A Deep Learning Model to Triage Screening Mammograms: A Simulation Study

Potential Impact of an Artificial Intelligence-based Mammography Triage Algorithm on Performance and Workload in a Population-based Screening Sample

OBJECTIVE

To evaluate potential screening mammography performance and workload impact using a commercial artificial intelligence (AI)-based triage device in a population-based screening sample.

METHODS

In this retrospective study, a sample of 2129 women who underwent screening mammograms were evaluated. The performance of a commercial AI-based triage device was compared with radiologists' reports, actual outcomes, and national benchmarks using commonly used mammography metrics. Up to 5 years of follow-up examination results were evaluated in cases to establish benignity. The algorithm sorted cases into groups of "suspicious" and "low suspicion." A theoretical workload reduction was calculated by subtracting cases triaged as "low suspicion" from the sample.

RESULTS

At the default 93% sensitivity setting, there was significant improvement (P <.05) in the following triage simulation mean performance measures compared with actual outcome: 45.5% improvement in recall rate (13.4% to 7.3%; 95% CI, 6.2-8.3), 119% improvement in positive predictive value (PPV) 1 (5.3% to 11.6%; 95% CI, 9.96-13.4), 28.5% improvement in PPV2 (24.6% to 31.6%; 95% CI, 24.8-39.1), 20% improvement in sensitivity (83.3% to 100%; 95% CI, 100-100), and 7.2% improvement in specificity (87.2% to 93.5%; 95% CI, 92.4-94.5). A theoretical 62.5% workload reduction was possible. At the ultrahigh 99% sensitivity setting, a theoretical 27% workload reduction was possible. No cancers were missed by the algorithm at either sensitivity.

CONCLUSION

Artificial intelligence-based triage in this simulation demonstrated potential for significant improvement in mammography performance and predicted substantial theoretical workload reduction without any missed cancers.

Diagnostic capabilities of artificial intelligence as an additional reader in a breast cancer screening program

OBJECTIVES

We aimed to evaluate the early-detection capabilities of AI in a screening program over its duration, with a specific focus on the detection of interval cancers, the early detection of cancers with the assistance of AI from prior visits, and its impact on workload for various reading scenarios.

MATERIALS AND METHODS

The study included 22,621 mammograms of 8825 women within a 10-year biennial two-reader screening program. The statistical analysis focused on 5136 mammograms from 4282 women due to data retrieval issues, among whom 105 were diagnosed with breast cancer. The AI software assigned scores from 1 to 100. Histopathology results determined the ground truth, and Youden's index was used to establish a threshold. Tumor characteristics were analyzed with ANOVA and chi-squared test, and different workflow scenarios were evaluated using bootstrapping.

RESULTS

The AI software achieved an AUC of 89.6% (86.1-93.2%, 95% CI). The optimal threshold was 30.44, yielding 72.38% sensitivity and 92.86% specificity. Initially, AI identified 57 screening-detected cancers (83.82%), 15 interval cancers (51.72%), and 4 missed cancers (50%). AI as a second reader could have led to earlier diagnosis in 24 patients (average 29.92 ± 19.67 months earlier). No significant differences were found in cancer-characteristics groups. A hybrid triage workflow scenario showed a potential 69.5% reduction in workload and a 30.5% increase in accuracy.

CONCLUSION

This AI system exhibits high sensitivity and specificity in screening mammograms, effectively identifying interval and missed cancers and identifying 23% of cancers earlier in prior mammograms. Adopting AI as a triage mechanism has the potential to reduce workload by nearly 70%.

CLINICAL RELEVANCE STATEMENT

The study proposes a more efficient method for screening programs, both in terms of workload and accuracy.

KEY POINTS

• Incorporating AI as a triage tool in screening workflow improves sensitivity (72.38%) and specificity (92.86%), enhancing detection rates for interval and missed cancers. • AI-assisted triaging is effective in differentiating low and high-risk cases, reduces radiologist workload, and potentially enables broader screening coverage. • AI has the potential to facilitate early diagnosis compared to human reading.

Artificial intelligence for triaging of breast cancer screening mammograms and workload reduction: A meta-analysis of a deep learning software

OBJECTIVE

Deep learning (DL) has shown promising results for improving mammographic breast cancer diagnosis. However, the impact of artificial intelligence (AI) on the breast cancer screening process has not yet been fully elucidated in terms of potential workload reduction. We aim to assess if AI-based triaging of breast cancer screening mammograms could reduce the radiologist's workload with non-inferior sensitivity.

METHODS

PubMed, EMBASE, Cochrane Central, and Web of Science databases were systematically searched for studies that evaluated AI algorithms on computer-aided triage of breast cancer screening mammograms. We extracted data from homogenous studies and performed a proportion meta-analysis with a random-effects model to examine the radiologist's workload reduction (proportion of low-risk mammograms that could be theoretically ruled out from human's assessment) and the software's sensitivity to breast cancer detection.

RESULTS

Thirteen studies were selected for full review, and three studies that used the same commercially available DL algorithm were included in the meta-analysis. In the 156,852 examinations included, the threshold of 7 was identified as optimal. With these parameters, radiologist workload decreased by 68.3% (95%CI 0.655-0.711, I² = 98.76%, p < 0.001), while achieving a sensitivity of 93.1% (95%CI 0.882-0.979, I² = 83.86%, p = 0.002) and a specificity of 68.7% (95% CI 0.684-0.723, I² = 97.5%, p < 0.01).

CONCLUSIONS

The deployment of DL computer-aided triage of breast cancer screening mammograms reduces the radiology workload while maintaining high sensitivity. Although the implementation of AI remains complex and heterogeneous, it is a promising tool to optimize healthcare resources.

Comparison of AI-integrated pathways with human-AI interaction in population mammographic screening for breast cancer

Artificial intelligence (AI) readers of mammograms compare favourably to individual radiologists in detecting breast cancer. However, AI readers cannot perform at the level of multi-reader systems used by screening programs in countries such as Australia, Sweden, and the UK. Therefore, implementation demands human-AI collaboration. Here, we use a large, high-quality retrospective mammography dataset from Victoria, Australia to conduct detailed simulations of five potential AI-integrated screening pathways, and examine human-AI interaction effects to explore automation bias. Operating an AI reader as a second reader or as a high confidence filter improves current screening outcomes by 1.9-2.5% in sensitivity and up to 0.6% in specificity, achieving 4.6-10.9% reduction in assessments and 48-80.7% reduction in human reads. Automation bias degrades performance in multi-reader settings but improves it for single-readers. This study provides insight into feasible approaches for AI-integrated screening pathways and prospective studies necessary prior to clinical adoption.

The digital revolution in pathology: Towards a smarter approach to research and treatment

Artificial intelligence (AI) applications in oncology are at the forefront of transforming healthcare during the Fourth Industrial Revolution, driven by the digital data explosion. This review provides an accessible introduction to the field of AI, presenting a concise yet structured overview of the foundations of AI, including expert systems, classical machine learning, and deep learning, along with their contextual application in clinical research and healthcare. We delve into the current applications of AI in oncology, with a particular focus on diagnostic imaging and pathology. Numerous AI tools have already received regulatory approval, and more are under active development, bringing clear benefits but not without challenges. We discuss the importance of data security, the need for transparent and interpretable models, and the ethical considerations that must guide AI development in healthcare. By providing a perspective on the opportunities and challenges, this review aims to inform and guide researchers, clinicians, and policymakers in the adoption of AI in oncology.

Reproducibility and Explainability of Deep Learning in Mammography: A Systematic Review of Literature

Background  Although abundant literature is currently available on the use of deep learning for breast cancer detection in mammography, the quality of such literature is widely variable. Purpose  To evaluate published literature on breast cancer detection in mammography for reproducibility and to ascertain best practices for model design. Methods  The PubMed and Scopus databases were searched to identify records that described the use of deep learning to detect lesions or classify images into cancer or noncancer. A modification of Quality Assessment of Diagnostic Accuracy Studies (mQUADAS-2) tool was developed for this review and was applied to the included studies. Results of reported studies (area under curve [AUC] of receiver operator curve [ROC] curve, sensitivity, specificity) were recorded. Results  A total of 12,123 records were screened, of which 107 fit the inclusion criteria. Training and test datasets, key idea behind model architecture, and results were recorded for these studies. Based on mQUADAS-2 assessment, 103 studies had high risk of bias due to nonrepresentative patient selection. Four studies were of adequate quality, of which three trained their own model, and one used a commercial network. Ensemble models were used in two of these. Common strategies used for model training included patch classifiers, image classification networks (ResNet in 67%), and object detection networks (RetinaNet in 67%). The highest reported AUC was 0.927 ± 0.008 on a screening dataset, while it reached 0.945 (0.919-0.968) on an enriched subset. Higher values of AUC (0.955) and specificity (98.5%) were reached when combined radiologist and Artificial Intelligence readings were used than either of them alone. None of the studies provided explainability beyond localization accuracy. None of the studies have studied interaction between AI and radiologist in a real world setting. Conclusion  While deep learning holds much promise in mammography interpretation, evaluation in a reproducible clinical setting and explainable networks are the need of the hour.

Impact of real-life use of artificial intelligence as support for human reading in a population-based breast cancer screening program with mammography and tomosynthesis

OBJECTIVES

To evaluate the impact of using an artificial intelligence (AI) system as support for human double reading in a real-life scenario of a breast cancer screening program with digital mammography (DM) or digital breast tomosynthesis (DBT).

