Is single reading with computer-aided detection (CAD) as good as double reading in mammography screening? A systematic review

Detection of breast cancer in digital breast tomosynthesis with vision transformers

Digital Breast Tomosynthesis (DBT) has revolutionized more traditional breast imaging through its three-dimensional (3D) visualization capability that significantly enhances lesion discernibility, reduces tissue overlap, and improves diagnostic precision as compared to conventional two-dimensional (2D) mammography. In this study, we propose an advanced Computer-Aided Detection (CAD) system that harnesses the power of vision transformers to augment DBT's diagnostic efficiency. This scheme uses a neural network to glean attributes from the 2D slices of DBT followed by post-processing that considers features from neighboring slices to categorize the entire 3D scan. By leveraging a transfer learning technique, we trained and validated our CAD framework on a unique dataset consisting of 3,831 DBT scans and subsequently tested it on 685 scans. Of the architectures tested, the Swin Transformer outperformed the ResNet101 and vanilla Vision Transformer. It achieved an impressive AUC score of 0.934 ± 0.026 at a resolution of 384 × 384. Increasing the image resolution from 224 to 384 not only maintained vital image attributes but also led to a marked improvement in performance (p-value = 0.0003). The Mean Teacher algorithm, a semi-supervised method using both labeled and unlabeled DBT slices, showed no significant improvement over the supervised approach. Comprehensive analyses across different lesion types, sizes, and patient ages revealed consistent performance. The integration of attention mechanisms yielded a visual narrative of the model's decision-making process that highlighted the prioritized regions during assessments. These findings should significantly propel the methodologies employed in DBT image analysis by setting a new benchmark for breast cancer diagnostic precision.

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.

The use of artificial intelligence tools in cancer detection compared to the traditional diagnostic imaging methods: An overview of the systematic reviews

BACKGROUND AND PURPOSE

In comparison to conventional medical imaging diagnostic modalities, the aim of this overview article is to analyze the accuracy of the application of Artificial Intelligence (AI) techniques in the identification and diagnosis of malignant tumors in adult patients.

DATA SOURCES

The acronym PIRDs was used and a comprehensive literature search was conducted on PubMed, Cochrane, Scopus, Web of Science, LILACS, Embase, Scielo, EBSCOhost, and grey literature through Proquest, Google Scholar, and JSTOR for systematic reviews of AI as a diagnostic model and/or detection tool for any cancer type in adult patients, compared to the traditional diagnostic radiographic imaging model. There were no limits on publishing status, publication time, or language. For study selection and risk of bias evaluation, pairs of reviewers worked separately.

RESULTS

In total, 382 records were retrieved in the databases, 364 after removing duplicates, 32 satisfied the full-text reading criterion, and 09 papers were considered for qualitative synthesis. Although there was heterogeneity in terms of methodological aspects, patient differences, and techniques used, the studies found that several AI approaches are promising in terms of specificity, sensitivity, and diagnostic accuracy in the detection and diagnosis of malignant tumors. When compared to other machine learning algorithms, the Super Vector Machine method performed better in cancer detection and diagnosis. Computer-assisted detection (CAD) has shown promising in terms of aiding cancer detection, when compared to the traditional method of diagnosis.

CONCLUSIONS

The detection and diagnosis of malignant tumors with the help of AI seems to be feasible and accurate with the use of different technologies, such as CAD systems, deep and machine learning algorithms and radiomic analysis when compared with the traditional model, although these technologies are not capable of to replace the professional radiologist in the analysis of medical images. Although there are limitations regarding the generalization for all types of cancer, these AI tools might aid professionals, serving as an auxiliary and teaching tool, especially for less trained professionals. Therefore, further longitudinal studies with a longer follow-up duration are required for a better understanding of the clinical application of these artificial intelligence systems.

TRIAL REGISTRATION

Systematic review registration. Prospero registration number: CRD42022307403.

Artificial intelligence in cancer diagnosis and therapy: Current status and future perspective

Artificial intelligence (AI) in healthcare plays a pivotal role in combating many fatal diseases, such as skin, breast, and lung cancer. AI is an advanced form of technology that uses mathematical-based algorithmic principles similar to those of the human mind for cognizing complex challenges of the healthcare unit. Cancer is a lethal disease with many etiologies, including numerous genetic and epigenetic mutations. Cancer being a multifactorial disease is difficult to be diagnosed at an early stage. Therefore, genetic variations and other leading factors could be identified in due time through AI and machine learning (ML). AI is the synergetic approach for mining the drug targets, their mechanism of action, and drug-organism interaction from massive raw data. This synergetic approach is also facing several challenges in data mining but computational algorithms from different scientific communities for multi-target drug discovery are highly helpful to overcome the bottlenecks in AI for drug-target discovery. AI and ML could be the epicenter in the medical world for the diagnosis, treatment, and evaluation of almost any disease in the near future. In this comprehensive review, we explore the immense potential of AI and ML when integrated with the biological sciences, specifically in the context of cancer research. Our goal is to illuminate the many ways in which AI and ML are being applied to the study of cancer, from diagnosis to individualized treatment. We highlight the prospective role of AI in supporting oncologists and other medical professionals in making informed decisions and improving patient outcomes by examining the intersection of AI and cancer control. Although AI-based medical therapies show great potential, many challenges must be overcome before they can be implemented in clinical practice. We critically assess the current hurdles and provide insights into the future directions of AI-driven approaches, aiming to pave the way for enhanced cancer interventions and improved patient care.

Measurement of malignant spiculated mass lesions on mammogram: Do we include the length of the spicules?

AIM

The aim of this study is to determine if the core size or size with spicules has a better correlation with the final histologic size of spiculated mass lesions.

METHODS

A retrospective study of 48-month duration from January 2014 to December 2017 of biopsy-proven invasive ductal carcinoma presenting as spiculated mass lesions on mammogram was conducted.

RESULTS

There were 195 patients in the study. The mean of the core size was 16.6 mm; when spicules were included the mean size was 27.4mm and final histologic size 21.1 mm. Using unpaired Student 't' test difference in the means was statistically significant (p<0.0001). Pearson number (R) core size versus final histologic size was 0.535 (P < 0.001) and for size with spicules versus final histologic size was 0.495 (P < 0.001).

