Computer-aided detection in the United Kingdom National Breast Screening Programme: prospective study

Factors Associated With False-Positive Recalls in Mammography Screening

BACKGROUND

We aimed to identify factors associated with false-positive recalls in mammography screening compared with women who were not recalled and those who received true-positive recalls.

METHODS

We included 29,129 women, aged 40 to 74 years, who participated in the Karolinska Mammography Project for Risk Prediction of Breast Cancer (KARMA) between 2011 and 2013 with follow-up until the end of 2017. Nonmammographic factors were collected from questionnaires, mammographic factors were generated from mammograms, and genotypes were determined using the OncoArray or an Illumina custom array. By the use of conditional and regular logistic regression models, we investigated the association between breast cancer risk factors and risk models and false-positive recalls.

RESULTS

Women with a history of benign breast disease, high breast density, masses, microcalcifications, high Tyrer-Cuzick 10-year risk scores, KARMA 2-year risk scores, and polygenic risk scores were more likely to have mammography recalls, including both false-positive and true-positive recalls. Further analyses restricted to women who were recalled found that women with a history of benign breast disease and dense breasts had a similar risk of having false-positive and true-positive recalls, whereas women with masses, microcalcifications, high Tyrer-Cuzick 10-year risk scores, KARMA 2-year risk scores, and polygenic risk scores were more likely to have true-positive recalls than false-positive recalls.

CONCLUSIONS

We found that risk factors associated with false-positive recalls were also likely, or even more likely, to be associated with true-positive recalls in mammography screening.

Deep Learning-Based Artificial Intelligence for Mammography

During the past decade, researchers have investigated the use of computer-aided mammography interpretation. With the application of deep learning technology, artificial intelligence (AI)-based algorithms for mammography have shown promising results in the quantitative assessment of parenchymal density, detection and diagnosis of breast cancer, and prediction of breast cancer risk, enabling more precise patient management. AI-based algorithms may also enhance the efficiency of the interpretation workflow by reducing both the workload and interpretation time. However, more in-depth investigation is required to conclusively prove the effectiveness of AI-based algorithms. This review article discusses how AI algorithms can be applied to mammography interpretation as well as the current challenges in its implementation in real-world practice.

A review of the PERFORMS scheme in breast screening

This review details the aetiology of the PERFORMS self-assessment scheme in breast screening, together with its subsequent development, current implementation and future function. The purpose of the scheme is examined and the importance of its continuing role in a changing screening service described, together with current evolution.

Interpretation time for screening mammography as a function of the number of computer-aided detection marks

Purpose: Computer-aided detection (CAD) alerts radiologists to findings potentially associated with breast cancer but is notorious for creating false-positive marks. Although a previous paper found that radiologists took more time to interpret mammograms with more CAD marks, our impression was that this was not true in actual interpretation. We hypothesized that radiologists would selectively disregard these marks when present in larger numbers.

Approach: We performed a retrospective review of bilateral digital screening mammograms. We use a mixed linear regression model to assess the relationship between number of CAD marks and ln (interpretation time) after adjustment for covariates. Both readers and mammograms were treated as random sampling units.

Results: Ten radiologists, with median experience after residency of 12.5 years (range 6 to 24) interpreted 1832 mammograms. After accounting for number of images, Breast Imaging Reporting and Data System category, and breast density, the number of CAD marks was positively associated with longer interpretation time, with each additional CAD mark proportionally increasing median interpretation time by 4.35% for a typical reader.

Conclusions: We found no support for our hypothesis that radiologists will selectively disregard CAD marks when they are present in larger numbers.

Artificial Intelligence for Mammography and Digital Breast Tomosynthesis: Current Concepts and Future Perspectives

Although computer-aided diagnosis (CAD) is widely used in mammography, conventional CAD programs that use prompts to indicate potential cancers on the mammograms have not led to an improvement in diagnostic accuracy. Because of the advances in machine learning, especially with use of deep (multilayered) convolutional neural networks, artificial intelligence has undergone a transformation that has improved the quality of the predictions of the models. Recently, such deep learning algorithms have been applied to mammography and digital breast tomosynthesis (DBT). In this review, the authors explain how deep learning works in the context of mammography and DBT and define the important technical challenges. Subsequently, they discuss the current status and future perspectives of artificial intelligence-based clinical applications for mammography, DBT, and radiomics. Available algorithms are advanced and approach the performance of radiologists-especially for cancer detection and risk prediction at mammography. However, clinical validation is largely lacking, and it is not clear how the power of deep learning should be used to optimize practice. Further development of deep learning models is necessary for DBT, and this requires collection of larger databases. It is expected that deep learning will eventually have an important role in DBT, including the generation of synthetic images.

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.

Designing and incorporating a real world data approach to international drug development and use: what the UK offers

Assessments of the safety, efficacy and appropriate use of new medicines lie at the heart of treatment development and subsequent adoption in clinical practice. Highly controlled randomised clinical trials routinely inform decisions on the approval, coverage and use of a medicine. Researchers and decision makers have become increasingly aware that these experimental data alone are insufficient to address those decisions fully. Real world data recorded from routine healthcare delivery by healthcare professionals and patients help provide a more complete picture of care. The UK, with its connectivity and rich longitudinal patient records, accumulated research and informatics experience and National Health Service, provides an exemplar of how real world data address a wide range of challenges across drug development.