MATERIAL AND METHODS

We analyzed the performance of double reading screening with mammography and tomosynthesis after implementarion of AI as decision support. The study group consisted of a consecutive cohort of 1 year screening between March 2021 and March 2022 where double reading was performed with concurrent AI support that automatically detects and highlights lesions suspicious of breast cancer in mammography and tomosynthesis. Screening performance was measured as cancer detection rate (CDR), recall rate (RR), and positive predictive value (PPV) of recalls. Performance in the study group was compared using a McNemar test to a control group that included a screening cohort of the same size, recorded just prior to the implementation of AI.

RESULTS

A total of 11,998 women (mean age 57.59 years ± 5.8 [sd]) were included in the study group (5049 DM and 6949 DBT). Comparing global results (including DM and DBT) of double reading with vs. without AI support, we observed an increase in CDR, PPV, and RR by 3.2/‰ (5.8 vs. 9; p < 0.001), 4% (10.6 vs. 14.6; p < 0.001), and 0.7% (5.4 vs. 6.1; p < 0.001) respectively.

CONCLUSION

AI used as support for human double reading in a real-life breast cancer screening program with DM and DBT increases CDR and PPV of the recalled women.

CLINICAL RELEVANCE STATEMENT

Artificial intelligence as support for human double reading improves accuracy in a real-life breast cancer screening program both in digital mammography and digital breast tomosynthesis.

KEY POINTS

• AI systems based on deep learning technology offer potential for improving breast cancer screening programs. • Using artificial intelligence as support for reading improves radiologists' performance in breast cancer screening programs with mammography or tomosynthesis. • Artificial intelligence used concurrently with human reading in clinical screening practice increases breast cancer detection rate and positive predictive value of the recalled women.

New Frontiers in Breast Cancer Imaging: The Rise of AI

Artificial intelligence (AI) has been implemented in multiple fields of medicine to assist in the diagnosis and treatment of patients. AI implementation in radiology, more specifically for breast imaging, has advanced considerably. Breast cancer is one of the most important causes of cancer mortality among women, and there has been increased attention towards creating more efficacious methods for breast cancer detection utilizing AI to improve radiologist accuracy and efficiency to meet the increasing demand of our patients. AI can be applied to imaging studies to improve image quality, increase interpretation accuracy, and improve time efficiency and cost efficiency. AI applied to mammography, ultrasound, and MRI allows for improved cancer detection and diagnosis while decreasing intra- and interobserver variability. The synergistic effect between a radiologist and AI has the potential to improve patient care in underserved populations with the intention of providing quality and equitable care for all. Additionally, AI has allowed for improved risk stratification. Further, AI application can have treatment implications as well by identifying upstage risk of ductal carcinoma in situ (DCIS) to invasive carcinoma and by better predicting individualized patient response to neoadjuvant chemotherapy. AI has potential for advancement in pre-operative 3-dimensional models of the breast as well as improved viability of reconstructive grafts.

A Semiautonomous Deep Learning System to Reduce False Positives in Screening Mammography

Purpose To evaluate the ability of a semiautonomous artificial intelligence (AI) model to identify screening mammograms not suspicious for breast cancer and reduce the number of false-positive examinations. Materials and Methods The deep learning algorithm was trained using 123 248 two-dimensional digital mammograms (6161 cancers) and a retrospective study was performed on three nonoverlapping datasets of 14 831 screening mammography examinations (1026 cancers) from two U.S. institutions and one U.K. institution (2008-2017). The stand-alone performance of humans and AI was compared. Human plus AI performance was simulated to examine reductions in the cancer detection rate, number of examinations, false-positive callbacks, and benign biopsies. Metrics were adjusted to mimic the natural distribution of a screening population, and bootstrapped CIs and P values were calculated. Results Retrospective evaluation on all datasets showed minimal changes to the cancer detection rate with use of the AI device (noninferiority margin of 0.25 cancers per 1000 examinations: U.S. dataset 1, P = .02; U.S. dataset 2, P < .001; U.K. dataset, P < .001). On U.S. dataset 1 (11 592 mammograms; 101 cancers; 3810 female patients; mean age, 57.3 years ± 10.0 [SD]), the device reduced screening examinations requiring radiologist interpretation by 41.6% (95% CI: 40.6%, 42.4%; P < .001), diagnostic examinations callbacks by 31.1% (95% CI: 28.7%, 33.4%; P < .001), and benign needle biopsies by 7.4% (95% CI: 4.1%, 12.4%; P < .001). U.S. dataset 2 (1362 mammograms; 330 cancers; 1293 female patients; mean age, 55.4 years ± 10.5) was reduced by 19.5% (95% CI: 16.9%, 22.1%; P < .001), 11.9% (95% CI: 8.6%, 15.7%; P < .001), and 6.5% (95% CI: 0.0%, 19.0%; P = .08), respectively. The U.K. dataset (1877 mammograms; 595 cancers; 1491 female patients; mean age, 63.5 years ± 7.1) was reduced by 36.8% (95% CI: 34.4%, 39.7%; P < .001), 17.1% (95% CI: 5.9%, 30.1%: P < .001), and 5.9% (95% CI: 2.9%, 11.5%; P < .001), respectively. Conclusion This work demonstrates the potential of a semiautonomous breast cancer screening system to reduce false positives, unnecessary procedures, patient anxiety, and medical expenses. Keywords: Artificial Intelligence, Semiautonomous Deep Learning, Breast Cancer, Screening Mammography Supplemental material is available for this article. Published under a CC BY 4.0 license.

Unsupervised feature correlation model to predict breast abnormal variation maps in longitudinal mammograms

Breast cancer continues to be a significant cause of mortality among women globally. Timely identification and precise diagnosis of breast abnormalities are critical for enhancing patient prognosis. In this study, we focus on improving the early detection and accurate diagnosis of breast abnormalities, which is crucial for improving patient outcomes and reducing the mortality rate of breast cancer. To address the limitations of traditional screening methods, a novel unsupervised feature correlation network was developed to predict maps indicating breast abnormal variations using longitudinal 2D mammograms. The proposed model utilizes the reconstruction process of current year and prior year mammograms to extract tissue from different areas and analyze the differences between them to identify abnormal variations that may indicate the presence of cancer. The model incorporates a feature correlation module, an attention suppression gate, and a breast abnormality detection module, all working together to improve prediction accuracy. The proposed model not only provides breast abnormal variation maps but also distinguishes between normal and cancer mammograms, making it more advanced compared to the state-of-the-art baseline models. The results of the study show that the proposed model outperforms the baseline models in terms of Accuracy, Sensitivity, Specificity, Dice score, and cancer detection rate.

Transfer learning for the generalization of artificial intelligence in breast cancer detection: a case-control study

BACKGROUND

Some researchers have questioned whether artificial intelligence (AI) systems maintain their performance when used for women from populations not considered during the development of the system.

PURPOSE

To evaluate the impact of transfer learning as a way of improving the generalization of AI systems in the detection of breast cancer.

MATERIAL AND METHODS

This retrospective case-control Finnish study involved 191 women diagnosed with breast cancer and 191 matched healthy controls. We selected a state-of-the-art AI system for breast cancer detection trained using a large US dataset. The selected baseline system was evaluated in two experimental settings. First, we examined our private Finnish sample as an independent test set that had not been considered in the development of the system (unseen population). Second, the baseline system was retrained to attempt to improve its performance in the unseen population by means of transfer learning. To analyze performance, we used areas under the receiver operating characteristic curve (AUCs) with DeLong's test.

RESULTS

Two versions of the baseline system were considered: ImageOnly and Heatmaps. The ImageOnly and Heatmaps versions yielded mean AUC values of 0.82±0.008 and 0.88±0.003 in the US dataset and 0.56 (95% CI=0.50-0.62) and 0.72 (95% CI=0.67-0.77) when evaluated in the unseen population, respectively. The retrained systems achieved AUC values of 0.61 (95% CI=0.55-0.66) and 0.69 (95% CI=0.64-0.75), respectively. There was no statistical difference between the baseline system and the retrained system.

CONCLUSION

Transfer learning with a small study sample did not yield a significant improvement in the generalization of the system.

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.

The role of an artificial intelligence software in clinical senology: a mammography multi-reader study

PURPOSE

To evaluate the diagnostic role of a dedicated AI software in detecting anomalous breast findings on mammography and tomosynthesis images in the clinical setting, stand-alone and as aid of four readers.

METHODS

A total of 210 patients with complete clinical and radiologic records were retrospectively analyzed. Pathology was used as the reference standard for patients undergoing surgery or biopsy, and a 1-year follow-up was used to confirm no change in the remaining patients. The image evaluation was performed by four readers with different levels of experience (a junior and three senior breast radiologists) using a 5-point Likert scale moving from 1 (definitively no cancer) to 5 (definitively cancer). The positivity of mammograms was assessed on the presence of any breast lesion (masses, architectural distortions, asymmetries, calcifications), including malignant and benign ones. A multi-reader multi-case analysis was performed. A p value < 0.05 was considered statistically significant.

RESULTS

The stand-alone AI system achieved an accuracy of 71% (69% sensitivity and 73% specificity), which is overall lower than the value achieved by readers without AI. However, with the aid of AI, a significant increase of accuracy (p value  = 0.004) and specificity (p value  = 0.04) was achieved for the less experienced radiologist and a senior one.

CONCLUSION

The use of AI software as a second reader for breast lesions assessment could play a crucial role in the clinical setting, by increasing sensitivity and specificity, especially for less experienced radiologists.

Systematic reviews of machine learning in healthcare: a literature review

INTRODUCTION

The increasing availability of data and computing power has made machine learning (ML) a viable approach to faster, more efficient healthcare delivery.