CONCLUSION

Our study demonstrated that the core size has a stronger positive correlation to final histologic size and should be used preoperatively in decision-making about surgery.

Framing the fallibility of Computer-Aided Detection aids cancer detection

Computer-Aided Detection (CAD) has been proposed to help operators search for cancers in mammograms. Previous studies have found that although accurate CAD leads to an improvement in cancer detection, inaccurate CAD leads to an increase in both missed cancers and false alarms. This is known as the over-reliance effect. We investigated whether providing framing statements of CAD fallibility could keep the benefits of CAD while reducing over-reliance. In Experiment 1, participants were told about the benefits or costs of CAD, prior to the experiment. Experiment 2 was similar, except that participants were given a stronger warning and instruction set in relation to the costs of CAD. The results showed that although there was no effect of framing in Experiment 1, a stronger message in Experiment 2 led to a reduction in the over-reliance effect. A similar result was found in Experiment 3 where the target had a lower prevalence. The results show that although the presence of CAD can result in over-reliance on the technology, these effects can be mitigated by framing and instruction sets in relation to CAD fallibility.

Development and validation of a deep learning model for detection of breast cancers in mammography from multi-institutional datasets

Objectives

The objective of this study was to develop and validate a state-of-the-art, deep learning (DL)-based model for detecting breast cancers on mammography.

Methods

Mammograms in a hospital development dataset, a hospital test dataset, and a clinic test dataset were retrospectively collected from January 2006 through December 2017 in Osaka City University Hospital and Medcity21 Clinic. The hospital development dataset and a publicly available digital database for screening mammography (DDSM) dataset were used to train and to validate the RetinaNet, one type of DL-based model, with five-fold cross-validation. The model’s sensitivity and mean false positive indications per image (mFPI) and partial area under the curve (AUC) with 1.0 mFPI for both test datasets were externally assessed with the test datasets.

Results

The hospital development dataset, hospital test dataset, clinic test dataset, and DDSM development dataset included a total of 3179 images (1448 malignant images), 491 images (225 malignant images), 2821 images (37 malignant images), and 1457 malignant images, respectively. The proposed model detected all cancers with a 0.45–0.47 mFPI and had partial AUCs of 0.93 in both test datasets.

Conclusions

The DL-based model developed for this study was able to detect all breast cancers with a very low mFPI. Our DL-based model achieved the highest performance to date, which might lead to improved diagnosis for breast cancer.

The optimal use of computer aided detection to find low prevalence cancers

People miss a high proportion of targets that only appear rarely. This low prevalence (LP) effect has implications for applied search tasks such as the clinical reading of mammograms. Computer aided detection (CAD) has been used to help radiologists search mammograms by highlighting areas likely to contain a cancer. Previous research has found a benefit in search when CAD cues were correct but a cost to search when CAD cues were incorrect. The current research investigated whether there is an optimal way to present CAD to ensure low error rates when CAD is both correct and incorrect. Experiment 1 compared an automatic condition, where CAD appeared simultaneously with the display to an interactive condition, where participants could choose to use CAD. Experiment 2 compared the automatic condition to a confirm condition, where participants searched the display first before being shown the CAD cues. The results showed that miss errors were reduced overall in the confirm condition, with no cost to false alarms. Furthermore, having CAD be interactive, resulted in a low uptake where it was only used in 34% of trials. The results showed that the presentation mode of CAD can affect decision-making in LP search.

SCREENING MAMMOGRAPHY: DIAGNOSTIC EFFICACY-ISSUES AND CONSIDERATIONS FOR THE 2020S

Diagnostic efficacy in medical imaging is ultimately a reflection of radiologist performance. This can be influenced by numerous factors, some of which are patient related, such as the physical size and density of the breast, and machine related, where some lesions are difficult to visualise on traditional imaging techniques. Other factors are human reader errors that occur during the diagnostic process, which relate to reader experience and their perceptual and cognitive oversights. Given the large-scale nature of breast cancer screening, even small increases in diagnostic performance equate to large numbers of women saved. It is important to identify the causes of diagnostic errors and how detection efficacy can be improved. This narrative review will therefore explore the various factors that influence mammographic performance and the potential solutions used in an attempt to ameliorate the errors made.

Legal and Regulatory Framework for AI Solutions in Healthcare in EU, US, China, and Russia: New Scenarios after a Pandemic

The COVID-19 crisis has exposed some of the most pressing challenges affecting healthcare and highlighted the benefits that robust integration of digital and AI technologies in the healthcare setting may bring. Although medical solutions based on AI are growing rapidly, regulatory issues and policy initiatives including ownership and control of data, data sharing, privacy protection, telemedicine, and accountability need to be carefully and continually addressed as AI research requires robust and ethical guidelines, demanding an update of the legal and regulatory framework all over the world. Several recently proposed regulatory frameworks provide a solid foundation but do not address a number of issues that may prevent algorithms from being fully trusted. A global effort is needed for an open, mature conversation about the best possible way to guard against and mitigate possible harms to realize the potential of AI across health systems in a respectful and ethical way. This conversation must include national and international policymakers, physicians, digital health and machine learning leaders from industry and academia. If this is done properly and in a timely fashion, the potential of AI in healthcare will be realized.

Clinical applications of artificial intelligence and machine learning in cancer diagnosis: looking into the future

Artificial intelligence (AI) is the use of mathematical algorithms to mimic human cognitive abilities and to address difficult healthcare challenges including complex biological abnormalities like cancer. The exponential growth of AI in the last decade is evidenced to be the potential platform for optimal decision-making by super-intelligence, where the human mind is limited to process huge data in a narrow time range. Cancer is a complex and multifaced disorder with thousands of genetic and epigenetic variations. AI-based algorithms hold great promise to pave the way to identify these genetic mutations and aberrant protein interactions at a very early stage. Modern biomedical research is also focused to bring AI technology to the clinics safely and ethically. AI-based assistance to pathologists and physicians could be the great leap forward towards prediction for disease risk, diagnosis, prognosis, and treatments. Clinical applications of AI and Machine Learning (ML) in cancer diagnosis and treatment are the future of medical guidance towards faster mapping of a new treatment for every individual. By using AI base system approach, researchers can collaborate in real-time and share knowledge digitally to potentially heal millions. In this review, we focused to present game-changing technology of the future in clinics, by connecting biology with Artificial Intelligence and explain how AI-based assistance help oncologist for precise treatment.

