Test-set results can predict participants' development in breast-screen cancer detection: An observational cohort study.

Background and Aim: Test-sets are standardized assessments used to evaluate reader performance in breast screening. Understanding how test-set results affect real-world performance can help refine their use as a quality improvement tool. The aim of this study is to explore if mammographic test-set results could identify breast-screening readers who improved their cancer detection in association with test-set training. Methods: Test-set results of 41 participants were linked to their annual cancer detection rate change in two periods oriented around their first test-set participation year. Correlation tests and a multiple linear regression model investigated the relationship between each metric in the test-set results and the change in detection rates. Additionally, participants were divided based on their improvement status between the two periods, and Mann-Whitney U test was used to determine if the subgroups differed in their test-set metrics. Results: Test-set records indicated multiple significant correlations with the change in breast cancer detection rate: a moderate positive correlation with sensitivity (0.688, p < 0.001), a moderate negative correlation with specificity (-0.528, p < 0.001), and a low to moderate positive correlation with lesion sensitivity (0.469, p = 0.002), and the number of years screen-reading mammograms (0.365, p = 0.02). In addition, the overall regression was statistically significant (F (2,38) = 18.456 p < 0.001), with an R² of 0.493 (adjusted R² = 0.466) based on sensitivity (F = 27.132, p < 0.001) and specificity (F = 9.78, p = 0.003). Subgrouping the cohort based on the change in cancer detection indicated that the improved group is significantly higher in sensitivity (p < 0.001) and lesion sensitivity (p = 0.02) but lower in specificity (p = 0.003). Conclusion: Sensitivity and specificity are the strongest test-set performance measures to predict the change in breast cancer detection in real-world breast screening settings following test-set participation.

BRAIxDet: Learning to detect malignant breast lesion with incomplete annotations.

Methods to detect malignant lesions from screening mammograms are usually trained with fully annotated datasets, where images are labelled with the localisation and classification of cancerous lesions. However, real-world screening mammogram datasets commonly have a subset that is fully annotated and another subset that is weakly annotated with just the global classification (i.e., without lesion localisation). Given the large size of such datasets, researchers usually face a dilemma with the weakly annotated subset: to not use it or to fully annotate it. The first option will reduce detection accuracy because it does not use the whole dataset, and the second option is too expensive given that the annotation needs to be done by expert radiologists. In this paper, we propose a middle-ground solution for the dilemma, which is to formulate the training as a weakly- and semi-supervised learning problem that we refer to as malignant breast lesion detection with incomplete annotations. To address this problem, our new method comprises two stages, namely: (1) pre-training a multi-view mammogram classifier with weak supervision from the whole dataset, and (2) extending the trained classifier to become a multi-view detector that is trained with semi-supervised student-teacher learning, where the training set contains fully and weakly-annotated mammograms. We provide extensive detection results on two real-world screening mammogram datasets containing incomplete annotations and show that our proposed approach achieves state-of-the-art results in the detection of malignant breast lesions with incomplete annotations.

An Interpretable and Accurate Deep-Learning Diagnosis Framework Modeled With Fully and Semi-Supervised Reciprocal Learning.

The deployment of automated deep-learning classifiers in clinical practice has the potential to streamline the diagnosis process and improve the diagnosis accuracy, but the acceptance of those classifiers relies on both their accuracy and interpretability. In general, accurate deep-learning classifiers provide little model interpretability, while interpretable models do not have competitive classification accuracy. In this paper, we introduce a new deep-learning diagnosis framework, called InterNRL, that is designed to be highly accurate and interpretable. InterNRL consists of a student-teacher framework, where the student model is an interpretable prototype-based classifier (ProtoPNet) and the teacher is an accurate global image classifier (GlobalNet). The two classifiers are mutually optimised with a novel reciprocal learning paradigm in which the student ProtoPNet learns from optimal pseudo labels produced by the teacher GlobalNet, while GlobalNet learns from ProtoPNet's classification performance and pseudo labels. This reciprocal learning paradigm enables InterNRL to be flexibly optimised under both fully- and semi-supervised learning scenarios, reaching state-of-the-art classification performance in both scenarios for the tasks of breast cancer and retinal disease diagnosis. Moreover, relying on weakly-labelled training images, InterNRL also achieves superior breast cancer localisation and brain tumour segmentation results than other competing methods.

Australian women's judgements about using artificial intelligence to read mammograms in breast cancer screening.

Objective: Mammographic screening for breast cancer is an early use case for artificial intelligence (AI) in healthcare. This is an active area of research, mostly focused on the development and evaluation of individual algorithms. A growing normative literature argues that AI systems should reflect human values, but it is unclear what this requires in specific AI implementation scenarios. Our objective was to understand women's values regarding the use of AI to read mammograms in breast cancer screening. Methods: We ran eight online discussion groups with a total of 50 women, focused on their expectations and normative judgements regarding the use of AI in breast screening. Results: Although women were positive about the potential of breast screening AI, they argued strongly that humans must remain as central actors in breast screening systems and consistently expressed high expectations of the performance of breast screening AI. Women expected clear lines of responsibility for decision-making, to be able to contest decisions, and for AI to perform equally well for all programme participants. Women often imagined both that AI might replace radiographers and that AI implementation might allow more women to be screened: screening programmes will need to communicate carefully about these issues. Conclusions: To meet women's expectations, screening programmes should delay implementation until there is strong evidence that the use of AI systems improves screening performance, should ensure that human expertise and responsibility remain central in screening programmes, and should avoid using AI in ways that exacerbate inequities.