METHODS

A systematic literature review (SLR) of published SLRs evaluating ML applications in healthcare settings published between1 January 2010 and 27 March 2023 was conducted.

RESULTS

In total 220 SLRs covering 10,462 ML algorithms were reviewed. The main application of AI in medicine related to the clinical prediction and disease prognosis in oncology and neurology with the use of imaging data. Accuracy, specificity, and sensitivity were provided in 56%, 28%, and 25% SLRs respectively. Internal and external validation was reported in 53% and less than 1% of the cases respectively. The most common modeling approach was neural networks (2,454 ML algorithms), followed by support vector machine and random forest/decision trees (1,578 and 1,522 ML algorithms, respectively).

EXPERT OPINION

The review indicated considerable reporting gaps in terms of the ML's performance, both internal and external validation. Greater accessibility to healthcare data for developers can ensure the faster adoption of ML algorithms into clinical practice.

A deep learning-based semiautomated workflow for triaging follow-up MR scans in treated nasopharyngeal carcinoma

It is imperative to optimally utilize virtues and obviate defects of fully automated analysis and expert knowledge in new paradigms of healthcare. We present a deep learning-based semiautomated workflow (RAINMAN) with 12,809 follow-up scans among 2,172 patients with treated nasopharyngeal carcinoma from three centers (ChiCTR.org.cn, Chi-CTR2200056595). A boost of diagnostic performance and reduced workload was observed in RAINMAN compared with the original manual interpretations (internal vs. external: sensitivity, 2.5% [p = 0.500] vs. 3.2% [p = 0.031]; specificity, 2.9% [p < 0.001] vs. 0.3% [p = 0.302]; workload reduction, 79.3% vs. 76.2%). The workflow also yielded a triaging performance of 83.6%, with increases of 1.5% in sensitivity (p = 1.000) and 0.6%-1.3% (all p < 0.05) in specificity compared to three radiologists in the reader study. The semiautomated workflow shows its unique superiority in reducing radiologist's workload by eliminating negative scans while retaining the diagnostic performance of radiologists.

Machine learning-based models for the prediction of breast cancer recurrence risk

Breast cancer is the most common malignancy diagnosed in women worldwide. The prevalence and incidence of breast cancer is increasing every year; therefore, early diagnosis along with suitable relapse detection is an important strategy for prognosis improvement. This study aimed to compare different machine algorithms to select the best model for predicting breast cancer recurrence. The prediction model was developed by using eleven different machine learning (ML) algorithms, including logistic regression (LR), random forest (RF), support vector classification (SVC), extreme gradient boosting (XGBoost), gradient boosting decision tree (GBDT), decision tree, multilayer perceptron (MLP), linear discriminant analysis (LDA), adaptive boosting (AdaBoost), Gaussian naive Bayes (GaussianNB), and light gradient boosting machine (LightGBM), to predict breast cancer recurrence. The area under the curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and F1 score were used to evaluate the performance of the prognostic model. Based on performance, the optimal ML was selected, and feature importance was ranked by Shapley Additive Explanation (SHAP) values. Compared to the other 10 algorithms, the results showed that the AdaBoost algorithm had the best prediction performance for successfully predicting breast cancer recurrence and was adopted in the establishment of the prediction model. Moreover, CA125, CEA, Fbg, and tumor diameter were found to be the most important features in our dataset to predict breast cancer recurrence. More importantly, our study is the first to use the SHAP method to improve the interpretability of clinicians to predict the recurrence model of breast cancer based on the AdaBoost algorithm. The AdaBoost algorithm offers a clinical decision support model and successfully identifies the recurrence of breast cancer.

Deep learning performance for detection and classification of microcalcifications on mammography

BACKGROUND

Breast cancer screening through mammography is crucial for early detection, yet the demand for mammography services surpasses the capacity of radiologists. Artificial intelligence (AI) can assist in evaluating microcalcifications on mammography. We developed and tested an AI model for localizing and characterizing microcalcifications.

METHODS

Three expert radiologists annotated a dataset of mammograms using histology-based ground truth. The dataset was partitioned for training, validation, and testing. Three neural networks (AlexNet, ResNet18, and ResNet34) were trained and evaluated using specific metrics including receiver operating characteristics area under the curve (AUC), sensitivity, and specificity. The reported metrics were computed on the test set (10% of the whole dataset).

RESULTS

The dataset included 1,000 patients aged 21-73 years and 1,986 mammograms (180 density A, 220 density B, 380 density C, and 220 density D), with 389 malignant and 611 benign groups of microcalcifications. AlexNet achieved the best performance with 0.98 sensitivity, 0.89 specificity of, and 0.98 AUC for microcalcifications detection and 0.85 sensitivity, 0.89 specificity, and 0.94 AUC of for microcalcifications classification. For microcalcifications detection, ResNet18 and ResNet34 achieved 0.96 and 0.97 sensitivity, 0.91 and 0.90 specificity and 0.98 and 0.98 AUC, retrospectively. For microcalcifications classification, ResNet18 and ResNet34 exhibited 0.75 and 0.84 sensitivity, 0.85 and 0.84 specificity, and 0.88 and 0.92 AUC, respectively.

CONCLUSIONS

The developed AI models accurately detect and characterize microcalcifications on mammography.

RELEVANCE STATEMENT

AI-based systems have the potential to assist radiologists in interpreting microcalcifications on mammograms. The study highlights the importance of developing reliable deep learning models possibly applied to breast cancer screening.

KEY POINTS

• A novel AI tool was developed and tested to aid radiologists in the interpretation of mammography by accurately detecting and characterizing microcalcifications. • Three neural networks (AlexNet, ResNet18, and ResNet34) were trained, validated, and tested using an annotated dataset of 1,000 patients and 1,986 mammograms. • The AI tool demonstrated high accuracy in detecting/localizing and characterizing microcalcifications on mammography, highlighting the potential of AI-based systems to assist radiologists in the interpretation of mammograms.

External Evaluation of a Mammography-based Deep Learning Model for Predicting Breast Cancer in an Ethnically Diverse Population

Purpose

To externally evaluate a mammography-based deep learning (DL) model (Mirai) in a high-risk racially diverse population and compare its performance with other mammographic measures.

Materials and Methods

A total of 6435 screening mammograms in 2096 female patients (median age, 56.4 years ± 11.2 [SD]) enrolled in a hospital-based case-control study from 2006 to 2020 were retrospectively evaluated. Pathologically confirmed breast cancer was the primary outcome. Mirai scores were the primary predictors. Breast density and Breast Imaging Reporting and Data System (BI-RADS) assessment categories were comparative predictors. Performance was evaluated using area under the receiver operating characteristic curve (AUC) and concordance index analyses.

Results

Mirai achieved 1- and 5-year AUCs of 0.71 (95% CI: 0.68, 0.74) and 0.65 (95% CI: 0.64, 0.67), respectively. One-year AUCs for nondense versus dense breasts were 0.72 versus 0.58 (P = .10). There was no evidence of a difference in near-term discrimination performance between BI-RADS and Mirai (1-year AUC, 0.73 vs 0.68; P = .34). For longer-term prediction (2-5 years), Mirai outperformed BI-RADS assessment (5-year AUC, 0.63 vs 0.54; P < .001). Using only images of the unaffected breast reduced the discriminatory performance of the DL model (P < .001 at all time points), suggesting that its predictions are likely dependent on the detection of ipsilateral premalignant patterns.

Conclusion

A mammography DL model showed good performance in a high-risk external dataset enriched for African American patients, benign breast disease, and BRCA mutation carriers, and study findings suggest that the model performance is likely driven by the detection of precancerous changes.Keywords: Breast, Cancer, Computer Applications, Convolutional Neural Network, Deep Learning Algorithms, Informatics, Epidemiology, Machine Learning, Mammography, Oncology, Radiomics Supplemental material is available for this article. © RSNA, 2023See also commentary by Kontos and Kalpathy-Cramer in this issue.

Deep learning for detection of iso-dense, obscure masses in mammographically dense breasts

OBJECTIVES

To analyze the performance of deep learning in isodense/obscure masses in dense breasts. To build and validate a deep learning (DL) model using core radiology principles and analyze its performance in isodense/obscure masses. To show performance on screening mammography as well as diagnostic mammography distribution.

METHODS

This was a retrospective, single-institution, multi-centre study with external validation. For model building, we took a 3-pronged approach. First, we explicitly taught the network to learn features other than density differences: such as spiculations and architectural distortion. Second, we used the opposite breast to enable the detection of asymmetries. Third, we systematically enhanced each image by piece-wise-linear transformation. We tested the network on a diagnostic mammography dataset (2569 images with 243 cancers, January to June 2018) and a screening mammography dataset (2146 images with 59 cancers, patient recruitment from January to April 2021) from a different centre (external validation).

RESULTS

When trained with our proposed technique (and compared with baseline network), sensitivity for malignancy increased from 82.7 to 84.7% at 0.2 False positives per image (FPI) in the diagnostic mammography dataset, 67.9 to 73.8% in the subset of patients with dense breasts, 74.6 to 85.3 in the subset of patients with isodense/obscure cancers and 84.9 to 88.7 in an external validation test set with a screening mammography distribution. We showed that our sensitivity exceeded currently reported values (0.90 at 0.2 FPI) on a public benchmark dataset (INBreast).

CONCLUSION

Modelling traditional mammographic teaching into a DL framework can help improve cancer detection accuracy in dense breasts.

CLINICAL RELEVANCE STATEMENT

Incorporating medical knowledge into neural network design can help us overcome some limitations associated with specific modalities. In this paper, we show how one such deep neural network can help improve performance on mammographically dense breasts.