WHAT CAN WE ACTUALLY SEE USING COMPUTER AIDED DETECTION IN MAMMOGRAPHY?

The main goal of this study was to compare the results of computer aided detection (CAD) analysis in screening mammography with the results independently obtained by two radiologists for the same samples and to determine the sensitivity and specificity of CAD for breast lesions. A total of 436 mammograms were analyzed with CAD. For each screening mammogram, the changes in breast tissue recognized by CAD were compared to the interpretations of two radiologists. The sensitivity and specificity of CAD for breast lesions were calculated using contingency table. The sensitivity of CAD for all lesions was 54% and specificity 16%. CAD sensitivity for suspicious lesions only was 86%. CAD sensitivity for microcalcifications was 100% and specificity 45%. CAD mainly ‘mistook’ glandular parenchyma, connective tissue and blood vessels for breast lesions, and blood vessel calcifications and axillary folds for microcalcifications. In this study, we confirmed CAD as an excellent tool for recognizing microcalcifications with 100% sensitivity. However, it should not be used as a stand-alone tool in breast screening mammography due to the high rate of false-positive results.

A machine learning approach on multiscale texture analysis for breast microcalcification diagnosis

Background

Screening programs use mammography as primary diagnostic tool for detecting breast cancer at an early stage. The diagnosis of some lesions, such as microcalcifications, is still difficult today for radiologists. In this paper, we proposed an automatic binary model for discriminating tissue in digital mammograms, as support tool for the radiologists. In particular, we compared the contribution of different methods on the feature selection process in terms of the learning performances and selected features.

Results

For each ROI, we extracted textural features on Haar wavelet decompositions and also interest points and corners detected by using Speeded Up Robust Feature (SURF) and Minimum Eigenvalue Algorithm (MinEigenAlg). Then a Random Forest binary classifier is trained on a subset of a sub-set features selected by two different kinds of feature selection techniques, such as filter and embedded methods. We tested the proposed model on 260 ROIs extracted from digital mammograms of the BCDR public database. The best prediction performance for the normal/abnormal and benign/malignant problems reaches a median AUC value of 98.16% and 92.08%, and an accuracy of 97.31% and 88.46%, respectively. The experimental result was comparable with related work performance.

Conclusions

The best performing result obtained with embedded method is more parsimonious than the filter one. The SURF and MinEigen algorithms provide a strong informative content useful for the characterization of microcalcification clusters.

Digital Analysis in Breast Imaging

Breast imaging is a multimodal approach that plays an essential role in the diagnosis of breast cancer. Mammography, sonography, magnetic resonance, and image-guided biopsy are imaging techniques used to search for malignant changes in the breast or precursors of malignant changes in, e.g., screening programs or follow-ups after breast cancer treatment. However, these methods still have some disadvantages such as interobserver variability and the mammography sensitivity in women with radiologically dense breasts. In order to overcome these difficulties and decrease the number of false positive findings, improvements in imaging analysis with the help of artificial intelligence are constantly being developed and tested. In addition, the extraction and correlation of imaging features with special tumor characteristics and genetics of the patients in order to get more information about treatment response, prognosis, and also cancer risk are coming more and more in focus. The aim of this review is to address recent developments in digital analysis of images and demonstrate their potential value in multimodal breast imaging.

Performance and Reading Time of Automated Breast US with or without Computer-aided Detection

BackgroundComputer-aided detection (CAD) systems may be used to help radiologists interpret automated breast (AB) US images. However, the optimal use of CAD with AB US has, to the knowledge of the authors, not been determined.PurposeTo compare the performance and reading time of different readers by using AB US CAD system to detect breast cancer in different reading modes.Materials and MethodsIn this retrospective study, 1485 AB US images (282 with malignant lesions, 695 with benign lesions, and 508 healthy) in 1452 women (mean age, 43.7 years; age range, 19-82 years) including 529 (36.4%) women who were asymptomatic were collected between 2016 and 2017. A CAD system was used to interpret the images. Three novice readers with 1-3 years of US experience and three experienced readers with 5-10 years of US experience were assigned to read AB US images without CAD, at a second reading (after the reader completed a full unaided interpretation), and at concurrent reading (use of CAD at the start of the assessment). Diagnostic performances and reading times were compared by using analysis of variance.ResultsFor all readers, the mean area under the receiver operating characteristic curve improved from 0.88 (95% confidence interval [CI]: 0.85, 0.91) at without-CAD mode to 0.91 (95% CI: 0.89, 0.92; P < .001) at the second-reading mode and 0.90 (95% CI: 0.89, 0.92; P = .002) at the concurrent-reading mode. The mean sensitivity of novice readers in women who were asymptomatic improved from 67% (95% CI: 63%, 74%) at without-CAD mode to 88% (95% CI: 84%, 89%) at both the second-reading mode and the concurrent-reading mode (P = .003). Compared with the without-CAD and second-reading modes, the mean reading time per volume of concurrent reading was 16 seconds (95% CI: 11, 22; P < .001) and 27 seconds (95% CI: 21, 32; P < .001) shorter, respectively.ConclusionComputer-aided detection (CAD) was helpful for novice readers to improve cancer detection at automated breast US in women who were asymptomatic. CAD was more efficient when used concurrently for all readers.© RSNA, 2019Online supplemental material is available for this article.See also the editorial by Slanetz in this issue.

Benefits of Independent Double Reading in Digital Mammography: A Theoretical Evaluation of All Possible Pairing Methodologies

RATIONALE AND OBJECTIVES

To establish the efficacy of pairing readers randomly and evaluate the merits of developing optimal pairing methodologies.

MATERIALS AND METHODS

Sensitivity, specificity, and proportion correct were computed for three different case sets that were independently read by 16 radiologists. Performance of radiologists as single readers was compared to expected double reading performance. We theoretically evaluated all possible pairing methodologies. Bootstrap resampling methods were used for statistical analyses.