ADMANI: Annotated Digital Mammograms and Associated Non-Image Datasets.

Supplemental material is available for this article. Keywords: Mammography, Screening, Convolutional Neural Network (CNN) Published under a CC BY 4.0 license. See also the commentary by Cadrin-Chênevert in this issue.

Utopia versus dystopia: Professional perspectives on the impact of healthcare artificial intelligence on clinical roles and skills.

BACKGROUND: Alongside the promise of improving clinical work, advances in healthcare artificial intelligence (AI) raise concerns about the risk of deskilling clinicians. This purpose of this study is to examine the issue of deskilling from the perspective of diverse group of professional stakeholders with knowledge and/or experiences in the development, deployment and regulation of healthcare AI. METHODS: We conducted qualitative, semi-structured interviews with 72 professionals with AI expertise and/or professional or clinical expertise who were involved in development, deployment and/or regulation of healthcare AI. Data analysis using combined constructivist grounded theory and framework approach was performed concurrently with data collection. FINDINGS: Our analysis showed participants had diverse views on three contentious issues regarding AI and deskilling. The first involved competing views about the proper extent of AI-enabled automation in healthcare work, and which clinical tasks should or should not be automated. We identified a cluster of characteristics of tasks that were considered more suitable for automation. The second involved expectations about the impact of AI on clinical skills, and whether AI-enabled automation would lead to worse or better quality of healthcare. The third tension implicitly contrasted two models of healthcare work: a human-centric model and a technology-centric model. These models assumed different values and priorities for healthcare work and its relationship to AI-enabled automation. CONCLUSION: Our study shows that a diverse group of professional stakeholders involved in healthcare AI development, acquisition, deployment and regulation are attentive to the potential impact of healthcare AI on clinical skills, but have different views about the nature and valence (positive or negative) of this impact. Detailed engagement with different types of professional stakeholders allowed us to identify relevant concepts and values that could guide decisions about AI algorithm development and deployment.

Exploration of potential indicators of burnout, psychological distress and post-traumatic stress disorder, among Australian female first responders.

There is limited research on the psychological wellbeing of female first responders (FRs) and therefore we explore potential indicators of burnout, psychological distress and post-traumatic stress disorder among Australian female FRs. We conducted an online health survey among Australian female FRs (fire, police, paramedical, aeromedical, remote area and other e.g., State Emergency Service). Of the 422 eligible participants who submitted the online survey, 286 completed at least 80% of all survey questions and were used in the final analyses. The main outcomes of interest were moderate burnout (compared to low burnout) and high scores for combined PCL-5/K10 (compared to low scores). Using logistical regression stepwise regression models, we analysed associations between the outcomes of interest and various work-psychosocial factors. Results showed the strongest indicators of moderate burnout to be, 1) returning to work with <12-hour break, 2) exposure to gossip and slander, 3) not enough time to do things, 4) and having experienced rape/sexual assault. The strongest indicators of higher PCL-5/K10 scores were, 1) exposure to unpleasant teasing, 2) pressure at work and home, 3) having experienced physical violence (e.g., beaten/mugged), and 4) someone close to them died unexpectedly. These findings show workforce stressors have more impact on female FRs psychological wellbeing, compared to lifetime traumatic exposures.

Quality use of artificial intelligence in medical imaging: What do radiologists need to know?

The application of artificial intelligence, and in particular machine learning, to the practice of radiology, is already impacting the quality of imaging care. It will increasingly do so in the future. Radiologists need to be aware of factors that govern the quality of these tools at the development, regulatory and clinical implementation stages in order to make judicious decisions about their use in daily practice.

Evaluation of deep learning-based artificial intelligence techniques for breast cancer detection on mammograms: Results from a retrospective study using a BreastScreen Victoria dataset.

INTRODUCTION: This study aims to evaluate deep learning (DL)-based artificial intelligence (AI) techniques for detecting the presence of breast cancer on a digital mammogram image. METHODS: We evaluated several DL-based AI techniques that employ different approaches and backbone DL models and tested the effect on performance of using different data-processing strategies on a set of digital mammographic images with annotations of pathologically proven breast cancer. RESULTS: Our evaluation uses the area under curve (AUC) and accuracy (ACC) for performance measurement. The best evaluation result, based on 349 test cases (930 test images), was an AUC of 0.8979 [95% confidence interval (CI) 0.873, 0.923] and ACC of 0.8178 [95% CI 0.785, 0.850]. This was achieved by an AI technique that utilises a certain family of DL models, namely ResNet, as its backbone, combines the global features extracted from the whole mammogram and the local features extracted from the automatically detected cancer and non-cancer local regions in the whole image, and leverages background cropping and text removal, contrast adjustment and more training data. CONCLUSION: DL-based AI techniques have shown promising results in retrospective studies for many medical image analysis applications. Our study demonstrates a significant opportunity to boost the performance of such techniques applied to breast cancer detection by exploring different types of approaches, backbone DL models and data-processing strategies. The promising results we have obtained suggest further development of AI reading services could transform breast cancer screening in the future.