KEY POINTS

• Although state-of-the-art deep learning networks achieve good results in cancer detection in mammography in general, isodense, obscure masses and mammographically dense breasts posed a challenge to deep learning networks. • Collaborative network design and incorporation of traditional radiology teaching into the deep learning approach helped mitigate the problem. • The accuracy of deep learning networks may be translatable to different patient distributions. We showed the results of our network on screening as well as diagnostic mammography datasets.

Deep learning-based recurrence detector on magnetic resonance scans in nasopharyngeal carcinoma: A multicenter study

OBJECTIVES

Accuracy in the detection of recurrent nasopharyngeal carcinoma (NPC) on follow-up magnetic resonance (MR) scans needs to be improved.

MATERIAL AND METHODS

A total of 5 035 follow-up MR scans from 5 035 survivors with treated NPC between April 2007 and July 2020 were retrospectively collected from three cancer centers for developing and evaluating the deep learning (DL) model MODERN (MR-based Deep learning model for dEtecting Recurrent Nasopharyngeal carcinoma). In a reader study with 220 scans, the accuracy of two radiologists in detecting recurrence on scans with vs without MODERN was evaluated. The performance was measured using the area under the receiver operating characteristic curve (ROC-AUC) and accuracy with a 95% confidence interval (CI).

RESULTS

MODERN exhibited sound performance in the validation cohort (internal: ROC-AUC, 0.88, 95% CI, 0.86-0.90; external 1: ROC-AUC, 0.88, 95% CI, 0.86-0.90; external 2: ROC-AUC, 0.85, 95% CI, 0.82-0.88). In a reader study, MODERN alone achieved reliable accuracy compared to that of radiologists (MODERN: 84.1%, 95% CI, 79.3%-88.9%; competent: 80.9%, 95% CI, 75.7%-86.1%, P < 0.001; expert: 85.9%, 95% CI, 81.3%-90.5%, P < 0.001). The accuracy of radiologists was boosted by the MODERN score (competent with MODERN score: 84.6%, 95% CI, 79.8%-89.3%, P < 0.001; expert with MODERN score: 87.7%, 95% CI, 83.4%-92.1%, P < 0.001).

CONCLUSION

We developed a DL model for recurrence detection with reliable performance. Computer-human collaboration has the potential to refine the workflow in interpreting surveillant MR scans among patients with treated NPC.

Mammography Breast Cancer Screening Triage Using Deep Learning: A UK Retrospective Study

Background Breast screening enables early detection of cancers; however, most women have normal mammograms, resulting in repetitive and resource-intensive reading tasks. Purpose To investigate if deep learning (DL) algorithms can be used to triage mammograms by identifying normal results to reduce workload or flag cancers that may be overlooked. Materials and Methods In this retrospective study, three commercial DL algorithms were investigated using consecutive mammograms from two UK Breast Screening Program sites from January 2015 to December 2017 and January 2017 to December 2018 on devices from two mammography vendors. Normal mammograms with a 3-year follow-up and histopathologically proven cancer detected at screening, the subsequent round, or in the 3-year interval were included. Two algorithm thresholds were set: in scenario A, 99.0% sensitivity for rule-out triage to a lone reader, and in scenario B, approximately 1.0% additional recall providing a rule-in triage for further assessment. Both thresholds were then applied to the screening workflow in scenario C. The sensitivity and specificity were used to assess the overall predictive performance of each DL algorithm. Results The data set comprised 78 849 patients (median age, 59 years [IQR, 53-63 years]) and 887 screening-detected, 439 interval, and 688 subsequent screening round-detected cancers. In scenario A (rule-out triage), models DL-1, DL-2, and DL-3 triaged 35.0% (27 565 of 78 849), 53.2% (41 937 of 78 849), and 55.6% (43 869 of 78 849) of mammograms, respectively, with 0.0% (0 of 887) to 0.1% (one of 887) of screening-detected cancers undetected. In scenario B, DL algorithms triaged in 4.6% (20 of 439) to 8.2% (36 of 439) of interval and 5.2% (36 of 688) to 6.1% (42 of 688) of subsequent-round cancers when applied after the routine double-reading workflow. Combining both approaches in scenario C resulted in an overall noninferior specificity (difference, -0.9%; P < .001) and superior sensitivity (difference, 2.7%; P < .001) for the adaptive workflow compared with routine double reading for all three algorithms. Conclusion Rule-out and rule-in DL-adapted triage workflows can improve the efficiency and efficacy of mammography breast cancer screening. © RSNA, 2023 Supplemental material is available for this article. See also the editorial by Nishikawa and Lu in this issue.

Mammography diagnosis of breast cancer screening through machine learning: a systematic review and meta-analysis

Breast cancer was the fourth leading cause of cancer-related death worldwide, and early mammography screening could decrease the breast cancer mortality. Artificial intelligence (AI)-assisted diagnose system based on machine learning (ML) methods can help improve the screening accuracy and efficacy. This study aimed to systematically review and make a meta-analysis on the diagnostic accuracy of mammography diagnosis of breast cancer through various ML methods. Springer Link, Science Direct (Elsevier), IEEE Xplore, PubMed and Web of Science were searched for relevant studies published from January 2000 to September 2021. The study was registered with the PROSPERO International Prospective Register of Systematic Reviews (protocol no. CRD42021284227). A Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) was used to assess the included studies, and reporting was evaluated using the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA). The pooled summary estimates for sensitivity, specificity, the area under the receiver operating characteristic curve (AUC) for three ML methods (convolutional neural network [CNN], artificial neural network [ANN], support vector machine [SVM]) were calculated. A total of 32 studies with 23,804 images were included in the meta-analysis. The overall pooled estimate for sensitivity, specificity and AUC was 0.914 [95% CI 0.868-0.945], 0.916 [95% CI 0.873-0.945] and 0.945 for mammography diagnosis of breast cancer through three ML methods. The pooled sensitivity, specificity and AUC of CNN were 0.961 [95% CI 0.886-0.988], 0.950 [95% CI 0.924-0.967] and 0.974. The pooled sensitivity, specificity and AUC of ANN were 0.837 [95% CI 0.772-0.886], 0.894 [95% CI 0.764-0.957] and 0.881. The pooled sensitivity, specificity and AUC of SVM were 0.889 [95% CI 0.807-0.939], 0.843 [95% CI 0.724-0.916] and 0.913. Machine learning methods (especially CNN) show excellent performance in mammography diagnosis of breast cancer screening based on retrospective studies. More rigorous prospective studies are needed to evaluate the longitudinal performance of AI.

Effect of artificial intelligence-based computer-aided diagnosis on the screening outcomes of digital mammography: a matched cohort study

OBJECTIVE

To investigate whether artificial intelligence-based computer-aided diagnosis (AI-CAD) can improve radiologists' performance when used to support radiologists' interpretation of digital mammography (DM) in breast cancer screening.

METHODS

A retrospective database search identified 3158 asymptomatic Korean women who consecutively underwent screening DM between January and December 2019 without AI-CAD support, and screening DM between February and July 2020 with image interpretation aided by AI-CAD in a tertiary referral hospital using single reading. Propensity score matching was used to match the DM with AI-CAD group in a 1:1 ratio with the DM without AI-CAD group according to age, breast density, experience level of the interpreting radiologist, and screening round. Performance measures were compared with the McNemar test and generalized estimating equations.

RESULTS

A total of 1579 women who underwent DM with AI-CAD were matched with 1579 women who underwent DM without AI-CAD. Radiologists showed higher specificity (96% [1500 of 1563] vs 91.6% [1430 of 1561]; p < 0.001) and lower abnormal interpretation rates (AIR) (4.9% [77 of 1579] vs 9.2% [145 of 1579]; p < 0.001) with AI-CAD than without. There was no significant difference in the cancer detection rate (CDR) (AI-CAD vs no AI-CAD, 8.9 vs 8.9 per 1000 examinations; p = 0.999), sensitivity (87.5% vs 77.8%; p = 0.999), and positive predictive value for biopsy (PPV3) (35.0% vs 35.0%; p = 0.999) according to AI-CAD support.

CONCLUSIONS

AI-CAD increases the specificity for radiologists without decreasing sensitivity as a supportive tool in the single reading of DM for breast cancer screening.

CLINICAL RELEVANCE STATEMENT

This study shows that AI-CAD could improve the specificity of radiologists' DM interpretation in the single reading system without decreasing sensitivity, suggesting that it can benefit patients by reducing false positive and recall rates.

KEY POINTS

• In this retrospective-matched cohort study (DM without AI-CAD vs DM with AI-CAD), radiologists showed higher specificity and lower AIR when AI-CAD was used to support decision-making in DM screening. • CDR, sensitivity, and PPV for biopsy did not differ with and without AI-CAD support.

Overview of trials on artificial intelligence algorithms in breast cancer screening - A roadmap for international evaluation and implementation

Accumulating evidence from retrospective studies demonstrate at least non-inferior performance when using AI algorithms with different strategies versus double-reading in mammography screening. In addition, AI algorithms for mammography screening can reduce work load by moving to single human reading. Prospective trials are essential to avoid unintended adverse consequences before incorporation of AI algorithms into UK's National Health Service (NHS) Breast Screening Programme (BSP). A stakeholders' meeting was organized in Newnham College, Cambridge, UK to undertake a review of the current evidence to enable consensus discussion on next steps required before implementation into a screening programme. It was concluded that a multicentre multivendor testing platform study with opt-out consent is preferred. AI thresholds from different vendors should be determined while maintaining non-inferior screening performance results, particularly ensuring recall rates are not increased. Automatic recall of cases using an agreed high sensitivity AI score versus automatic rule out with a low AI score set at a high sensitivity could be used. A human reader should still be involved in decision making with AI-only recalls requiring human arbitration. Standalone AI algorithms used without prompting maintain unbiased screening reading performance, but reading with prompts should be tested prospectively and ideally provided for arbitration.