RESULTS

Significant improvements in expected performance for double versus single reading (ie, delta performance) were shown for all performance measures and case-sets (p ≤ .003), with overall delta performance across all theoretically possible pairing schemes (n = 10,395) ranging between .05 and .08. Delta performance for the 20 best pairing schemes was significant (p < .001) and ranged between .07 and .10. Delta performance for 20 random pairing schemes was also significant (p ≤ .003) and ranged between .05 and .08. Delta performance for the 20 worst pairing schemes ranged between .03 and .06, reaching significance in delta proportion correct (p ≤ .021) for all three case-sets and in delta specificity for two case-sets (p ≤ .033) but not for a third case-set (p = .131), and not reaching significance in delta sensitivity for any of the three case-sets (.098 ≥ p ≥ .067).

CONCLUSION

Significant benefits accrue from double reading, and while random reader pairing achieves most double reading benefits, a strategic pairing approach may maximize the benefits of double reading.

A new approach for detecting abnormalities in mammograms using a computer-aided windowing system based on Otsu's method

Breast cancer is the most common cancer and the leading cause of cancer deaths in women worldwide. This study aimed to provide an automatic windowing method in mammograms, based on the principles of Otsu's thresholding function, to help radiologists more easily detect abnormalities on mammograms. A total of 322 mammographic images from the Mammographic Image Analysis Society (MIAS) database were used in the present study. The image background was removed based on Otsu's method. After selecting the threshold in the computer-aided windowing (CAW) system, the pixel values were kept larger than the threshold and displayed on a grayscale. A radiologist evaluated images randomly before and after CAW. Using CAW, the radiologist correctly diagnosed all healthy images (207 images). A total of 115 mammograms were evaluated to differentiate malignancy from benign masses. All 63 benign images were accurately diagnosed after using CAW. Moreover, of 52 malignant images, all were accurately recognized as malignant except one, which was recognized as benign. Therefore, specificity and sensitivity were significantly improved to 98% and 99.6%, respectively, and the area under the receiver operating characteristic (ROC) curve was calculated to be 0.99. The study showed that the use of CAW can potentially lead to quicker image assessment and improve the diagnostic accuracy of radiologists in differentiating between benign and malignant masses on mammograms.

Governance of automated image analysis and artificial intelligence analytics in healthcare

The hype over artificial intelligence (AI) has spawned claims that clinicians (particularly radiologists) will become redundant. It is still moot as to whether AI will replace radiologists in day-to-day clinical practice, but more AI applications are expected to be incorporated into the workflows in the foreseeable future. These applications could produce significant ethical and legal issues in healthcare if they cause abrupt disruptions to its contextual integrity and relational dynamics. Sustaining trust and trustworthiness is a key goal of governance, which is necessary to promote collaboration among all stakeholders and to ensure the responsible development and implementation of AI in radiology and other areas of clinical work. In this paper, the nature of AI governance in biomedicine is discussed along with its limitations. It is argued that radiologists must assume a more active role in propelling medicine into the digital age. In this respect, professional responsibilities include inquiring into the clinical and social value of AI, alleviating deficiencies in technical knowledge in order to facilitate ethical evaluation, supporting the recognition, and removal of biases, engaging the "black box" obstacle, and brokering a new social contract on informational use and security. In essence, a much closer integration of ethics, laws, and good practices is needed to ensure that AI governance achieves its normative goals.

The efficacy of using computer-aided detection (CAD) for detection of breast cancer in mammography screening: a systematic review

BACKGROUND

Early detection of breast cancer (BC) is crucial in lowering the mortality.

PURPOSE

To present an overview of studies concerning computer-aided detection (CAD) in screening mammography for early detection of BC and compare diagnostic accuracy and recall rates (RR) of single reading (SR) with SR + CAD and double reading (DR) with SR + CAD.

MATERIAL AND METHODS

PRISMA guidelines were used as a review protocol. Articles on clinical trials concerning CAD for detection of BC in a screening population were included. The literature search resulted in 1522 records. A total of 1491 records were excluded by abstract and 18 were excluded by full text reading. A total of 13 articles were included.

RESULTS

All but two studies from the SR vs. SR + CAD group showed an increased sensitivity and/or cancer detection rate (CDR) when adding CAD. The DR vs. SR + CAD group showed no significant differences in sensitivity and CDR. Adding CAD to SR increased the RR and decreased the specificity in all but one study. For the DR vs. SR + CAD group only one study reported a significant difference in RR.

CONCLUSION

All but two studies showed an increase in RR, sensitivity and CDR when adding CAD to SR. Compared to DR no statistically significant differences in sensitivity or CDR were reported. Additional studies based on organized population-based screening programs, with longer follow-up time, high-volume readers, and digital mammography are needed to evaluate the efficacy of CAD.

Detection of Breast Cancer with Mammography: Effect of an Artificial Intelligence Support System

Purpose To compare breast cancer detection performance of radiologists reading mammographic examinations unaided versus supported by an artificial intelligence (AI) system. Materials and Methods An enriched retrospective, fully crossed, multireader, multicase, HIPAA-compliant study was performed. Screening digital mammographic examinations from 240 women (median age, 62 years; range, 39-89 years) performed between 2013 and 2017 were included. The 240 examinations (100 showing cancers, 40 leading to false-positive recalls, 100 normal) were interpreted by 14 Mammography Quality Standards Act-qualified radiologists, once with and once without AI support. The readers provided a Breast Imaging Reporting and Data System score and probability of malignancy. AI support provided radiologists with interactive decision support (clicking on a breast region yields a local cancer likelihood score), traditional lesion markers for computer-detected abnormalities, and an examination-based cancer likelihood score. The area under the receiver operating characteristic curve (AUC), specificity and sensitivity, and reading time were compared between conditions by using mixed-models analysis dof variance and generalized linear models for multiple repeated measurements. Results On average, the AUC was higher with AI support than with unaided reading (0.89 vs 0.87, respectively; P = .002). Sensitivity increased with AI support (86% [86 of 100] vs 83% [83 of 100]; P = .046), whereas specificity trended toward improvement (79% [111 of 140]) vs 77% [108 of 140]; P = .06). Reading time per case was similar (unaided, 146 seconds; supported by AI, 149 seconds; P = .15). The AUC with the AI system alone was similar to the average AUC of the radiologists (0.89 vs 0.87). Conclusion Radiologists improved their cancer detection at mammography when using an artificial intelligence system for support, without requiring additional reading time. Published under a CC BY 4.0 license. See also the editorial by Bahl in this issue.