Role of computer-aided detection in very small screening detected invasive breast cancers

This study aims to assess computer-aided detection (CAD) performance with full-field digital mammography (FFDM) in very small (equal to or less than 1 cm) invasive breast cancers. Sixty-eight invasive breast cancers less than or equal to 1 cm were retrospectively studied. All cases were detected with FFDM in women aged 49-69 years from our breast cancer screening program. Radiological characteristics of lesions following BI-RADS descriptors were recorded and compared with CAD sensitivity. Age, size, BI-RADS classification, breast density type, histological type of the neoplasm, and role of the CAD were also assessed. Per-study specificity and mass false-positive rate were determined by using 100 normal consecutive studies. Thirty-seven (54.4 %) masses, 17 (25 %) calcifications, 6 (8.8 %) masses with calcifications, 7 (10.3 %) architectural distortions, and 1 asymmetry (1.5 %) were found. CAD showed an overall sensitivity of 86.7 % (masses, 86.5 %; calcifications, 100 %; masses with calcifications, 100 %; and architectural distortion, 57.14 %), CAD failed to detect 9 out of 68 cases: 5 of 37 masses, 3 of 7 architectural distortions, and 1 of 1 asymmetry. Fifteen out of 37 masses were hyperdense, and all of them were detected by CAD. No association was seen among mass morphology or margins and detectability. Per-study specificity and CAD false-positive rate was 26 % and 1.76 false marks per study. In conclusion, CAD shows a high sensitivity and a low specificity. Lesion size, histology, and breast density do not influence sensitivity. Mammographic features, mass density, and thickness of the spicules in architectural distortions do influence.

False-positive reduction in mammography using multiscale spatial Weber law descriptor and support vector machines

In a CAD system for the detection of masses, segmentation of mammograms yields regions of interest (ROIs), which are not only true masses but also suspicious normal tissues that result in false positives. We introduce a new method for false-positive reduction in this paper. The key idea of our approach is to exploit the textural properties of mammograms and for texture description, to use Weber law descriptor (WLD), which outperforms state-of-the-art best texture descriptors. The basic WLD is a holistic descriptor by its construction because it integrates the local information content into a single histogram, which does not take into account the spatial locality of micropatterns. We extend it into a multiscale spatial WLD (MSWLD) that better characterizes the texture micro structures of masses by incorporating the spatial locality and scale of microstructures. The dimension of the feature space generated by MSWLD becomes high; it is reduced by selecting features based on their significance. Finally, support vector machines are employed to classify ROIs as true masses or normal parenchyma. The proposed approach is evaluated using 1024 ROIs taken from digital database for screening mammography and an accuracy of Az = 0.99 ± 0.003 (area under receiver operating characteristic curve) is obtained. A comparison reveals that the proposed method has significant improvement over the state-of-the-art best methods for false-positive reduction problem.

An interactive system for computer-aided diagnosis of breast masses

Although mammography is the only clinically accepted imaging modality for screening the general population to detect breast cancer, interpreting mammograms is difficult with lower sensitivity and specificity. To provide radiologists "a visual aid" in interpreting mammograms, we developed and tested an interactive system for computer-aided detection and diagnosis (CAD) of mass-like cancers. Using this system, an observer can view CAD-cued mass regions depicted on one image and then query any suspicious regions (either cued or not cued by CAD). CAD scheme automatically segments the suspicious region or accepts manually defined region and computes a set of image features. Using content-based image retrieval (CBIR) algorithm, CAD searches for a set of reference images depicting "abnormalities" similar to the queried region. Based on image retrieval results and a decision algorithm, a classification score is assigned to the queried region. In this study, a reference database with 1,800 malignant mass regions and 1,800 benign and CAD-generated false-positive regions was used. A modified CBIR algorithm with a new function of stretching the attributes in the multi-dimensional space and decision scheme was optimized using a genetic algorithm. Using a leave-one-out testing method to classify suspicious mass regions, we compared the classification performance using two CBIR algorithms with either equally weighted or optimally stretched attributes. Using the modified CBIR algorithm, the area under receiver operating characteristic curve was significantly increased from 0.865 ± 0.006 to 0.897 ± 0.005 (p < 0.001). This study demonstrated the feasibility of developing an interactive CAD system with a large reference database and achieving improved performance.

Value of one-view breast tomosynthesis versus two-view mammography in diagnostic workup of women with clinical signs and symptoms and in women recalled from screening

OBJECTIVE

The purpose of this study is to compare the diagnostic value of one-view digital breast tomosynthesis versus two-view full-field digital mammography (FFDM) alone, and versus a combined reading of both modalities.

MATERIALS AND METHODS

The datasets of one-view digital breast tomosynthesis and two-view FFDM of abnormal mammograms in 144 consecutive women admitted for diagnostic workup with clinical signs and symptoms (n = 78) or recalled from screening (n = 66) were read alone and in a combined setting. The malignant or benign nature of the lesions was established by histologic analysis of biopsied lesions or by 12-16-month follow-up.

