Large Language Model Use in Radiology Residency Applications: Unwelcomed but Inevitable.

OBJECTIVE: This study explores radiology program directors' perspectives on the impact of large language model (LLM) use among residency applicants to craft personal statements. METHODS: Eight program directors from the Radiology Residency Education Research Alliance (RRERA) participated in a mixed-methods study, which included a survey regarding impressions of AI-generated personal statements and focus group discussions (July 2023). Each director reviewed four personal statement variations for five applicants, blinded to author type: the original and three ChatGPT-4.0 versions generated with varying prompts, aggregated for analysis. A 5-point Likert scale surveyed the writing quality, including voice, clarity, engagement, organization, and the perceived origin of each statement. An experienced qualitative researcher facilitated focus group discussions. Data analysis was performed using a rapid analytic approach with a coding template capturing key areas related to residency applications. RESULTS: GPT-generated statement (GPT) ratings were more often average or worse in quality (56%, 268/475) than ratings of human-authored statements (Hu) (29% [45/160]). Although reviewers were not confident in their ability to distinguish the origin of personal statements, they did so reliably and consistently, identifying the human-authored personal statements at 95% (38/40) as probably or definitely original. Focus group discussions highlighted the inevitable use of AI in crafting personal statements and concerns about its impact on the authenticity and the value of the personal statement in residency selections. Program directors were divided on the appropriate use and regulation of AI. DISCUSSION: Radiology residency program directors rated LLM-generated personal statements as lower in quality and expressed concern about the loss of the applicant's voice but acknowledged the inevitability of increased AI use in the generation of application statements.

Cross-site Validation of AI Segmentation and Harmonization in Breast MRI.

This work aims to perform a cross-site validation of automated segmentation for breast cancers in MRI and to compare the performance to radiologists. A three-dimensional (3D) U-Net was trained to segment cancers in dynamic contrast-enhanced axial MRIs using a large dataset from Site 1 (n = 15,266; 449 malignant and 14,817 benign). Performance was validated on site-specific test data from this and two additional sites, and common publicly available testing data. Four radiologists from each of the three clinical sites provided two-dimensional (2D) segmentations as ground truth. Segmentation performance did not differ between the network and radiologists on the test data from Sites 1 and 2 or the common public data (median Dice score Site 1, network 0.86 vs. radiologist 0.85, n = 114; Site 2, 0.91 vs. 0.91, n = 50; common: 0.93 vs. 0.90). For Site 3, an affine input layer was fine-tuned using segmentation labels, resulting in comparable performance between the network and radiologist (0.88 vs. 0.89, n = 42). Radiologist performance differed on the common test data, and the network numerically outperformed 11 of the 12 radiologists (median Dice: 0.85-0.94, n = 20). In conclusion, a deep network with a novel supervised harmonization technique matches radiologists' performance in MRI tumor segmentation across clinical sites. We make code and weights publicly available to promote reproducible AI in radiology.

Improving Computer-aided Detection for Digital Breast Tomosynthesis by Incorporating Temporal Change.

Purpose To develop a deep learning algorithm that uses temporal information to improve the performance of a previously published framework of cancer lesion detection for digital breast tomosynthesis. Materials and Methods This retrospective study analyzed the current and the 1-year-prior Hologic digital breast tomosynthesis screening examinations from eight different institutions between 2016 and 2020. The dataset contained 973 cancer and 7123 noncancer cases. The front end of this algorithm was an existing deep learning framework that performed single-view lesion detection followed by ipsilateral view matching. For this study, PriorNet was implemented as a cascaded deep learning module that used the additional growth information to refine the final probability of malignancy. Data from seven of the eight sites were used for training and validation, while the eighth site was reserved for external testing. Model performance was evaluated using localization receiver operating characteristic curves. Results On the validation set, PriorNet showed an area under the receiver operating characteristic curve (AUC) of 0.931 (95% CI: 0.930, 0.931), which outperformed both baseline models using single-view detection (AUC, 0.892 [95% CI: 0.891, 0.892]; P < .001) and ipsilateral matching (AUC, 0.915 [95% CI: 0.914, 0.915]; P < .001). On the external test set, PriorNet achieved an AUC of 0.896 (95% CI: 0.885, 0.896), outperforming both baselines (AUC, 0.846 [95% CI: 0.846, 0.847]; P < .001 and AUC, 0.865 [95% CI: 0.865, 0.866]; P < .001, respectively). In the high sensitivity range of 0.9 to 1.0, the partial AUC of PriorNet was significantly higher (P < .001) relative to both baselines. Conclusion PriorNet using temporal information further improved the breast cancer detection performance of an existing digital breast tomosynthesis cancer detection framework. Keywords: Digital Breast Tomosynthesis, Computer-aided Detection, Breast Cancer, Deep Learning © RSNA, 2024 See also commentary by Lee in this issue.