Factors influencing the recruitment and retention of female first responders: a mixed methods systematic review protocol.

OBJECTIVE: The objective of this review is to produce a set of integrated findings of quantitative and qualitative evidence regarding workplace recruitment and retention factors (including departure) of female first responders to inform recommendations for policy and practice. INTRODUCTION: Historically, first responder workforces such as police officers, firefighters, search and rescue personnel, medical technicians, and paramedics have been largely male dominated. Over the past few decades, however, there has been a steady increase in the number of women entering this field. However, there is minimal research examining factors that influence the recruitment/retention of female first responders. The intention of this review is to identify barriers and facilitators to recruitment and retention of female first responders and to identify any differences between sexes/genders. INCLUSION CRITERIA: This review will summarize qualitative and quantitative research examining factors influencing the recruitment/retention of female first responders. Such factors may include job satisfaction, quality of work life, burnout, compassion fatigue, and intent to remain in the workforce. METHODS: MEDLINE (PubMed), CINAHL (EBSCO), PsycINFO (APA), PTSDpubs (formerly PILOTS; ProQuest), Embase (Elsevier), and Scopus (Elsevier) will be searched for studies published in English from 2009 to the present. Unpublished studies will be searched in Google Scholar, and ProQuest Dissertations and Theses Global. Both quantitative and qualitative studies will be screened for inclusion and critically appraised for methodological quality by two independent reviewers. Both types of data will be extracted using JBI tools for mixed methods systematic reviews. A convergent integrated approach to synthesis and integration will be used. SYSTEMATIC REVIEW REGISTRATION NUMBER: PROSPERO CRD42020156524.

The ethical, legal and social implications of using artificial intelligence systems in breast cancer care.

Breast cancer care is a leading area for development of artificial intelligence (AI), with applications including screening and diagnosis, risk calculation, prognostication and clinical decision-support, management planning, and precision medicine. We review the ethical, legal and social implications of these developments. We consider the values encoded in algorithms, the need to evaluate outcomes, and issues of bias and transferability, data ownership, confidentiality and consent, and legal, moral and professional responsibility. We consider potential effects for patients, including on trust in healthcare, and provide some social science explanations for the apparent rush to implement AI solutions. We conclude by anticipating future directions for AI in breast cancer care. Stakeholders in healthcare AI should acknowledge that their enterprise is an ethical, legal and social challenge, not just a technical challenge. Taking these challenges seriously will require broad engagement, imposition of conditions on implementation, and pre-emptive systems of oversight to ensure that development does not run ahead of evaluation and deliberation. Once artificial intelligence becomes institutionalised, it may be difficult to reverse: a proactive role for government, regulators and professional groups will help ensure introduction in robust research contexts, and the development of a sound evidence base regarding real-world effectiveness. Detailed public discussion is required to consider what kind of AI is acceptable rather than simply accepting what is offered, thus optimising outcomes for health systems, professionals, society and those receiving care.

Improvement of Cancer Detection on Mammograms via BREAST Test Sets.

BACKGROUND: Breast Screen Reader Assessment Strategy (BREAST) is an innovative training and research program for radiologists in Australia and New Zealand. The aim of this study is to evaluate the efficacy of BREAST test sets in improving readers' performance in detecting cancers on mammograms. MATERIALS AND METHODS: Between 2011 and 2018, 50 radiologists (40 fellows, 10 registrars) completed three BREAST test sets and 17 radiologists completed four test sets. Each test set contained 20 biopsy-proven cancer and 40 normal cases. Immediate image-based feedback was available to readers after they completed each test set which allowed the comparison of their selections with the truth. Case specificity, case sensitivity, lesion sensitivity, the Receiver Operating Characteristic (ROC) Area Under the Curve (AUC) and Jackknife Free-Response Receiver Operating Characteristic (JAFROC) Figure of Merit (FOM) were calculated for each reader. Kruskal-Wallis test was utilized to compare scores of the radiologist and registrars across all test-sets whilst Wilcoxon signed rank test was to compare the scores between pairs of test sets. RESULTS: Significant improvements in lesion sensitivity ranging from 21% to 31% were found in radiologists completing later test sets compared to first test set (p ≤ 0.01). Eighty three percent of radiologists achieved higher performance in lesion sensitivity after they completed the first read. Registrars had significantly better scores in the third test set compared to the first set with mean increases of 79% in lesion sensitivity (p = 0.005) and 37% in JAFROC (p = 0.02). Sixty percent and 100% of registrars increased their scores in lesion sensitivity in the second and third reads compared to the first read while the percentage of registrars with higher scores in JAFROC was 80%. CONCLUSION: Introduction of BREAST into national training programs appears to have an important impact in promoting diagnostic efficacy amongst radiologists and radiology registrars undergoing mammographic readings.