using the optimal imaging dose. The tracer uptake differed between tumor lesions 5-fold within and 9-fold between patients. During AMG 211 treatment, tracer was present in the blood pool, whereas tumor lesions were not visualized, possibly reflecting target saturation.

CONCLUSIONS

This first-in-human study shows high, specific ⁸⁹Zr-AMG 211 accumulation in CD3-rich lymphoid tissues, as well as a clear, inter- and intraindividual heterogeneous tumor uptake.

T1 vs. T2 weighted magnetic resonance imaging to assess total kidney volume in patients with autosomal dominant polycystic kidney disease

Purpose

In ADPKD patients total kidney volume (TKV) measurement using MRI is performed to predict rate of disease progression. Historically T1 weighted images (T1) were used, but the methodology of T2 weighted imaging (T2) has evolved. We compared the performance of both sequences.

Methods

40 ADPKD patients underwent an abdominal MRI at baseline and follow-up. TKV was measured by manual tracing with Analyze Direct 11.0 software. Three readers established intra- and interreader coefficients of variation (CV). T1 and T2 measured kidney volumes and growth rates were compared with ICC and Bland–Altman analyses.

Results

Participants were 49.7 ± 7.0 years of age, 55.0% female, with estimated GFR of 50.1 ± 11.5 mL/min/1.73 m². CVs were low and comparable for T2 and T1 (intrareader: 0.83% [0.48–1.79] vs. 1.15% [0.34–1.77], P = 0.9, interreader: 2.18% [1.59–2.61] vs. 1.69% [1.07–3.87], P = 0.9). TKV was clinically similar, but statistically significantly different between T2 and T1: 1867 [1172–2721] vs. 1932 [1180–2551] mL, respectively (P = 0.006), with a bias of only 0.8% and high agreement (ICC 0.997). Percentage kidney growth during 2.2 ± 0.3 years was similar for T2 and T1 (9.3 ± 10.6% vs. 7.8 ± 9.9%, P = 0.1, respectively), with a bias of 1.5% and high agreement (ICC 0.843). T2 was more often of sufficient quality for volume measurement (86.7% vs. 71.1%, P < 0.001).

Conclusions

In patients with ADPKD, measurement of kidney volume and growth rate performs similarly when using T2 compared to T1 weighted images, although T2 performs better on secondary outcome parameters; they are more often of sufficient quality for volume measurement and result in slightly lower intra- and interreader variability.

Electronic supplementary material

The online version of this article (doi:10.1007/s00261-017-1285-2) contains supplementary material, which is available to authorized users.

89Zr-Bevacizumab PET: Potential Early Indicator of Everolimus Efficacy in Patients with Metastatic Renal Cell Carcinoma

Currently, biomarkers that predict the efficacy of everolimus in metastatic renal cell carcinoma (mRCC) patients are lacking. Everolimus inhibits vascular endothelial growth factor A (VEGF-A) expression. We performed PET scans on mRCC patients with ⁸⁹Zr-bevacizumab, a VEGF-A-binding antibody tracer. The aims were to determine a change in tumor tracer uptake after the start of everolimus and to explore whether ⁸⁹Zr-bevacizumab PET can identify patients with early disease progression. Methods:⁸⁹Zr-bevacizumab PET was done before and 2 and 6 wk after the start of everolimus, 10 mg/d, in mRCC patients. Routine CT scans were performed at baseline and every 3 mo thereafter. Tumor tracer uptake was quantified using SUV_max The endpoints were a change in tumor tracer uptake and treatment response on CT after 3 mo. Results: Thirteen patients participated. The median SUV_max of 94 tumor lesions was 7.3 (range, 1.6-59.5). Between patients, median tumor SUV_max varied up to 8-fold. After 2 wk, median SUV_max was 6.3 (1.7-62.3), corresponding to a mean decrease of 9.1% (P < 0.0001). Three patients discontinued everolimus early. At 6 wk, a mean decrease in SUV_max of 23.4% compared with baseline was found in 70 evaluable lesions of 10 patients, with a median SUV_max of 5.4 (1.1-49.4, P < 0.0001). All 10 patients who continued treatment had stable disease at 3 mo. Conclusion: Everolimus decreases ⁸⁹Zr-bevacizumab tumor uptake. Further studies are warranted to evaluate the predictive value of ⁸⁹Zr-bevacizumab PET for everolimus antitumor efficacy.