Disease-specific survival outcomes for patients after locoregional treatment for ductal carcinoma in situ: observational cohort study.

BACKGROUND: Breast-conserving surgery alone, breast-conserving surgery with adjuvant radiation treatment, and mastectomy are guideline-concordant treatments for ductal carcinoma in situ. The aim of this study was to compare survival outcomes between these treatment options. METHODS: A stratified random sample of patients diagnosed with pure ductal carcinoma in situ between 2008 and 2014 was selected from 1330 sites in the USA. Data on diagnosis, treatment, and follow-up were abstracted by local cancer registrars. Population-averaged marginal estimates of disease-specific survival and overall survival for breast-conserving surgery alone, breast-conserving surgery with radiation treatment, and mastectomy were obtained by combining sampling and overlap weights. RESULTS: A total of 18 442 women were included, with a median follow-up of 67.8 (interquartile range 46.1-93.5) months. A total of 35 women died from breast cancer, at a median age of 62 (interquartile range 50-74) years. Population-averaged 8-year rates of disease-specific survival were 99.6% or higher for all treatment groups, with no significant differences between groups (breast-conserving surgery alone versus breast-conserving surgery with radiation treatment, HR 1.19 (95% c.i. 0.29 to 4.85); and mastectomy versus breast-conserving surgery with radiation treatment, HR 1.74 (95% c.i. 0.53 to 5.72). There was no difference in overall survival between the patients who underwent a mastectomy and the patients who underwent breast-conserving surgery with radiation treatment (HR 1.09 (95% c.i. 0.83 to 1.43)). Patients who underwent breast-conserving surgery alone had lower overall survival compared with the patients who underwent breast-conserving surgery with radiation treatment (HR 1.29 (95% c.i. 1.00 to 1.67)). This survival difference vanished for all but one subgroup, namely patients less than 65 years (HR 1.86 (95% c.i. 1.15 to 3.00)). CONCLUSION: There was no statistically significant difference in disease-specific survival between women operated with breast-conserving surgery alone, breast-conserving surgery with radiation treatment, or mastectomy for ductal carcinoma in situ. Given the low absolute risk of disease-specific mortality, these results provide confidence in offering individualized locoregional treatment without fear of compromising survival.

Survey on Current Utilization and Perception of Synthesized Mammography.

OBJECTIVE: To assess utilization and perceptions of 2D synthesized mammography (SM) for digital breast tomosynthesis (DBT) among practicing U.S. breast radiologists. METHODS: An IRB-exempt 23-question anonymized survey was developed by the Society of Breast Imaging (SBI) Patient Care and Delivery Committee and emailed to practicing U.S. radiologist SBI members on October 9, 2023. Questions assessed respondents' demographics, current mammographic screening protocol, confidence interpreting SM for mammographic findings, and perceived advantages and disadvantages of SM. RESULTS: Response rate was 13.4% (371/2771). Of 371 respondents, 208 were currently screening with DBT/SM (56.1%), 98 with DBT/SM/digital mammography (DM) (26.4%), 61 with DBT/DM (16.4%), and 4 with DM (1.1%). Most respondents felt confident using DBT/SM to evaluate masses (254/319, 79.6%), asymmetries (247/319, 77.4%), and distortions (265/318, 83.3%); however, confidence was mixed for calcifications (agreement 130/320, 40.6%; disagreement 156/320, 48.8%; neutral 34/320, 10.6%). The most frequently cited disadvantage and advantage of SM were reconstruction algorithm false-positive results (199/347, 57.4%) and lower radiation dose (281/346, 81.2%), respectively. Higher confidence and fewer disadvantages were reported by radiologists who had more SM experience, screened with DBT/SM, or exclusively used Hologic vendor (all P <.05). CONCLUSION: For most survey respondents (56.1%), SM has replaced DM in DBT screening. Radiologists currently screening with DBT/SM or with more SM experience reported greater confidence in SM with fewer perceived disadvantages.