Artificial intelligence in medical imaging: threat or opportunity? Radiologists again at the forefront of innovation in medicine

One of the most promising areas of health innovation is the application of artificial intelligence (AI), primarily in medical imaging. This article provides basic definitions of terms such as “machine/deep learning” and analyses the integration of AI into radiology. Publications on AI have drastically increased from about 100–150 per year in 2007–2008 to 700–800 per year in 2016–2017. Magnetic resonance imaging and computed tomography collectively account for more than 50% of current articles. Neuroradiology appears in about one-third of the papers, followed by musculoskeletal, cardiovascular, breast, urogenital, lung/thorax, and abdomen, each representing 6–9% of articles. With an irreversible increase in the amount of data and the possibility to use AI to identify findings either detectable or not by the human eye, radiology is now moving from a subjective perceptual skill to a more objective science. Radiologists, who were on the forefront of the digital era in medicine, can guide the introduction of AI into healthcare. Yet, they will not be replaced because radiology includes communication of diagnosis, consideration of patient’s values and preferences, medical judgment, quality assurance, education, policy-making, and interventional procedures. The higher efficiency provided by AI will allow radiologists to perform more value-added tasks, becoming more visible to patients and playing a vital role in multidisciplinary clinical teams.

Deep neural network improves fracture detection by clinicians

Suspected fractures are among the most common reasons for patients to visit emergency departments (EDs), and X-ray imaging is the primary diagnostic tool used by clinicians to assess patients for fractures. Missing a fracture in a radiograph often has severe consequences for patients, resulting in delayed treatment and poor recovery of function. Nevertheless, radiographs in emergency settings are often read out of necessity by emergency medicine clinicians who lack subspecialized expertise in orthopedics, and misdiagnosed fractures account for upward of four of every five reported diagnostic errors in certain EDs. In this work, we developed a deep neural network to detect and localize fractures in radiographs. We trained it to accurately emulate the expertise of 18 senior subspecialized orthopedic surgeons by having them annotate 135,409 radiographs. We then ran a controlled experiment with emergency medicine clinicians to evaluate their ability to detect fractures in wrist radiographs with and without the assistance of the deep learning model. The average clinician's sensitivity was 80.8% (95% CI, 76.7-84.1%) unaided and 91.5% (95% CI, 89.3-92.9%) aided, and specificity was 87.5% (95 CI, 85.3-89.5%) unaided and 93.9% (95% CI, 92.9-94.9%) aided. The average clinician experienced a relative reduction in misinterpretation rate of 47.0% (95% CI, 37.4-53.9%). The significant improvements in diagnostic accuracy that we observed in this study show that deep learning methods are a mechanism by which senior medical specialists can deliver their expertise to generalists on the front lines of medicine, thereby providing substantial improvements to patient care.

Artificial intelligence as a medical device in radiology: ethical and regulatory issues in Europe and the United States

Abstract

Worldwide interest in artificial intelligence (AI) applications is growing rapidly. In medicine, devices based on machine/deep learning have proliferated, especially for image analysis, presaging new significant challenges for the utility of AI in healthcare. This inevitably raises numerous legal and ethical questions. In this paper we analyse the state of AI regulation in the context of medical device development, and strategies to make AI applications safe and useful in the future. We analyse the legal framework regulating medical devices and data protection in Europe and in the United States, assessing developments that are currently taking place. The European Union (EU) is reforming these fields with new legislation (General Data Protection Regulation [GDPR], Cybersecurity Directive, Medical Devices Regulation, In Vitro Diagnostic Medical Device Regulation). This reform is gradual, but it has now made its first impact, with the GDPR and the Cybersecurity Directive having taken effect in May, 2018. As regards the United States (U.S.), the regulatory scene is predominantly controlled by the Food and Drug Administration. This paper considers issues of accountability, both legal and ethical. The processes of medical device decision-making are largely unpredictable, therefore holding the creators accountable for it clearly raises concerns. There is a lot that can be done in order to regulate AI applications. If this is done properly and timely, the potentiality of AI based technology, in radiology as well as in other fields, will be invaluable.

Teaching Points

• AI applications are medical devices supporting detection/diagnosis, work-flow, cost-effectiveness.

• Regulations for safety, privacy protection, and ethical use of sensitive information are needed.

• EU and U.S. have different approaches for approving and regulating new medical devices.

• EU laws consider cyberattacks, incidents (notification and minimisation), and service continuity.

• U.S. laws ask for opt-in data processing and use as well as for clear consumer consent.

Transfer Representation Learning using Inception-V3 for the Detection of Masses in Mammography

Breast cancer is the most prevalent cancer among women. The most common method to detect breast cancer is mammography. However, interpreting mammography is a challenging task that requires high skills and is timeconsuming. In this work, we propose a computer-aided diagnosis (CAD) scheme for mammography based on transfer representation learning using the Inception-V3 architecture. We evaluate the performance of the proposed scheme using the INBreast database, where the features are extracted from different layers of the architecture. In order to cope with the small dataset size limitation, we expand the training dataset by generating artificial mammograms and employing different augmentation techniques. The proposed scheme shows great potential with a maximal area under the receiver operating characteristics curve of 0.91.

Characteristics of screen-detected cancers following concordant or discordant recalls at blinded double reading in biennial digital screening mammography

OBJECTIVES

To analyse which mammographic and tumour characteristics led to concordant versus discordant recalls at blinded double reading to further optimise our breast cancer screening programme.

METHODS

We included a consecutive series of 99,013 screening mammograms obtained between July 2013 and January 2015. All mammograms were double read in a blinded fashion. Discordant readings were routinely recalled without consensus or arbitration. During the 2-year follow-up, relevant data of the recalled women were collected. We compared mammographic characteristics, screening outcome and tumour characteristics between concordant and discordant recalls.