FSE-Net: feature selection and enhancement network for mammogram classification

Objective. Early detection and diagnosis allow for intervention and treatment at an early stage of breast cancer. Despite recent advances in computer aided diagnosis systems based on convolutional neural networks for breast cancer diagnosis, improving the classification performance of mammograms remains a challenge due to the various sizes of breast lesions and difficult extraction of small lesion features. To obtain more accurate classification results, many studies choose to directly classify region of interest (ROI) annotations, but labeling ROIs is labor intensive. The purpose of this research is to design a novel network to automatically classify mammogram image as cancer and no cancer, aiming to mitigate or address the above challenges and help radiologists perform mammogram diagnosis more accurately.Approach. We propose a novel feature selection and enhancement network (FSE-Net) to fully exploit the features of mammogram images, which requires only mammogram images and image-level labels without any bounding boxes or masks. Specifically, to obtain more contextual information, an effective feature selection module is proposed to adaptively select the receptive fields and fuse features from receptive fields of different scales. Moreover, a feature enhancement module is designed to explore the correlation between feature maps of different resolutions and to enhance the representation capacity of low-resolution feature maps with high-resolution feature maps.Main results. The performance of the proposed network has been evaluated on the CBIS-DDSM dataset and INbreast dataset. It achieves an accuracy of 0.806 with an AUC of 0.866 on the CBIS-DDSM dataset and an accuracy of 0.956 with an AUC of 0.974 on the INbreast dataset.Significance. Through extensive experiments and saliency map visualization analysis, the proposed network achieves the satisfactory performance in the mammogram classification task, and can roughly locate suspicious regions to assist in the final prediction of the entire images.

Effect of Benign Biopsy Findings on an Artificial Intelligence-Based Cancer Detector in Screening Mammography: Retrospective Case-Control Study

BACKGROUND

Artificial intelligence (AI)-based cancer detectors (CAD) for mammography are starting to be used for breast cancer screening in radiology departments. It is important to understand how AI CAD systems react to benign lesions, especially those that have been subjected to biopsy.

OBJECTIVE

Our goal was to corroborate the hypothesis that women with previous benign biopsy and cytology assessments would subsequently present increased AI CAD abnormality scores even though they remained healthy.

METHODS

This is a retrospective study applying a commercial AI CAD system (Insight MMG, version 1.1.4.3; Lunit Inc) to a cancer-enriched mammography screening data set of 10,889 women (median age 56, range 40-74 years). The AI CAD generated a continuous prediction score for tumor suspicion between 0.00 and 1.00, where 1.00 represented the highest level of suspicion. A binary read (flagged or not flagged) was defined on the basis of a predetermined cutoff threshold (0.40). The flagged median and proportion of AI scores were calculated for women who were healthy, those who had a benign biopsy finding, and those who were diagnosed with breast cancer. For women with a benign biopsy finding, the interval between mammography and the biopsy was used for stratification of AI scores. The effect of increasing age was examined using subgroup analysis and regression modeling.

RESULTS

Of a total of 10,889 women, 234 had a benign biopsy finding before or after screening. The proportions of flagged healthy women were 3.5%, 11%, and 84% for healthy women without a benign biopsy finding, those with a benign biopsy finding, and women with breast cancer, respectively (P<.001). For the 8307 women with complete information, radiologist 1, radiologist 2, and the AI CAD system flagged 8.5%, 6.8%, and 8.5% of examinations of women who had a prior benign biopsy finding. The AI score correlated only with increasing age of the women in the cancer group (P=.01).

CONCLUSIONS

Compared to healthy women without a biopsy, the examined AI CAD system flagged a much larger proportion of women who had or would have a benign biopsy finding based on a radiologist's decision. However, the flagging rate was not higher than that for radiologists. Further research should be focused on training the AI CAD system taking prior biopsy information into account.

Expanding Horizons: The Realities of CAD, the Promise of Artificial Intelligence, and Machine Learning's Role in Breast Imaging beyond Screening Mammography

Artificial intelligence (AI) applications in mammography have gained significant popular attention; however, AI has the potential to revolutionize other aspects of breast imaging beyond simple lesion detection. AI has the potential to enhance risk assessment by combining conventional factors with imaging and improve lesion detection through a comparison with prior studies and considerations of symmetry. It also holds promise in ultrasound analysis and automated whole breast ultrasound, areas marked by unique challenges. AI's potential utility also extends to administrative tasks such as MQSA compliance, scheduling, and protocoling, which can reduce the radiologists' workload. However, adoption in breast imaging faces limitations in terms of data quality and standardization, generalizability, benchmarking performance, and integration into clinical workflows. Developing methods for radiologists to interpret AI decisions, and understanding patient perspectives to build trust in AI results, will be key future endeavors, with the ultimate aim of fostering more efficient radiology practices and better patient care.

Deep learning for predicting future lesion emergence in high-risk breast MRI screening: a feasibility study

BACKGROUND

International societies have issued guidelines for high-risk breast cancer (BC) screening, recommending contrast-enhanced magnetic resonance imaging (CE-MRI) of the breast as a supplemental diagnostic tool. In our study, we tested the applicability of deep learning-based anomaly detection to identify anomalous changes in negative breast CE-MRI screens associated with future lesion emergence.

METHODS

In this prospective study, we trained a generative adversarial network on dynamic CE-MRI of 33 high-risk women who participated in a screening program but did not develop BC. We defined an anomaly score as the deviation of an observed CE-MRI scan from the model of normal breast tissue variability. We evaluated the anomaly score's association with future lesion emergence on the level of local image patches (104,531 normal patches, 455 patches of future lesion location) and entire CE-MRI exams (21 normal, 20 with future lesion). Associations were analyzed by receiver operating characteristic (ROC) curves on the patch level and logistic regression on the examination level.

RESULTS

The local anomaly score on image patches was a good predictor for future lesion emergence (area under the ROC curve 0.804). An exam-level summary score was significantly associated with the emergence of lesions at any location at a later time point (p = 0.045).

CONCLUSIONS

Breast cancer lesions are associated with anomalous appearance changes in breast CE-MRI occurring before the lesion emerges in high-risk women. These early image signatures are detectable and may be a basis for adjusting individual BC risk and personalized screening.

RELEVANCE STATEMENT

Anomalies in screening MRI preceding lesion emergence in women at high-risk of breast cancer may inform individualized screening and intervention strategies.

KEY POINTS

• Breast lesions are associated with preceding anomalies in CE-MRI of high-risk women. • Deep learning-based anomaly detection can help to adjust risk assessment for future lesions. • An appearance anomaly score may be used for adjusting screening interval times.

Integration of Artificial Intelligence Decision Aids to Reduce Workload and Enhance Efficiency in Thyroid Nodule Management

Importance

To optimize the integration of artificial intelligence (AI) decision aids and reduce workload in thyroid nodule management, it is critical to incorporate personalized AI into the decision-making processes of radiologists with varying levels of expertise.

Objective

To develop an optimized integration of AI decision aids for reducing radiologists' workload while maintaining diagnostic performance compared with traditional AI-assisted strategy.

Design, Setting, and Participants

In this diagnostic study, a retrospective set of 1754 ultrasonographic images of 1048 patients with 1754 thyroid nodules from July 1, 2018, to July 31, 2019, was used to build an optimized strategy based on how 16 junior and senior radiologists incorporated AI-assisted diagnosis results with different image features. In the prospective set of this diagnostic study, 300 ultrasonographic images of 268 patients with 300 thyroid nodules from May 1 to December 31, 2021, were used to compare the optimized strategy with the traditional all-AI strategy in terms of diagnostic performance and workload reduction. Data analyses were completed in September 2022.

Main Outcomes and Measures

The retrospective set of images was used to develop an optimized integration of AI decision aids for junior and senior radiologists based on the selection of AI-assisted significant or nonsignificant features. In the prospective set of images, the diagnostic performance, time-based cost, and assisted diagnosis were compared between the optimized strategy and the traditional all-AI strategy.

Results

The retrospective set included 1754 ultrasonographic images from 1048 patients (mean [SD] age, 42.1 [13.2] years; 749 women [71.5%]) with 1754 thyroid nodules (mean [SD] size, 16.4 [10.6] mm); 748 nodules (42.6%) were benign, and 1006 (57.4%) were malignant. The prospective set included 300 ultrasonographic images from 268 patients (mean [SD] age, 41.7 [14.1] years; 194 women [72.4%]) with 300 thyroid nodules (mean [SD] size, 17.2 [6.8] mm); 125 nodules (41.7%) were benign, and 175 (58.3%) were malignant. For junior radiologists, the ultrasonographic features that were not improved by AI assistance included cystic or almost completely cystic nodules, anechoic nodules, spongiform nodules, and nodules smaller than 5 mm, whereas for senior radiologists the features that were not improved by AI assistance were cystic or almost completely cystic nodules, anechoic nodules, spongiform nodules, very hypoechoic nodules, nodules taller than wide, lobulated or irregular nodules, and extrathyroidal extension. Compared with the traditional all-AI strategy, the optimized strategy was associated with increased mean task completion times for junior radiologists (reader 11, from 15.2 seconds [95% CI, 13.2-17.2 seconds] to 19.4 seconds [95% CI, 15.6-23.3 seconds]; reader 12, from 12.7 seconds [95% CI, 11.4-13.9 seconds] to 15.6 seconds [95% CI, 13.6-17.7 seconds]), but shorter times for senior radiologists (reader 14, from 19.4 seconds [95% CI, 18.1-20.7 seconds] to 16.8 seconds [95% CI, 15.3-18.3 seconds]; reader 16, from 12.5 seconds [95% CI, 12.1-12.9 seconds] to 10.0 seconds [95% CI, 9.5-10.5 seconds]). There was no significant difference in sensitivity (range, 91%-100%) or specificity (range, 94%-98%) between the 2 strategies for readers 11 to 16.