Radiology for Ductal Carcinoma In Situ of the Breast: Updates on Invasive Cancer Progression and Active Monitoring.

Ductal carcinoma in situ (DCIS) accounts for approximately 30% of new breast cancer diagnoses. However, our understanding of how normal breast tissue evolves into DCIS and invasive cancers remains insufficient. Further, conclusions regarding the mechanisms of disease progression in terms of histopathology, genetics, and radiology are often conflicting and have implications for treatment planning. Moreover, the increase in DCIS diagnoses since the adoption of organized breast cancer screening programs has raised concerns about overdiagnosis and subsequent overtreatment. Active monitoring, a nonsurgical management strategy for DCIS, avoids surgery in favor of close imaging follow-up to de-escalate therapy and provides more treatment options. However, the two major challenges in active monitoring are identifying occult invasive cancer and patients at risk of invasive cancer progression. Subsequently, four prospective active monitoring trials are ongoing to determine the feasibility of active monitoring and refine the patient eligibility criteria and follow-up intervals. Radiologists play a major role in determining eligibility for active monitoring and reviewing surveillance images for disease progression. Trial results published over the next few years would support a new era of multidisciplinary DCIS care.

Adjectives in the Summary Paragraph of the MSPE: Are They Telling Us Something?

RATIONALE AND OBJECTIVES: Many US medical schools do not provide a summative, comparative assessment of students (class rank), instead utilizing descriptive adjectives in the summary paragraph of the Medical Student Performance Evaluation (MSPE). The objective of this study was to determine whether those adjectives correlate with student performance. MATERIALS AND METHODS: Applications from all US allopathic medical schools received by a single diagnostic radiology program in the 2023-24 cycle were reviewed. The final adjectives from schools that rank students were coded as positive, negative, or neutral. For students from non-ranking schools, descriptive adjectives from the MSPE summary paragraphs were extracted and categorized based on the final adjective coding schema along with a library of commonly used positive adjectives. The frequency of adjectives was correlated with applicant academic and demographic factors using multivariable logistic regression. RESULTS: Applications from 97% (147/151) of US allopathic medical schools were received. 60.5% (89/147) of schools rank their students with 27.9% (41/147) using a final adjective coded to performance tier. Of the 58 non-ranking schools, 56.9% (33/58) used descriptive adjectives in the MSPE's summary paragraph. There was no association with academic performance metrics and either generally positive adjectives or coded descriptive adjectives. There was a greater association with positive descriptive adjectives for non-white applicants (p = 0.011) and generally positive adjectives for higher-ranked schools (p = 0.004). CONCLUSION: Undefined descriptive adjectives in the MSPE's summary paragraph, when benchmarked to final adjectives, do not correlate with academic performance, and may instead be used for student advocacy by medical school deans.

Patient Characteristics Impact Performance of AI Algorithm in Interpreting Negative Screening Digital Breast Tomosynthesis Studies.