RESULTS

There were 2,543 concordant recalls (71.4%) and 997 discordant recalls (28.0%). The positive predictive value of a concordant recall was significantly higher (23.5% vs. 10.0%, p < 0.001). The proportion of BI-RADS 0 was significantly higher in the discordant recall group (75.7% vs. 56.3%, p < 0.001). Discordant recalls were more often an asymmetry or architectural distortion (21.8% vs. 13.2% and 9.3% vs. 6.5%, respectively, p < 0.001). There were no differences in the distribution of DCIS and invasive cancers and tumour characteristics were comparable for the two groups, except for a more favourable tumour grade in the discordant recall group (54.7% vs. 39.9% grade I tumours, p = 0.022).

CONCLUSIONS

Screen-detected cancers detected by a discordant reading show a more favourable tumour grade than cancers diagnosed after a concordant recall. The higher proportion of asymmetries and architectural distortions in this group provide a possible target for improving screening programmes by additional training of screening radiologists and the implementation of digital breast tomosynthesis.

KEY POINTS

• With blinded double reading of screening mammograms, screen-detected cancers detected by a discordant reading show a more favourable tumour grade than cancers diagnosed after a concordant recall. • The proportions of asymmetries and architectural distortions are higher in case of a discordant reading. • Possible improvement strategies could target additional training of screening radiologists and the implementation of digital breast tomosynthesis in breast cancer screening programmes.

Utilization of Computer-Aided Detection for Digital Screening Mammography in the United States, 2008 to 2016

PURPOSE

Computer-aided detection (CAD) for screening mammography is a software technology designed to improve radiologists' reading performance. Since 2007, multiple Breast Cancer Surveillance Consortium research papers have shown that CAD decreases performance by increasing recalls and decreasing the detection of invasive cancer while increasing the detection of ductal carcinoma in situ. The aim of this study was to test the hypothesis that CAD use by digital mammography facilities would decrease over time.

METHODS

In August 2007, August 2011, and March 2016, the FDA database of certified mammography facilities was accessed, and a random sample of 400 of approximately 8,500 total facilities was generated. In 2008 and 2011, a telephone survey was conducted of the facilities regarding digital mammography and CAD use. In 2016, facility websites were reviewed before calling the facilities. Bonferroni-corrected P values were used to assess statistical differences in the proportion of CAD at digital facilities for the three surveys.

RESULTS

The mean proportion of digital facilities using CAD was 91.4%, including 91.4% (128 of 140) in 2008, 90.2% (238 of 264) in 2011, and 92.3% (358 of 388) in 2016. The difference for 2008 versus 2011 was 1.3% (95% confidence interval [CI], -0.5% to 7.7%), for 2011 versus 2016 was -2.1% (95% CI, -6.9% to 2.7%), and for 2008 versus 2016 was -0.8% (95% CI, -6.7% to 5.0%).

CONCLUSIONS

In three national surveys, it was found that CAD use at US digital screening mammography facilities was stable from 2008 to 2016. This persistent utilization is relevant to the debate on the value of targeting ductal carcinoma in situ in screening.

Effectiveness and cost-effectiveness of double reading in digital mammography screening: A systematic review and meta-analysis

PURPOSE

Double reading is the strategy of choice for mammogram interpretation in screening programmes. It remains, however, unknown whether double reading is still the strategy of choice in the context of digital mammography. Our aim was to determine the effectiveness and cost-effectiveness of double reading versus single reading of digital mammograms in screening programmes.

METHODS

We performed a systematic review by searching the PubMed, Embase, and Cochrane Library databases up to April 2017. We used the QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies) tool and CHEERS (Consolidated Health Economic Evaluation Reporting Standards) checklist to assess the methodological quality of the diagnostic studies and economic evaluations, respectively. A proportion's meta-analysis approach, 95% Confidence Intervals (95% CI) and test of heterogeneity (P values) were used for pooled results. Costs are expressed US$ PPP (United States Dollar purchasing power parities). The PROSPERO ID of this Systematic Review's protocol is CRD42014013804.

RESULTS

Of 1473 potentially relevant hits, four high-quality studies were included. The pooled cancer detection rate of double reading was 6.01 per 1000 screens (CI: 4.47‰-7.77‰), and it was 5.65 per 1000 screens (CI: 3.95‰-7.65‰) for single reading (P=0.76). The pooled proportion of false-positives of double reading was 47.03 per 1000 screens (CI: 39.13‰-55.62‰) and it was 40.60 per 1000 screens (CI: 38.58‰-42.67‰) for single reading (P=0.12). One study reported, for double reading, an ICER (Incremental Cost-Effectiveness Ratio) of 16,684 Euros (24,717 US$ PPP; 2015 value) per detected cancer. Single reading+CAD (computer-aided-detection) was cost-effective in Japan.

CONCLUSION

The evidence of benefit for double reading compared to single reading for digital mammography interpretation is scarce. Double reading seems to increase operational costs, have a not significantly higher false-positive rate, and a similar cancer detection rate.

A sequential test for assessing observed agreement between raters

Assessing the agreement between two or more raters is an important topic in medical practice. Existing techniques, which deal with categorical data, are based on contingency tables. This is often an obstacle in practice as we have to wait for a long time to collect the appropriate sample size of subjects to construct the contingency table. In this paper, we introduce a nonparametric sequential test for assessing agreement, which can be applied as data accrues, does not require a contingency table, facilitating a rapid assessment of the agreement. The proposed test is based on the cumulative sum of the number of disagreements between the two raters and a suitable statistic representing the waiting time until the cumulative sum exceeds a predefined threshold. We treat the cases of testing two raters' agreement with respect to one or more characteristics and using two or more classification categories, the case where the two raters extremely disagree, and finally the case of testing more than two raters' agreement. The numerical investigation shows that the proposed test has excellent performance. Compared to the existing methods, the proposed method appears to require significantly smaller sample size with equivalent power. Moreover, the proposed method is easily generalizable and brings the problem of assessing the agreement between two or more raters and one or more characteristics under a unified framework, thus providing an easy to use tool to medical practitioners.

