Checklist for Reproducibility of Deep Learning in Medical Imaging.

The application of deep learning (DL) in medicine introduces transformative tools with the potential to enhance prognosis, diagnosis, and treatment planning. However, ensuring transparent documentation is essential for researchers to enhance reproducibility and refine techniques. Our study addresses the unique challenges presented by DL in medical imaging by developing a comprehensive checklist using the Delphi method to enhance reproducibility and reliability in this dynamic field. We compiled a preliminary checklist based on a comprehensive review of existing checklists and relevant literature. A panel of 11 experts in medical imaging and DL assessed these items using Likert scales, with two survey rounds to refine responses and gauge consensus. We also employed the content validity ratio with a cutoff of 0.59 to determine item face and content validity. Round 1 included a 27-item questionnaire, with 12 items demonstrating high consensus for face and content validity that were then left out of round 2. Round 2 involved refining the checklist, resulting in an additional 17 items. In the last round, 3 items were deemed non-essential or infeasible, while 2 newly suggested items received unanimous agreement for inclusion, resulting in a final 26-item DL model reporting checklist derived from the Delphi process. The 26-item checklist facilitates the reproducible reporting of DL tools and enables scientists to replicate the study's results.

Environmental Sustainability and AI in Radiology: A Double-Edged Sword.

According to the World Health Organization, climate change is the single biggest health threat facing humanity. The global health care system, including medical imaging, must manage the health effects of climate change while at the same time addressing the large amount of greenhouse gas (GHG) emissions generated in the delivery of care. Data centers and computational efforts are increasingly large contributors to GHG emissions in radiology. This is due to the explosive increase in big data and artificial intelligence (AI) applications that have resulted in large energy requirements for developing and deploying AI models. However, AI also has the potential to improve environmental sustainability in medical imaging. For example, use of AI can shorten MRI scan times with accelerated acquisition times, improve the scheduling efficiency of scanners, and optimize the use of decision-support tools to reduce low-value imaging. The purpose of this Radiology in Focus article is to discuss this duality at the intersection of environmental sustainability and AI in radiology. Further discussed are strategies and opportunities to decrease AI-related emissions and to leverage AI to improve sustainability in radiology, with a focus on health equity. Co-benefits of these strategies are explored, including lower cost and improved patient outcomes. Finally, knowledge gaps and areas for future research are highlighted.

An Efficient Deep Neural Network to Classify Large 3D Images With Small Objects.

3D imaging enables accurate diagnosis by providing spatial information about organ anatomy. However, using 3D images to train AI models is computationally challenging because they consist of 10x or 100x more pixels than their 2D counterparts. To be trained with high-resolution 3D images, convolutional neural networks resort to downsampling them or projecting them to 2D. We propose an effective alternative, a neural network that enables efficient classification of full-resolution 3D medical images. Compared to off-the-shelf convolutional neural networks, our network, 3D Globally-Aware Multiple Instance Classifier (3D-GMIC), uses 77.98%-90.05% less GPU memory and 91.23%-96.02% less computation. While it is trained only with image-level labels, without segmentation labels, it explains its predictions by providing pixel-level saliency maps. On a dataset collected at NYU Langone Health, including 85,526 patients with full-field 2D mammography (FFDM), synthetic 2D mammography, and 3D mammography, 3D-GMIC achieves an AUC of 0.831 (95% CI: 0.769-0.887) in classifying breasts with malignant findings using 3D mammography. This is comparable to the performance of GMIC on FFDM (0.816, 95% CI: 0.737-0.878) and synthetic 2D (0.826, 95% CI: 0.754-0.884), which demonstrates that 3D-GMIC successfully classified large 3D images despite focusing computation on a smaller percentage of its input compared to GMIC. Therefore, 3D-GMIC identifies and utilizes extremely small regions of interest from 3D images consisting of hundreds of millions of pixels, dramatically reducing associated computational challenges. 3D-GMIC generalizes well to BCS-DBT, an external dataset from Duke University Hospital, achieving an AUC of 0.848 (95% CI: 0.798-0.896).

Screening mammographic performance by race and age in the National Mammography Database: 29,479,665 screening mammograms from 13,181,241 women.