Background Artificial intelligence (AI) is increasingly used to manage radiologists' workloads. The impact of patient characteristics on AI performance has not been well studied. Purpose To understand the impact of patient characteristics (race and ethnicity, age, and breast density) on the performance of an AI algorithm interpreting negative screening digital breast tomosynthesis (DBT) examinations. Materials and Methods This retrospective cohort study identified negative screening DBT examinations from an academic institution from January 1, 2016, to December 31, 2019. All examinations had 2 years of follow-up without a diagnosis of atypia or breast malignancy and were therefore considered true negatives. A subset of unique patients was randomly selected to provide a broad distribution of race and ethnicity. DBT studies in this final cohort were interpreted by a U.S. Food and Drug Administration-approved AI algorithm, which generated case scores (malignancy certainty) and risk scores (1-year subsequent malignancy risk) for each mammogram. Positive examinations were classified based on vendor-provided thresholds for both scores. Multivariable logistic regression was used to understand relationships between the scores and patient characteristics. Results A total of 4855 patients (median age, 54 years [IQR, 46-63 years]) were included: 27% (1316 of 4855) White, 26% (1261 of 4855) Black, 28% (1351 of 4855) Asian, and 19% (927 of 4855) Hispanic patients. False-positive case scores were significantly more likely in Black patients (odds ratio [OR] = 1.5 [95% CI: 1.2, 1.8]) and less likely in Asian patients (OR = 0.7 [95% CI: 0.5, 0.9]) compared with White patients, and more likely in older patients (71-80 years; OR = 1.9 [95% CI: 1.5, 2.5]) and less likely in younger patients (41-50 years; OR = 0.6 [95% CI: 0.5, 0.7]) compared with patients aged 51-60 years. False-positive risk scores were more likely in Black patients (OR = 1.5 [95% CI: 1.0, 2.0]), patients aged 61-70 years (OR = 3.5 [95% CI: 2.4, 5.1]), and patients with extremely dense breasts (OR = 2.8 [95% CI: 1.3, 5.8]) compared with White patients, patients aged 51-60 years, and patients with fatty density breasts, respectively. Conclusion Patient characteristics influenced the case and risk scores of a Food and Drug Administration-approved AI algorithm analyzing negative screening DBT examinations. © RSNA, 2024.

Screening the Implant-Augmented Breast with Digital Breast Tomosynthesis: Is Tomosynthesis Necessary for Non-implant-Displaced Views?

OBJECTIVE: To determine cancer visualization utility and radiation dose for non-implant-displaced (ID) views using standard protocol with digital breast tomosynthesis (DBT) vs alternative protocol with 2D only when screening women with implant augmentation. METHODS: This retrospective cohort study identified women with implants who underwent screening DBT examinations that had abnormal findings from July 28, 2014, to December 31, 2021. Three fellowship-trained breast radiologists independently reviewed examinations retrospectively to determine if the initially identified abnormalities could be visualized on standard protocol (DBT with synthesized 2D (S2D) for ID and non-ID views) and alternate protocol (DBT with S2D for ID and only the S2D images for non-ID views). Estimated exam average glandular dose (AGD) and associations between cancer visualization with patient and implant characteristics for both protocols were evaluated. RESULTS: The study included 195 patients (mean age 55 years ± 10) with 223 abnormal findings. Subsequent biopsy was performed for 86 abnormalities: 59 (69%) benign, 8 (9%) high risk, and 19 (22%) malignant. There was no significant difference in malignancy visualization rate between standard (19/223, 8.5%) and alternate (18/223, 8.1%) protocols (P = .92), but inclusion of the DBT for non-ID views found one additional malignancy. Total examination AGD using standard protocol (21.9 mGy ± 5.0) was significantly higher than it would be for estimated alternate protocol (12.6 mGy ± 5.0, P <.001). This remained true when stratified by breast thickness: 6.0-7.9 cm, 8.0-9.9 cm, >10.0 cm (all P <.001). CONCLUSION: The inclusion of DBT for non-ID views did not significantly increase the cancer visualization rate but did significantly increase overall examination AGD.

Women in leadership matters: Achieving diversity in radiology residency recruitment.