Conclusions and Relevance

This diagnostic study suggests that an optimized AI strategy in thyroid nodule management may reduce diagnostic time-based costs without sacrificing diagnostic accuracy for senior radiologists, while the traditional all-AI strategy may still be more beneficial for junior radiologists.

Quality control system for mammographic breast positioning using deep learning

This study proposes a deep convolutional neural network (DCNN) classification for the quality control and validation of breast positioning criteria in mammography. A total of 1631 mediolateral oblique mammographic views were collected from an open database. We designed two main steps for mammographic verification: automated detection of the positioning part and classification of three scales that determine the positioning quality using DCNNs. After acquiring labeled mammograms with three scales visually evaluated based on guidelines, the first step was automatically detecting the region of interest of the subject part by image processing. The next step was classifying mammographic positioning accuracy into three scales using four representative DCNNs. The experimental results showed that the DCNN model achieved the best positioning classification accuracy of 0.7836 using VGG16 in the inframammary fold and a classification accuracy of 0.7278 using Xception in the nipple profile. Furthermore, using the softmax function, the breast positioning criteria could be evaluated quantitatively by presenting the predicted value, which is the probability of determining positioning accuracy. The proposed method can be quantitatively evaluated without the need for an individual qualitative evaluation and has the potential to improve the quality control and validation of breast positioning criteria in mammography.

Artificial intelligence in BreastScreen Norway: a retrospective analysis of a cancer-enriched sample including 1254 breast cancer cases

OBJECTIVES

To compare results of selected performance measures in mammographic screening for an artificial intelligence (AI) system versus independent double reading by radiologists.

METHODS

In this retrospective study, we analyzed data from 949 screen-detected breast cancers, 305 interval cancers, and 13,646 negative examinations performed in BreastScreen Norway during the period from 2010 to 2018. An AI system scored the examinations from 1 to 10, based on the risk of malignancy. Results from the AI system were compared to screening results after independent double reading. AI score 10 was set as the threshold. The results were stratified by mammographic density.

RESULTS

A total of 92.7% of the screen-detected and 40.0% of the interval cancers had an AI score of 10. Among women with a negative screening outcome, 9.1% had an AI score of 10. For women with the highest breast density, the AI system scored 100% of the screen-detected cancers and 48.6% of the interval cancers with an AI score of 10, which resulted in a sensitivity of 80.9% for women with the highest breast density for the AI system, compared to 62.8% for independent double reading. For women with screen-detected cancers who had prior mammograms available, 41.9% had an AI score of 10 at the prior screening round.

CONCLUSIONS

The high proportion of cancers with an AI score of 10 indicates a promising performance of the AI system, particularly for women with dense breasts. Results on prior mammograms with AI score 10 illustrate the potential for earlier detection of breast cancers by using AI in screen-reading.

KEY POINTS

• The AI system scored 93% of the screen-detected cancers and 40% of the interval cancers with AI score 10. • The AI system scored all screen-detected cancers and almost 50% of interval cancers among women with the highest breast density with AI score 10. • About 40% of the screen-detected cancers had an AI score of 10 on the prior mammograms, indicating a potential for earlier detection by using AI in screen-reading.

From patterns to patients: Advances in clinical machine learning for cancer diagnosis, prognosis, and treatment

Machine learning (ML) is increasingly used in clinical oncology to diagnose cancers, predict patient outcomes, and inform treatment planning. Here, we review recent applications of ML across the clinical oncology workflow. We review how these techniques are applied to medical imaging and to molecular data obtained from liquid and solid tumor biopsies for cancer diagnosis, prognosis, and treatment design. We discuss key considerations in developing ML for the distinct challenges posed by imaging and molecular data. Finally, we examine ML models approved for cancer-related patient usage by regulatory agencies and discuss approaches to improve the clinical usefulness of ML.

Prioritizing Screening Mammograms for Immediate Interpretation and Diagnostic Evaluation on the Basis of Risk for Recall

PURPOSE

The aim of this study was to develop a prioritization strategy for scheduling immediate screening mammographic interpretation and possible diagnostic evaluation.

METHODS

A population-based cohort with screening mammograms performed from 2012 to 2020 at 126 radiology facilities from 7 Breast Cancer Surveillance Consortium registries was identified. Classification trees identified combinations of clinical history (age, BI-RADS® density, time since prior mammogram, history of false-positive recall or biopsy result), screening modality (digital mammography, digital breast tomosynthesis), and facility characteristics (profit status, location, screening volume, practice type, academic affiliation) that grouped screening mammograms by recall rate, with ≥12/100 considered high and ≥16/100 very high. An efficiency ratio was estimated as the percentage of recalls divided by the percentage of mammograms.

RESULTS

The study cohort included 2,674,051 screening mammograms in 925,777 women, with 235,569 recalls. The most important predictor of recall was time since prior mammogram, followed by age, history of false-positive recall, breast density, history of benign biopsy, and screening modality. Recall rates were very high for baseline mammograms (21.3/100; 95% confidence interval, 19.7-23.0) and high for women with ≥5 years since prior mammogram (15.1/100; 95% confidence interval, 14.3-16.1). The 9.2% of mammograms in subgroups with very high and high recall rates accounted for 19.2% of recalls, an efficiency ratio of 2.1 compared with a random approach. Adding women <50 years of age with dense breasts accounted for 20.3% of mammograms and 33.9% of recalls (efficiency ratio = 1.7). Results including facility-level characteristics were similar.

CONCLUSIONS

Prioritizing women with baseline mammograms or ≥5 years since prior mammogram for immediate interpretation and possible diagnostic evaluation could considerably reduce the number of women needing to return for diagnostic imaging at another visit.

Comparison of Diagnostic Performance in Mammography Assessment: Radiologist with Reference to Clinical Information Versus Standalone Artificial Intelligence Detection

We compared diagnostic performances between radiologists with reference to clinical information and standalone artificial intelligence (AI) detection of breast cancer on digital mammography. This study included 392 women (average age: 57.3 ± 12.1 years, range: 30−94 years) diagnosed with malignancy between January 2010 and June 2021 who underwent digital mammography prior to biopsy. Two radiologists assessed mammographic findings based on clinical symptoms and prior mammography. All mammographies were analyzed via AI. Breast cancer detection performance was compared between radiologists and AI based on how the lesion location was concordant between each analysis method (radiologists or AI) and pathological results. Kappa coefficient was used to measure the concordance between radiologists or AI analysis and pathology results. Binominal logistic regression analysis was performed to identify factors influencing the concordance between radiologists’ analysis and pathology results. Overall, the concordance was higher in radiologists’ diagnosis than on AI analysis (kappa coefficient: 0.819 vs. 0.698). Impact of prior mammography (odds ratio (OR): 8.55, p < 0.001), clinical symptom (OR: 5.49, p < 0.001), and fatty breast density (OR: 5.18, p = 0.008) were important factors contributing to the concordance of lesion location between radiologists’ diagnosis and pathology results.

Beyond fine-tuning: Classifying high resolution mammograms using function-preserving transformations

The task of classifying mammograms is very challenging because the lesion is usually small in the high resolution image. The current state-of-the-art approaches for medical image classification rely on using the de-facto method for convolutional neural networks-fine-tuning. However, there are fundamental differences between natural images and medical images, which based on existing evidence from the literature, limits the overall performance gain when designed with algorithmic approaches. In this paper, we propose to go beyond fine-tuning by introducing a novel framework called MorphHR, in which we highlight a new transfer learning scheme. The idea behind the proposed framework is to integrate function-preserving transformations, for any continuous non-linear activation neurons, to internally regularise the network for improving mammograms classification. The proposed solution offers two major advantages over the existing techniques. Firstly and unlike fine-tuning, the proposed approach allows for modifying not only the last few layers but also several of the first ones on a deep ConvNet. By doing this, we can design the network front to be suitable for learning domain specific features. Secondly, the proposed scheme is scalable to hardware. Therefore, one can fit high resolution images on standard GPU memory. We show that by using high resolution images, one prevents losing relevant information. We demonstrate, through numerical and visual experiments, that the proposed approach yields to a significant improvement in the classification performance over state-of-the-art techniques, and is indeed on a par with radiology experts. Moreover and for generalisation purposes, we show the effectiveness of the proposed learning scheme on another large dataset, the ChestX-ray14, surpassing current state-of-the-art techniques.

Multicenter, Multivendor Validation of an FDA-approved Algorithm for Mammography Triage

OBJECTIVE

Artificial intelligence (AI)-based triage algorithms may improve cancer detection and expedite radiologist workflow. To this end, the performance of a commercial AI-based triage algorithm on screening mammograms was evaluated across breast densities and lesion types.

METHODS

This retrospective, IRB-exempt, multicenter, multivendor study examined 1255 screening 4-view mammograms (400 positive and 855 negative studies). Images were anonymized by providing institutions and analyzed by a commercially available AI algorithm (cmTriage, CureMetrix, La Jolla, CA) that performed retrospective triage at the study level by flagging exams as "suspicious" or not. Sensitivities and specificities with confidence intervals were derived from area under the curve (AUC) calculations.