Usefulness of grayscale inverted images in addition to standard images in digital mammography

BACKGROUND

Mammography is essential for early diagnosis of breast cancer, which is the most common type of cancer in females that is associated with a high mortality rate. We investigated whether evaluation of the grayscale inverted images of mammograms would aid in increasing the diagnostic sensitivity of the mammographic imaging technique.

METHODS

Our study included 636 mammograms of 159 women who had undergone digital mammography. Standard, grayscale inverted, and standard plus grayscale inverted images were sequentially examined three times, at 15-day intervals, for the presence or assessment of pathological changes in the skin, calcification, asymmetric density, mass lesions, structural distortions, and intramammary and axillary lymph nodes. To determine whether grayscale inverted image assessment improved detection rates, the results of the three assessment modes were compared using Cochran's Q test and the McNemar test (p < 0.05 was considered statistically significant).

RESULTS

The average age of 159 patients was 50.4 years (range, 35-80 years). There were significant differences among the three assessment modes with respect to calcification and intramammary lymph nodes (p < 0.05); however, no significant differences were observed for the detection of other parameters.

CONCLUSIONS

Assessment of grayscale inverted images in addition to standard images facilitates the detection of microcalcification.

Classification of normal screening mammograms is strongly influenced by perceived mammographic breast density

INTRODUCTION

To investigate how breast screen readers classify normal screening cases using descriptors of normal mammographic features and to assess test cases for suitability for a single reading strategy.

METHODS

Fifteen breast screen readers interpreted a test set of 29 normal screening cases and classified them by firstly rating their perceived difficulty to reach a 'normal' decision, secondly identifying the cases' salient normal mammographic features and thirdly assessing the cases' suitability for a single reading strategy.

RESULTS

The relationship between the perceived difficulty in making 'normal' decisions and the normal mammographic features was investigated. Regular ductal pattern (T_b  = -0.439, P = 0.001), uniform density (T_b  = -0.527, P < 0.001), non-dense breasts (T_b  = -0.736, P < 0.001), symmetrical mammographic features (T_b  = -0.474, P = 0.001) and overlapped density (T_b  = 0.630, P < 0.001) had a moderate to strong correlation with the difficulty to make 'normal' decisions. Cases with regular ductal pattern (T_b  = 0.447, P = 0.002), uniform density (T_b  = 0.550, P < 0.001), non-dense breasts (T_b  = 0.748, P < 0.001) and symmetrical mammographic features (T_b  = 0.460, P = 0.001) were considered to be more suitable for single reading, whereas cases with overlapped density were not (T_b  = -0.679, P < 0.001).

CONCLUSION

The findings suggest that perceived mammographic breast density has a major influence on the difficulty for readers to classify cases as normal and hence their suitability for single reading.

Cost-Effectiveness of Double Reading versus Single Reading of Mammograms in a Breast Cancer Screening Programme

OBJECTIVES

The usual practice in breast cancer screening programmes for mammogram interpretation is to perform double reading. However, little is known about its cost-effectiveness in the context of digital mammography. Our purpose was to evaluate the cost-effectiveness of double reading versus single reading of digital mammograms in a population-based breast cancer screening programme.

METHODS

Data from 28,636 screened women was used to establish a decision-tree model and to compare three strategies: 1) double reading; 2) double reading for women in their first participation and single reading for women in their subsequent participations; and 3) single reading. We calculated the incremental cost-effectiveness ratio (ICER), which was defined as the expected cost per one additionally detected cancer. We performed a deterministic sensitivity analysis to test the robustness of the ICER.

RESULTS

The detection rate of double reading (5.17‰) was similar to that of single reading (4.78‰; P = .768). The mean cost of each detected cancer was €8,912 for double reading and €8,287 for single reading. The ICER of double reading versus single reading was €16,684. The sensitivity analysis showed variations in the ICER according to the sensitivity of reading strategies. The strategy that combines double reading in first participation with single reading in subsequent participations was ruled out due to extended dominance.

CONCLUSIONS

From our results, double reading appears not to be a cost-effective strategy in the context of digital mammography. Double reading would eventually be challenged in screening programmes, as single reading might entail important net savings without significantly changing the cancer detection rate. These results are not conclusive and should be confirmed in prospective studies that investigate long-term outcomes like quality adjusted life years (QALYs).

Neoadjuvant chemotherapy of breast cancer with pirarubicin versus epirubicin in combination with cyclophosphamide and docetaxel

Breast cancers (BC) are treated with surgery, radiotherapy, and chemotherapy. Neoadjuvant chemotherapy (NACT) is an emerging treatment option in many cancers and is given before primary therapy to shrink tumor size. The efficacy of NACT in varied settings of BC, such as inoperable tumors, borderline resectable tumors, and breast-conserving surgery, has been debated extensively in literature, and the results remain unclear and depended on a wide variety of factors such as cancer type, disease extent, and the specific combination of chemotherapy drugs. This study was performed to examine the efficacy, toxicity, and tolerability of pirarubicin (THP) and epirubicin (EPI) in combination with docetaxel and cyclophosphamide in a NACT setting for BC. A total of 48 patients with stage II or III breast cancers were randomly divided into two groups: THP group and EPI group. The patients in THP group received 2-4 cycles of neoadjuvant chemotherapy with DTC regimen (docetaxel, THP, cyclophosphamide), while patients in the EPI group received 2-4 cycles of DEC regimen (docetaxel, EPI, cyclophosphamide) before surgery. The incidence of adverse reactions and the efficacy of the treatment regimen were compared between the two groups. Prognostic evaluation indexes were estimated by Kaplan-Meier survival analysis, including the 5-year disease-free survival (DFS) and overall survival (OS). The overall response rate in THP group was 83.3 %, and the EPI group showed a response rate of 79.2 %, with no statistically significant difference in response rate between the two groups. The incidence of cardiac toxicity, myelosuppression, nausea, and vomiting in the THP group was significantly lower than the EPI group (all P < 0.05). The incidence of hepatic toxicity, alopecia, and diarrhea in the THP group was also lower than the EPI group, but these differences were not statistically significant. The 5-year DFS and OS in THP versus EPI groups were 80 versus 76 % (DFS) and 86 versus 81 % (OS), respectively. Our study found that NACT with DTC regimen and DEC regimen were both very effective in treatment of BC. However, THP-based combination therapy was associated with significantly lower incidence of cardiac toxicity, myelosuppression, nausea, and vomiting.