RATIONALE AND OBJECTIVES: To assess whether academic radiology departments and residency programs with efforts toward supporting and augmenting Diversity, Equity, and Inclusion (DEI) are associated with a higher proportion of residents from diverse backgrounds. MATERIALS AND METHODS: Program Directors within the Radiology Residency Education Research Alliance were surveyed to gather information about program characteristics, incorporation of diversity in resident recruitment, the sponsoring department's commitment to efforts at expanding diversity, and a summary of their current and past residents, staff and faculty members (academic years 2020 and 2023) with respect to a list of diversity characteristics. RESULTS: Survey response rate was 51 %. Sixty-three percent (15/24) of participating programs have departmental committees dedicated to DEI work; 46 % (11/24) of programs' departments have a Vice Chair for DEI. Sixty percent (15/24) of programs use their social media accounts to advertise their DEI programming efforts. Ninety-six percent (23/24) of programs participating in the survey use diversity factors to select candidates for their program. Women Leadership was associated with above-median diversity of residents and faculty. CONCLUSION: This study of radiology residency programs encourages a more prominent role for women in leadership positions within academic radiology departments to drive diversity and inclusion efforts.

Otolaryngology residency program factors associated with female resident representation.

Objective: Female representation in the field of otolaryngology is lacking. Residency is the first point at which medical school graduates specialize in a chosen field and thus represents an opportunity to recruit and train more female otolaryngologists. This study sought to identify program factors associated with greater female representation among resident physicians. Methods: Departmental websites of all 124 otolaryngology residency programs in the United States and Puerto Rico were examined for a list of residents. For programs with a resident roster available, the genders of residents, faculty, program directors, and chairpersons were recorded. Location and city population for each program was also recorded, as was female resident representation. Programs were compared using Pearson Chi-squared univariate tests. Results: 1,632 residents and 2,605 faculty were included in the analysis of 109 programs. The median female resident representation was 40%. Programs with larger faculty sizes, more female faculty, and urban location were associated with an above-median female resident representation. Programs with a larger residency cohort approached significance regarding above-median female resident representation. Higher female faculty representation, program director gender, chairperson gender, and US region were not associated with variation in female resident representation. Conclusions: Greater female otolaryngology residency representation was associated with programs having an urban location and greater numbers of female and total faculty. It was also likely that a larger resident cohort size may affect female resident representation. The proportions of female faculty, program director, and chairperson gender, as well as the US region, were not associated with variation in female resident gender representation.

A publicly available deep learning model and dataset for segmentation of breast, fibroglandular tissue, and vessels in breast MRI.

Breast density, or the amount of fibroglandular tissue (FGT) relative to the overall breast volume, increases the risk of developing breast cancer. Although previous studies have utilized deep learning to assess breast density, the limited public availability of data and quantitative tools hinders the development of better assessment tools. Our objective was to (1) create and share a large dataset of pixel-wise annotations according to well-defined criteria, and (2) develop, evaluate, and share an automated segmentation method for breast, FGT, and blood vessels using convolutional neural networks. We used the Duke Breast Cancer MRI dataset to randomly select 100 MRI studies and manually annotated the breast, FGT, and blood vessels for each study. Model performance was evaluated using the Dice similarity coefficient (DSC). The model achieved DSC values of 0.92 for breast, 0.86 for FGT, and 0.65 for blood vessels on the test set. The correlation between our model's predicted breast density and the manually generated masks was 0.95. The correlation between the predicted breast density and qualitative radiologist assessment was 0.75. Our automated models can accurately segment breast, FGT, and blood vessels using pre-contrast breast MRI data. The data and the models were made publicly available.

Classification performance bias between training and test sets in a limited mammography dataset.

OBJECTIVES: To assess the performance bias caused by sampling data into training and test sets in a mammography radiomics study. METHODS: Mammograms from 700 women were used to study upstaging of ductal carcinoma in situ. The dataset was repeatedly shuffled and split into training (n = 400) and test cases (n = 300) forty times. For each split, cross-validation was used for training, followed by an assessment of the test set. Logistic regression with regularization and support vector machine were used as the machine learning classifiers. For each split and classifier type, multiple models were created based on radiomics and/or clinical features. RESULTS: Area under the curve (AUC) performances varied considerably across the different data splits (e.g., radiomics regression model: train 0.58-0.70, test 0.59-0.73). Performances for regression models showed a tradeoff where better training led to worse testing and vice versa. Cross-validation over all cases reduced this variability, but required samples of 500+ cases to yield representative estimates of performance. CONCLUSIONS: In medical imaging, clinical datasets are often limited to relatively small size. Models built from different training sets may not be representative of the whole dataset. Depending on the selected data split and model, performance bias could lead to inappropriate conclusions that might influence the clinical significance of the findings. ADVANCES IN KNOWLEDGE: Performance bias can result from model testing when using limited datasets. Optimal strategies for test set selection should be developed to ensure study conclusions are appropriate.