RESULTS

The algorithm demonstrated an AUC of 0.95 (95% CI: 0.94-0.96) for case identification. Area under the curve held across densities (0.95) and lesion types (masses: 0.94 [95% CI: 0.92-0.96] or microcalcifications: 0.97 [95% CI: 0.96-0.99]). The algorithm has a default sensitivity of 93% (95% CI: 95.6%-90.5%) with specificity of 76.3% (95% CI: 79.2%-73.4%). To evaluate real-world performance, a sensitivity of 86.9% (95% CI: 83.6%-90.2%) was tested, as observed for practicing radiologists by the Breast Cancer Surveillance Consortium (BCSC) study. The resulting specificity was 88.5% (95% CI: 86.4%-90.7%), similar to the BCSC specificity of 88.9%, indicating performance comparable to real-world results.

CONCLUSION

When tested for lesion detection, an AI-based triage software can perform at the level of practicing radiologists. Drawing attention to suspicious exams may improve reader specificity and help streamline radiologist workflow, enabling faster turnaround times and improving care.

Past, Present, and Future of Machine Learning and Artificial Intelligence for Breast Cancer Screening

Breast cancer screening has evolved substantially over the past few decades because of advancements in new image acquisition systems and novel artificial intelligence (AI) algorithms. This review provides a brief overview of the history, current state, and future of AI in breast cancer screening and diagnosis along with challenges involved in the development of AI systems. Although AI has been developing for interpretation tasks associated with breast cancer screening for decades, its potential to combat the subjective nature and improve the efficiency of human image interpretation is always expanding. The rapid advancement of computational power and deep learning has increased greatly in AI research, with promising performance in detection and classification tasks across imaging modalities. Most AI systems, based on human-engineered or deep learning methods, serve as concurrent or secondary readers, that is, as aids to radiologists for a specific, well-defined task. In the future, AI may be able to perform multiple integrated tasks, making decisions at the level of or surpassing the ability of humans. Artificial intelligence may also serve as a partial primary reader to streamline ancillary tasks, triaging cases or ruling out obvious normal cases. However, before AI is used as an independent, autonomous reader, various challenges need to be addressed, including explainability and interpretability, in addition to repeatability and generalizability, to ensure that AI will provide a significant clinical benefit to breast cancer screening across all populations.

Artificial Intelligence-Based Identification of Normal Chest Radiographs: A Simulation Study in a Multicenter Health Screening Cohort

Objective

This study aimed to investigate the feasibility of using artificial intelligence (AI) to identify normal chest radiography (CXR) from the worklist of radiologists in a health-screening environment.

Materials and Methods

This retrospective simulation study was conducted using the CXRs of 5887 adults (mean age ± standard deviation, 55.4 ± 11.8 years; male, 4329) from three health screening centers in South Korea using a commercial AI (Lunit INSIGHT CXR3, version 3.5.8.8). Three board-certified thoracic radiologists reviewed CXR images for referable thoracic abnormalities and grouped the images into those with visible referable abnormalities (identified as abnormal by at least one reader) and those with clearly visible referable abnormalities (identified as abnormal by at least two readers). With AI-based simulated exclusion of normal CXR images, the percentages of normal images sorted and abnormal images erroneously removed were analyzed. Additionally, in a random subsample of 480 patients, the ability to identify visible referable abnormalities was compared among AI-unassisted reading (i.e., all images read by human readers without AI), AI-assisted reading (i.e., all images read by human readers with AI assistance as concurrent readers), and reading with AI triage (i.e., human reading of only those rendered abnormal by AI).

Results

Of 5887 CXR images, 405 (6.9%) and 227 (3.9%) contained visible and clearly visible abnormalities, respectively. With AI-based triage, 42.9% (2354/5482) of normal CXR images were removed at the cost of erroneous removal of 3.5% (14/405) and 1.8% (4/227) of CXR images with visible and clearly visible abnormalities, respectively. In the diagnostic performance study, AI triage removed 41.6% (188/452) of normal images from the worklist without missing visible abnormalities and increased the specificity for some readers without decreasing sensitivity.

Conclusion

This study suggests the feasibility of sorting and removing normal CXRs using AI with a tailored cut-off to increase efficiency and reduce the workload of radiologists.

A deep learning model based on dynamic contrast-enhanced magnetic resonance imaging enables accurate prediction of benign and malignant breast lessons

Objectives

The study aims to investigate the value of a convolutional neural network (CNN) based on dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) in predicting malignancy of breast lesions.

Methods

We developed a CNN model based on DCE-MRI to characterize breast lesions. Between November 2018 and October 2019, 6,165 slices of 364 lesions (234 malignant, 130 benign) in 364 patients were pooled in the training/validation set. Lesions were semi-automatically segmented by two breast radiologists using ITK-SNAP software. The standard of reference was histologic consequences. Algorithm performance was evaluated in an independent testing set of 1,560 slices of 127 lesions in 127 patients using weighted sums of the area under the curve (AUC) scores.

Results

The area under the receiver operating characteristic (ROC) curve was 0.955 for breast cancer prediction while the accuracy, sensitivity, and specificity were 90.3, 96.2, and 79.0%, respectively, in the slice-based method. In the case-based method, the efficiency of the model changed by adjusting the standard for the number of positive slices. When a lesion with three or more positive slices was determined as malignant, the sensitivity was above 90%, with a specificity of nearly 60% and an accuracy higher than 80%.

Conclusion

The CNN model based on DCE-MRI demonstrated high accuracy for predicting malignancy among the breast lesions. This method should be validated in a larger and independent cohort.

Combining the strengths of radiologists and AI for breast cancer screening: a retrospective analysis

BACKGROUND

We propose a decision-referral approach for integrating artificial intelligence (AI) into the breast-cancer screening pathway, whereby the algorithm makes predictions on the basis of its quantification of uncertainty. Algorithmic assessments with high certainty are done automatically, whereas assessments with lower certainty are referred to the radiologist. This two-part AI system can triage normal mammography exams and provide post-hoc cancer detection to maintain a high degree of sensitivity. This study aimed to evaluate the performance of this AI system on sensitivity and specificity when used either as a standalone system or within a decision-referral approach, compared with the original radiologist decision.

METHODS

We used a retrospective dataset consisting of 1 193 197 full-field, digital mammography studies carried out between Jan 1, 2007, and Dec 31, 2020, from eight screening sites participating in the German national breast-cancer screening programme. We derived an internal-test dataset from six screening sites (1670 screen-detected cancers and 19 997 normal mammography exams), and an external-test dataset of breast cancer screening exams (2793 screen-detected cancers and 80 058 normal exams) from two additional screening sites to evaluate the performance of an AI algorithm on sensitivity and specificity when used either as a standalone system or within a decision-referral approach, compared with the original individual radiologist decision at the point-of-screen reading ahead of the consensus conference. Different configurations of the AI algorithm were evaluated. To account for the enrichment of the datasets caused by oversampling cancer cases, weights were applied to reflect the actual distribution of study types in the screening programme. Triaging performance was evaluated as the rate of exams correctly identified as normal. Sensitivity across clinically relevant subgroups, screening sites, and device manufacturers was compared between standalone AI, the radiologist, and decision referral. We present receiver operating characteristic (ROC) curves and area under the ROC (AUROC) to evaluate AI-system performance over its entire operating range. Comparison with radiologists and subgroup analysis was based on sensitivity and specificity at clinically relevant configurations.

FINDINGS

The exemplary configuration of the AI system in standalone mode achieved a sensitivity of 84·2% (95% CI 82·4-85·8) and a specificity of 89·5% (89·0-89·9) on internal-test data, and a sensitivity of 84·6% (83·3-85·9) and a specificity of 91·3% (91·1-91·5) on external-test data, but was less accurate than the average unaided radiologist. By contrast, the simulated decision-referral approach significantly improved upon radiologist sensitivity by 2·6 percentage points and specificity by 1·0 percentage points, corresponding to a triaging performance at 63·0% on the external dataset; the AUROC was 0·982 (95% CI 0·978-0·986) on the subset of studies assessed by AI, surpassing radiologist performance. The decision-referral approach also yielded significant increases in sensitivity for a number of clinically relevant subgroups, including subgroups of small lesion sizes and invasive carcinomas. Sensitivity of the decision-referral approach was consistent across the eight included screening sites and three device manufacturers.

INTERPRETATION

The decision-referral approach leverages the strengths of both the radiologist and AI, demonstrating improvements in sensitivity and specificity surpassing that of the individual radiologist and of the standalone AI system. This approach has the potential to improve the screening accuracy of radiologists, is adaptive to the requirements of screening, and could allow for the reduction of workload ahead of the consensus conference, without discarding the generalised knowledge of radiologists.

FUNDING

Vara.

The Economic Impact of AI on Breast Imaging

This article explores the development of computer-aided detection (CAD) and artificial or augmented intelligence (AI) for breast radiology examinations and describes the current applications of AI in breast imaging. Although radiologists in other subspecialties may be less familiar with the use of these technologies in their practices, CAD has been used in breast imaging for more than two decades, as mammography CAD programs have been commercially available in the United States since the late 1990s. Likewise, breast radiologists have seen payment for CAD in mammography and breast MRI evolve over time. With the rapid expansion of AI products in radiology in recent years, many new applications for these technologies in breast imaging have emerged. This article outlines the current state of reimbursement for breast radiology AI algorithms within the traditional fee-for-service model used by Medicare and commercial insurers as well as potential future payment pathways. In addition, the inherent challenges of employing the existing payment framework in the United States to AI programs in radiology are detailed for the reader. This article aims to give breast radiologists a better understanding of how AI will be reimbursed as they seek to further incorporate these emerging technologies into their practices to advance patient care and improve workflow efficiency.