Mammographic quantitative image analysis and biologic image composition for breast lesion characterization and classification

PURPOSE

To investigate whether biologic image composition of mammographic lesions can improve upon existing mammographic quantitative image analysis (QIA) in estimating the probability of malignancy.

METHODS

The study population consisted of 45 breast lesions imaged with dual-energy mammography prior to breast biopsy with final diagnosis resulting in 10 invasive ductal carcinomas, 5 ductal carcinomain situ, 11 fibroadenomas, and 19 other benign diagnoses. Analysis was threefold: (1) The raw low-energy mammographic images were analyzed with an established in-house QIA method, "QIA alone," (2) the three-compartment breast (3CB) composition measure-derived from the dual-energy mammography-of water, lipid, and protein thickness were assessed, "3CB alone", and (3) information from QIA and 3CB was combined, "QIA + 3CB." Analysis was initiated from radiologist-indicated lesion centers and was otherwise fully automated. Steps of the QIA and 3CB methods were lesion segmentation, characterization, and subsequent classification for malignancy in leave-one-case-out cross-validation. Performance assessment included box plots, Bland-Altman plots, and Receiver Operating Characteristic (ROC) analysis.

RESULTS

The area under the ROC curve (AUC) for distinguishing between benign and malignant lesions (invasive and DCIS) was 0.81 (standard error 0.07) for the "QIA alone" method, 0.72 (0.07) for "3CB alone" method, and 0.86 (0.04) for "QIA+3CB" combined. The difference in AUC was 0.043 between "QIA + 3CB" and "QIA alone" but failed to reach statistical significance (95% confidence interval [-0.17 to + 0.26]).

CONCLUSIONS

In this pilot study analyzing the new 3CB imaging modality, knowledge of the composition of breast lesions and their periphery appeared additive in combination with existing mammographic QIA methods for the distinction between different benign and malignant lesion types.

Mammographic images segmentation based on chaotic map clustering algorithm

BACKGROUND

This work investigates the applicability of a novel clustering approach to the segmentation of mammographic digital images. The chaotic map clustering algorithm is used to group together similar subsets of image pixels resulting in a medically meaningful partition of the mammography.

METHODS

The image is divided into pixels subsets characterized by a set of conveniently chosen features and each of the corresponding points in the feature space is associated to a map. A mutual coupling strength between the maps depending on the associated distance between feature space points is subsequently introduced. On the system of maps, the simulated evolution through chaotic dynamics leads to its natural partitioning, which corresponds to a particular segmentation scheme of the initial mammographic image.

RESULTS

The system provides a high recognition rate for small mass lesions (about 94% correctly segmented inside the breast) and the reproduction of the shape of regions with denser micro-calcifications in about 2/3 of the cases, while being less effective on identification of larger mass lesions.

CONCLUSIONS

We can summarize our analysis by asserting that due to the particularities of the mammographic images, the chaotic map clustering algorithm should not be used as the sole method of segmentation. It is rather the joint use of this method along with other segmentation techniques that could be successfully used for increasing the segmentation performance and for providing extra information for the subsequent analysis stages such as the classification of the segmented ROI.

Automated breast image classification using features from its discrete cosine transform

Purpose

This work aimed to improve breast screening program accuracy using automated classification. The goal was to determine if whole image features represented in the discrete cosine transform would provide a basis for classification. Priority was placed on avoiding false negative findings.

Methods

Online datasets were used for this work. No informed consent was required. Programs were developed in Mathematica and, where necessary to improve computational performance ported to C++. The use of a discrete cosine transform to separate normal from cancerous breast tissue was tested. Features (moments of the mean) were calculated in square sections of the transform centered on the origin. K-nearest neighbor and naive Bayesian classifiers were tested.

Results

Forty-one features were generated and tested singly, and in combination of two or three. Using a k-nearest neighbor classifier, sensitivities as high as 98% with a specificity of 66% were achieved. With a naive Bayesian classifier, sensitivities as high as 100% were achieved with a specificity of 64%.

Conclusion

Whole image classification based on discrete cosine transform (DCT) features was effectively implemented with a high level of sensitivity and specificity achieved. The high sensitivity attained using the DCT generated feature set implied that these classifiers could be used in series with other methods to increase specificity. Using a classifier with near 100% sensitivity, such as the one developed in this project, before applying a second classifier could only boost the accuracy of that classifier.

Automatic detection of anomalies in screening mammograms

Background

Diagnostic performance in breast screening programs may be influenced by the prior probability of disease. Since breast cancer incidence is roughly half a percent in the general population there is a large probability that the screening exam will be normal. That factor may contribute to false negatives. Screening programs typically exhibit about 83% sensitivity and 91% specificity. This investigation was undertaken to determine if a system could be developed to pre-sort screening-images into normal and suspicious bins based on their likelihood to contain disease. Wavelets were investigated as a method to parse the image data, potentially removing confounding information. The development of a classification system based on features extracted from wavelet transformed mammograms is reported.

Methods

In the multi-step procedure images were processed using 2D discrete wavelet transforms to create a set of maps at different size scales. Next, statistical features were computed from each map, and a subset of these features was the input for a concerted-effort set of naïve Bayesian classifiers. The classifier network was constructed to calculate the probability that the parent mammography image contained an abnormality. The abnormalities were not identified, nor were they regionalized.

The algorithm was tested on two publicly available databases: the Digital Database for Screening Mammography (DDSM) and the Mammographic Images Analysis Society’s database (MIAS). These databases contain radiologist-verified images and feature common abnormalities including: spiculations, masses, geometric deformations and fibroid tissues.

Results

The classifier-network designs tested achieved sensitivities and specificities sufficient to be potentially useful in a clinical setting. This first series of tests identified networks with 100% sensitivity and up to 79% specificity for abnormalities. This performance significantly exceeds the mean sensitivity reported in literature for the unaided human expert.

Conclusions

Classifiers based on wavelet-derived features proved to be highly sensitive to a range of pathologies, as a result Type II errors were nearly eliminated. Pre-sorting the images changed the prior probability in the sorted database from 37% to 74%.