PURPOSE: There are insufficient large-scale studies comparing the performance of screening mammography in women of different races. This study aims to compare the screening performance metrics across racial and age groups in the National Mammography Database (NMD). METHODS: All screening mammograms performed between January 1, 2008, and December 31, 2021, in women aged 30-100 years from 746 mammography facilities in 46 U.S. states in the NMD were included. Patients were stratified by 10-year age intervals and 5 racial groups (African American, American Indian, Asian, White, unknown). Incidence of risk factors (breast density, personal history, family history of breast cancer, age), and time since prior exams were compared. Five screening mammography metrics were calculated: recall rate (RR), cancer detection rate (CDR), positive predictive values for recalls (PPV1), biopsy recommended (PPV2) and biopsy performed (PPV3). RESULTS: 29,479,655 screening mammograms performed in 13,181,241 women between January 1, 2008, and December 31, 2021, from the NMD were analyzed. The overall mean performance metrics were RR 10.00% (95% CI 9.99-10.02), CDR 4.18/1000 (4.16-4.21), PPV1 4.18% (4.16-4.20), PPV2 25.84% (25.72-25.97), PPV3 25.78% (25.66-25.91). With advancing age, RR significantly decreases, while CDR, PPV1, PPV2, and PPV3 significantly increase. Incidence of personal/family history of breast cancer, breast density, age, prior mammogram availability, and time since prior mammogram were mostly similar across all races. Compared to White women, African American women had significantly higher RR, but lower CDR, PPV1, PPV2 and PPV3. CONCLUSIONS: Benefits of screening mammography increase with age, including for women age > 70 and across all races. Screening mammography is effective; with lower RR and higher CDR, PPV2, and PPV3 with advancing age. African American women have poorer outcomes from screening mammography (higher RR and lower CDR), compared to White and all women in the NMD. Racial disparity can be partly explained by higher rate of African American women lost to follow up.

Breast cancer outcomes based on method of detection in community-based breast cancer registry.

PURPOSE: The impact of opportunistic screening mammography in the United States is difficult to quantify, partially due to lack of inclusion regarding method of detection (MOD) in national registries. This study sought to determine the feasibility of MOD collection in a multicenter community registry and to compare outcomes and characteristics of breast cancer based on MOD. METHODS: We conducted a retrospective study of breast cancer patients from a multicenter tumor registry in Missouri from January 2004 - December 2018. Registry data were extracted by certified tumor registrars and included MOD, clinicopathologic information, and treatment. MOD was assigned as screen-detected or clinically detected. Data were analyzed at the patient level. Chi-squared tests were used for categorical variable comparison and Mann-Whitney-U test was used for numerical variable comparison. RESULTS: 5351 women (median age, 63 years; interquartile range, 53-73 years) were included. Screen-detected cancers were smaller than clinically detected cancers (median size 12 mm vs. 25 mm; P < .001) and more likely node-negative (81% vs. 54%; P < .001), lower grade (P < .001), and lower stage (P < .001). Screen-detected cancers were more likely treated with lumpectomy vs. mastectomy (73% vs. 41%; P < .001) and less likely to require chemotherapy (24% vs. 52%; P < .001). Overall survival for patients with invasive breast cancer was higher for screen-detected cancers (89% vs. 74%, P < .0001). CONCLUSION: MOD can be routinely collected and linked to breast cancer outcomes through tumor registries, with demonstration of significant differences in outcome and characteristics of breast cancers based on MOD. Routine inclusion of MOD in US tumor registries would help quantify the impact of opportunistic screening mammography in the US.

Problem-solving Breast MRI.

Breast MRI has high sensitivity and negative predictive value, making it well suited to problem solving when other imaging modalities or physical examinations yield results that are inconclusive for the presence of breast cancer. Indications for problem-solving MRI include equivocal or uncertain imaging findings at mammography and/or US; suspicious nipple discharge or skin changes suspected to represent an abnormality when conventional imaging results are negative for cancer; lesions categorized as Breast Imaging Reporting and Data System 4, which are not amenable to biopsy; and discordant radiologic-pathologic findings after biopsy. MRI should not precede or replace careful diagnostic workup with mammography and US and should not be used when a biopsy can be safely performed. The role of MRI in characterizing calcifications is controversial, and management of calcifications should depend on their mammographic appearance because ductal carcinoma in situ may not appear enhancing on MR images. In addition, ductal carcinoma in situ detected solely with MRI is not associated with a higher likelihood of an upgrade to invasive cancer compared with ductal carcinoma in situ detected with other modalities. MRI for triage of high-risk lesions is a subject of ongoing investigation, with a possible future role for MRI in decreasing excisional biopsies. The accuracy of MRI is likely to increase with the use of advanced techniques such as deep learning, which will likely expand the indications for problem-solving MRI. ©RSNA, 2023 Quiz questions for this article are available in the supplemental material.

Multisite MRI Intravoxel Incoherent Motion Repeatability and Reproducibility across 3 T Scanners in a Breast Diffusion Phantom: A BReast Intravoxel Incoherent Motion Multisite (BRIMM) Study.