Using deep learning to safely exclude lesions with only ultrafast breast MRI to shorten acquisition and reading time

OBJECTIVES

To investigate the feasibility of automatically identifying normal scans in ultrafast breast MRI with artificial intelligence (AI) to increase efficiency and reduce workload.

METHODS

In this retrospective analysis, 837 breast MRI examinations performed on 438 women from April 2016 to October 2019 were included. The left and right breasts in each examination were labelled normal (without suspicious lesions) or abnormal (with suspicious lesions) based on final interpretation. Maximum intensity projection (MIP) images of each breast were then used to train a deep learning model. A high sensitivity threshold was calculated based on the detection trade - off (DET) curve on the validation set. The performance of the model was evaluated by receiver operating characteristic analysis of the independent test set. The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) with the high sensitivity threshold were calculated.

RESULTS

The independent test set consisted of 178 examinations of 149 patients (mean age, 44 years ± 14 [standard deviation]). The trained model achieved an AUC of 0.81 (95% CI: 0.75-0.88) on the independent test set. Applying a threshold of 0.25 yielded a sensitivity of 98% (95% CI: 90%; 100%), an NPV of 98% (95% CI: 89%; 100%), a workload reduction of 15.7%, and a scan time reduction of 16.6%.

CONCLUSION

This deep learning model has a high potential to help identify normal scans in ultrafast breast MRI and thereby reduce radiologists' workload and scan time.

KEY POINTS

• Deep learning in TWIST may eliminate the necessity of additional sequences for identifying normal breasts during MRI screening. • Workload and scanning time reductions of 15.7% and 16.6%, respectively, could be achieved with the cost of 1 (1 of 55) false negative prediction.

Deep learning in breast imaging

Millions of breast imaging exams are performed each year in an effort to reduce the morbidity and mortality of breast cancer. Breast imaging exams are performed for cancer screening, diagnostic work-up of suspicious findings, evaluating extent of disease in recently diagnosed breast cancer patients, and determining treatment response. Yet, the interpretation of breast imaging can be subjective, tedious, time-consuming, and prone to human error. Retrospective and small reader studies suggest that deep learning (DL) has great potential to perform medical imaging tasks at or above human-level performance, and may be used to automate aspects of the breast cancer screening process, improve cancer detection rates, decrease unnecessary callbacks and biopsies, optimize patient risk assessment, and open up new possibilities for disease prognostication. Prospective trials are urgently needed to validate these proposed tools, paving the way for real-world clinical use. New regulatory frameworks must also be developed to address the unique ethical, medicolegal, and quality control issues that DL algorithms present. In this article, we review the basics of DL, describe recent DL breast imaging applications including cancer detection and risk prediction, and discuss the challenges and future directions of artificial intelligence-based systems in the field of breast cancer.

Anomaly Detection of Calcifications in Mammography Based on 11,000 Negative Cases

In mammography, calcifications are one of the most common signs of breast cancer. Detection of such lesions is an active area of research for computer-aided diagnosis and machine learning algorithms. Due to limited numbers of positive cases, many supervised detection models suffer from overfitting and fail to generalize. We present a one-class, semi-supervised framework using a deep convolutional autoencoder trained with over 50,000 images from 11,000 negative-only cases. Since the model learned from only normal breast parenchymal features, calcifications produced large signals when comparing the residuals between input and reconstruction output images. As a key advancement, a structural dissimilarity index was used to suppress non-structural noises. Our selected model achieved pixel-based AUROC of 0.959 and AUPRC of 0.676 during validation, where calcification masks were defined in a semi-automated process. Although not trained directly on any cancers, detection performance of calcification lesions on 1,883 testing images (645 malignant and 1238 negative) achieved 75% sensitivity at 2.5 false positives per image. Performance plateaued early when trained with only a fraction of the cases, and greater model complexity or a larger dataset did not improve performance. This study demonstrates the potential of this anomaly detection approach to detect mammographic calcifications in a semi-supervised manner with efficient use of a small number of labeled images, and may facilitate new clinical applications such as computer-aided triage and quality improvement.

An Artificial Intelligence-based Mammography Screening Protocol for Breast Cancer: Outcome and Radiologist Workload

Background Developments in artificial intelligence (AI) systems to assist radiologists in reading mammograms could improve breast cancer screening efficiency. Purpose To investigate whether an AI system could detect normal, moderate-risk, and suspicious mammograms in a screening sample to safely reduce radiologist workload and evaluate across Breast Imaging Reporting and Data System (BI-RADS) densities. Materials and Methods This retrospective simulation study analyzed mammographic examination data consecutively collected from January 2014 to December 2015 in the Danish Capital Region breast cancer screening program. All mammograms were scored from 0 to 10, representing the risk of malignancy, using an AI tool. During simulation, normal mammograms (score < 5) would be excluded from radiologist reading and suspicious mammograms (score > recall threshold [RT]) would be recalled. Two radiologists read the remaining mammograms. The RT was fitted using another independent cohort (same institution) by matching to the radiologist sensitivity. This protocol was further applied to each BI-RADS density. Screening outcomes were measured using the sensitivity, specificity, workload, and false-positive rate. The AI-based screening was tested for noninferiority sensitivity compared with radiologist screening using the Farrington-Manning test. Specificities were compared using the McNemar test. Results The study sample comprised 114 421 screenings for breast cancer in 114 421 women, resulting in 791 screen-detected, 327 interval, and 1473 long-term cancers and 2107 false-positive screenings. The mean age of the women was 59 years ± 6 (SD). The AI-based screening sensitivity was 69.7% (779 of 1118; 95% CI: 66.9, 72.4) and was noninferior (P = .02) to the radiologist screening sensitivity of 70.8% (791 of 1118; 95% CI: 68.0, 73.5). The AI-based screening specificity was 98.6% (111 725 of 113 303; 95% CI: 98.5, 98.7), which was higher (P < .001) than the radiologist specificity of 98.1% (111 196 of 113 303; 95% CI: 98.1, 98.2). The radiologist workload was reduced by 62.6% (71 585 of 114 421), and 25.1% (529 of 2107) of false-positive screenings were avoided. Screening results were consistent across BI-RADS densities, although not significantly so for sensitivity. Conclusion Artificial intelligence (AI)-based screening could detect normal, moderate-risk, and suspicious mammograms in a breast cancer screening program, which may reduce the radiologist workload. AI-based screening performed consistently across breast densities. © RSNA, 2022 Online supplemental material is available for this article.

Artificial Intelligence for Reducing Workload in Breast Cancer Screening with Digital Breast Tomosynthesis

Background Digital breast tomosynthesis (DBT) has higher diagnostic accuracy than digital mammography, but interpretation time is substantially longer. Artificial intelligence (AI) could improve reading efficiency. Purpose To evaluate the use of AI to reduce workload by filtering out normal DBT screens. Materials and Methods The retrospective study included 13 306 DBT examinations from 9919 women performed between June 2013 and November 2018 from two health care networks. The cohort was split into training, validation, and test sets (3948, 1661, and 4310 women, respectively). A workflow was simulated in which the AI model classified cancer-free examinations that could be dismissed from the screening worklist and used the original radiologists' interpretations on the rest of the worklist examinations. The AI system was also evaluated with a reader study of five breast radiologists reading the DBT mammograms of 205 women. The area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and recall rate were evaluated in both studies. Statistics were computed across 10 000 bootstrap samples to assess 95% CIs, noninferiority, and superiority tests. Results The model was tested on 4310 screened women (mean age, 60 years ± 11 [standard deviation]; 5182 DBT examinations). Compared with the radiologists' performance (417 of 459 detected cancers [90.8%], 477 recalls in 5182 examinations [9.2%]), the use of AI to automatically filter out cases would result in 39.6% less workload, noninferior sensitivity (413 of 459 detected cancers; 90.0%; P = .002), and 25% lower recall rate (358 recalls in 5182 examinations; 6.9%; P = .002). In the reader study, AUC was higher in the standalone AI compared with the mean reader (0.84 vs 0.81; P = .002). Conclusion The artificial intelligence model was able to identify normal digital breast tomosynthesis screening examinations, which decreased the number of examinations that required radiologist interpretation in a simulated clinical workflow. Published under a CC BY 4.0 license. Online supplemental material is available for this article. See also the editorial by Philpotts in this issue.

Application of Deep Learning in Breast Cancer Imaging

This review gives an overview of the current state of deep learning research in breast cancer imaging. Breast imaging plays a major role in detecting breast cancer at an earlier stage, as well as monitoring and evaluating breast cancer during treatment. The most commonly used modalities for breast imaging are digital mammography, digital breast tomosynthesis, ultrasound and magnetic resonance imaging. Nuclear medicine imaging techniques are used for detection and classification of axillary lymph nodes and distant staging in breast cancer imaging. All of these techniques are currently digitized, enabling the possibility to implement deep learning (DL), a subset of Artificial intelligence, in breast imaging. DL is nowadays embedded in a plethora of different tasks, such as lesion classification and segmentation, image reconstruction and generation, cancer risk prediction, and prediction and assessment of therapy response. Studies show similar and even better performances of DL algorithms compared to radiologists, although it is clear that large trials are needed, especially for ultrasound and magnetic resonance imaging, to exactly determine the added value of DL in breast cancer imaging. Studies on DL in nuclear medicine techniques are only sparsely available and further research is mandatory. Legal and ethical issues need to be considered before the role of DL can expand to its full potential in clinical breast care practice.