Current Practice and Variation in Same-Day Services in Breast Imaging: A Multi-Institutional National Survey of the Society of Breast Imaging Membership.

OBJECTIVE: The availability of same-day services in breast imaging is an important topic given potential advantages for timely diagnoses and patient experiences, but there are potential barriers that lead facilities to not offer these services. We sought to understand current practice patterns and radiologist perspectives on offering same-day services. METHODS: The Society of Breast Imaging (SBI) Patient Care & Delivery Committee developed a 19-question survey that was emailed to all 3449 active members of the SBI in May 2023. An exemption from the institutional review board was obtained at the lead author's institution. The survey consisted of 19 questions that were designed to understand the scope, perceptions, barriers, and logistics of same-day services. Comparisons were made between responses for offering same-day services (screening interpretation, diagnostic examinations, biopsies) and respondent demographics. RESULTS: A total of 437 American and Canadian members participated, yielding a response rate of 12.7%. Respondents were most commonly in private practice (43.0%, 188/437), working in an outpatient medical center-based clinic (41.9%, 183/437), and without trainees (64.5%, 282/437). Respondents estimated 12.1% of screening examinations were interpreted while patients waited, which was significantly more common in free-standing breast imaging clinics (P = .028) and practices without trainees (P = .036). Respondents estimated 15.0% of diagnostic examinations were performed same day, which was more common in academic and private practices (P = .03) and practices without trainees (P = .01). Respondents estimated 11.5% of biopsies were performed the same day as the recommendation, which had no association with practice type/context, presence of trainees, number of mammography units, number of radiologists, or number of technologists. Long patient travel distance and limited patient mobility were the most cited reasons for offering patients same-day services. CONCLUSION: Offering same-day breast imaging services varies among institutions and may be influenced by factors such as practice context and type and the presence of trainees.

Optimizing the Patient Experience for Women With Disabilities in the Breast Imaging Clinic.

While there are varying opinions on what age to begin and at what interval to perform breast cancer screening, screening mammography is recommended for all women irrespective of disability. Unfortunately, women with disabilities are more likely to present with later-stage disease and higher mortality owing to the barriers for more widespread screening in this population. Women with disabilities may experience challenges accessing breast imaging services, and imaging centers may have suboptimal facilities and staff who are inexperienced in caring for this population. Efforts to increase accessibility by employing universal design to increase ease of access and provide training to improve the patient experience will go far to improve outcomes for patients with disabilities. To date, there exists no comprehensive guidance on how to improve breast cancer screening programs for women with disabilities. The purpose of this paper is to review barriers to screening faced by patients with disabilities, describe strategies to overcome these barriers, and provide guidance for radiologists and referring providers in selecting the best exam for the individual patient.

Factors Contributing to Disproportionate Burnout in Women Breast Imaging Radiologists: A Review.

Physician burnout continues to increase in prevalence and disproportionately affects women physicians. Breast imaging is a woman-dominated subspeciality, and therefore, worsening burnout among women physicians may have significant repercussions on the future of the breast imaging profession. Systemic and organizational factors have been shown to be the greatest contributors to burnout beyond individual factors. Based on the Mayo Model, we review the evidence regarding the 7 major organizational contributors to physician burnout and their potential disproportionate impacts on women breast radiologists. The major organizational factors discussed are work-life integration, control and flexibility, workload and job demands, efficiency and resources, finding meaning in work, social support and community at work, and organizational culture and values. We also propose potential strategies for institutions and practices to mitigate burnout in women breast imaging radiologists. Many of these strategies could also benefit men breast imaging radiologists, who are at risk for burnout as well.