BACKGROUND: Monoexponential apparent diffusion coefficient (ADC) and biexponential intravoxel incoherent motion (IVIM) analysis of diffusion-weighted imaging is helpful in the characterization of breast tumors. However, repeatability/reproducibility studies across scanners and across sites are scarce. PURPOSE: To evaluate the repeatability and reproducibility of ADC and IVIM parameters (tissue diffusivity (Dt ), perfusion fraction (Fp ) and pseudo-diffusion (Dp )) within and across sites employing MRI scanners from different vendors utilizing 16-channel breast array coils in a breast diffusion phantom. STUDY TYPE: Phantom repeatability. PHANTOM: A breast phantom containing tubes of different polyvinylpyrrolidone (PVP) concentrations, water, fat, and sponge flow chambers, together with an MR-compatible liquid crystal (LC) thermometer. FIELD STRENGTH/SEQUENCE: Bipolar gradient twice-refocused spin echo sequence and monopolar gradient single spin echo sequence at 3 T. ASSESSMENT: Studies were performed twice in each of two scanners, located at different sites, on each of 2 days, resulting in four studies per scanner. ADCs of the PVP and water were normalized to the vendor-provided calibrated values at the temperature indicated by the LC thermometer for repeatability/reproducibility comparisons. STATISTICAL TESTS: ADC and IVIM repeatability and reproducibility within and across sites were estimated via the within-system coefficient of variation (wCV). Pearson correlation coefficient (r) was also computed between IVIM metrics and flow speed. A P value <0.05 was considered statistically significant. RESULTS: ADC and Dt demonstrated excellent repeatability (<2%; <3%, respectively) and reproducibility (both <5%) at the two sites. Fp and Dp exhibited good repeatability (mean of two sites 3.67% and 5.59%, respectively) and moderate reproducibility (mean of two sites 15.96% and 13.3%, respectively). The mean intersite reproducibility (%) of Fp /Dp /Dt was 50.96/13.68/5.59, respectively. Fp and Dt demonstrated high correlations with flow speed while Dp showed lower correlations. Fp correlations with flow speed were significant at both sites. DATA CONCLUSION: IVIM reproducibility results were promising and similar to ADC, particularly for Dt . The results were reproducible within both sites, and a progressive trend toward reproducibility across sites except for Fp . LEVEL OF EVIDENCE: 2 TECHNICAL EFFICACY: Stage 1.

Current Practices in Anticoagulation Management for Patients Undergoing Percutaneous Image-guided Breast Procedures.

Objective: Given variability in how practices manage patients on antithrombotic medications, we undertook this study to understand the current practice of antithrombotic management for patients undergoing percutaneous breast and axillary procedures. Methods: A 20-item survey with multiple-choice and write-in options was emailed to 2094 active North American members of the Society of Breast Imaging (SBI) in March 2021. Data were collected anonymously and analyzed quantitatively, with free-text responses categorized by themes. Results: Three-hundred twenty-six of 2094 members (15.6%) completed the survey. Eighty-seven percent (274/313) reported having a policy for managing antithrombotic medications. Fifty-nine percent (185/312) reported routinely withholding medications before biopsy, more commonly in the Northeast and South (P = 0.08). Withholding of medications did not vary by lesion location (182/308, 59%, breast vs 181/308, 58.7%, axillary; P = 0.81). Respondents were statistically more likely to withhold medications if using a vacuum-assisted device for all classes of antithrombotic medications (P < 0.001). Up to 50.2% (100/199) on warfarin and 33.6% (66/196) on direct oral anticoagulants had medications withheld more stringently than guidelines suggest. Conclusion: Based on a survey of SBI members, breast imaging practices vary widely in antithrombotic management for image-guided breast and axillary procedures. Of the 60% who withhold antithrombotic medications, a minority comply with recommended withhold guidelines, placing at least some patients at potential risk for thrombotic events. Breast imaging radiologists should weigh the risks and benefits of withholding these medications, and if they elect to withhold should closely follow evidence-based guidelines to minimize the risks of this practice.

PACS-integrated machine learning breast density classifier: clinical validation.

OBJECTIVE: To test the performance of a novel machine learning-based breast density tool. The tool utilizes a convolutional neural network to predict the BI-RADS based density assessment of a study. The clinical density assessments of 33,000 mammographic examinations (164,000 images) from one academic medical center (Site A) were used for training. MATERIALS AND METHODS: This was an IRB approved HIPAA compliant study performed at two academic medical centers. The validation data set was composed of 500 studies from one site (Site A) and 700 from another (Site B). At Site A, each study was assessed by three breast radiologists and the majority (consensus) assessment was used as truth. At Site B, if the tool agreed with the clinical reading, then it was considered to have correctly predicted the clinical reading. In cases where the tool and the clinical reading disagreed, then the study was evaluated by three radiologists and the consensus reading was used as the clinical reading. RESULTS: For the classification into the four categories of the Breast Imaging Reporting and Data System (BI-RADS®), the AI classifier had an accuracy of 84.6% at Site A and 89.7% at Site B. For binary classification (dense vs. non-dense), the AI classifier had an accuracy of 94.4% at Site A and 97.4% at Site B. In no case did the classifier disagree with the consensus reading by more than one category. CONCLUSIONS: The automated breast density tool showed high agreement with radiologists' assessments of breast density.