RESULTS

Eighty-six of the 144 patients were found to have breast cancer. The BI-RADS categories for one-view digital breast tomosynthesis were significantly better than those for two-view FFDM (p < 0.001) and were equal to those of the combined reading in both women admitted for diagnostic workup and women recalled from screening. The sensitivity and negative predictive values of digital breast tomosynthesis were superior to those of FFDM in fatty and dense breasts overall and in women admitted for diagnostic workup and in women recalled from screening. Only 11% of digital breast tomosynthesis examinations required additional imaging, compared with 23% of FFDMs.

CONCLUSION

In patients with abnormal mammograms, one-view digital breast tomosynthesis had better sensitivity and negative predictive value than did FFDM in patients with fatty and dense breasts. They also suggest that digital breast tomosynthesis would likely increase the predictive values if incorporated in routine screening.

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

BACKGROUND

In accordance with European guidelines, mammography screening comprises independent readings by two breast radiologists (double reading). CAD (computer-aided detection) has been suggested to complement or replace one of the two readers (single reading + CAD).The aim of this systematic review is to address the following question: Is the reading of mammographic x-ray images by a single breast radiologist together with CAD at least as accurate as double reading?

METHODS

The electronic literature search included the databases Pub Med, EMBASE and The Cochrane Library. Two independent reviewers assessed abstracts and full-text articles.

RESULTS

1049 abstracts were identified, of which 996 were excluded with reference to inclusion and exclusion criteria; 53 full-text articles were assessed for eligibility. Finally, four articles were included in the qualitative analysis, and one in a GRADE synthesis.

CONCLUSIONS

The scientific evidence is insufficient to determine whether the accuracy of single reading + CAD is at least equivalent to that obtained in standard practice, i.e. double reading where two breast radiologists independently read the mammographic images.

Clinically missed cancer: how effectively can radiologists use computer-aided detection?

OBJECTIVE

The purpose of this study was to determine the effectiveness with which radiologists can use computer-aided detection (CADe) to detect cancer missed at screening.

MATERIALS AND METHODS

An observer study was performed to measure the ability of radiologists to detect breast cancer on mammograms with and without CADe. The images in the study were from 300 analog mammographic examinations. In 234 cases the mammograms were read clinically as normal and free of cancer for at least 2 subsequent years. In the other 66 cases, cancers were missed clinically. In 256 cases, current and previous mammograms were available. Eight radiologists read the dataset and recorded a BI-RADS assessment, the location of the lesion, and their level of confidence that the patient should be recalled for diagnostic workup for each suspicious lesion. Jackknife alternative free-response receiver operating characteristic analysis was used.

RESULTS

The jackknife alternative free-response receiver operating characteristic figure of merit was 0.641 without aid and 0.659 with aid (p = 0.06; 95% CI, -0.001 to 0.036). The sensitivity increased 9.9% (95% CI, 3.4-19%) and the callback rate 12.1% (95% CI, 7.3-20%) with CADe. Both increases were statistically significant (p < 0.001). Radiologists on average ignored 71% of correct computer prompts.

CONCLUSION

Use of CADe can increase radiologist sensitivity 10% with a comparable increase in recall rate. There is potential for CADe to have a bigger clinical impact because radiologists failed to recognize a correct computer prompt in 71% of missed cancer cases [corrected].

CAD e mamografia: por que usar esta ferramenta?

A mamografia representa o melhor método de detecção precoce do câncer de mama, porém cerca de 10% a 30% das lesões mamárias são perdidas no rastreamento, devido a limitações próprias dos observadores humanos. A detecção auxiliada por computador (computer-aided detection - CAD) é uma tecnologia relativamente nova que tem sido implementada em alguns serviços de mamografia, com o intuito de prover uma dupla leitura. Estudos clínicos têm demonstrado que o CAD aumenta a sensibilidade de detecção do câncer da mama, por radiologistas, em até 21%. Um sistema CAD é útil em situações em que exista alta variabilidade interobservador, falta de observadores treinados, ou na impossibilidade de se realizar a dupla leitura com dois ou mais radiologistas. O objetivo desta revisão está baseado na necessidade de atualizar a comunidade médica acerca desta ferramenta, como um método auxiliar, quantitativo, não operador-dependente, e que visa a melhorar a qualidade do diagnóstico do câncer de mama.

False positive marks on unsuspicious screening mammography with computer-aided detection

The contribution of computer-aided detection (CAD) systems as an interpretive aid in screening mammography can be hampered by a high rate of false positive detections. Specificity, false positive rate, and ease of dismissing false positive marks from two CAD systems are retrospectively evaluated. One hundred screening mammographic studies with a BI-RADS assessment code of 1 or 2 and at least 2-year normal mammographic follow-up were retrospectively reviewed using two CAD systems. Breast density, CAD marks, and radiologist's ease of dismissing false positive marks were recorded. Specificities from the two CAD versions considering all marks were 23% and 15% (p value = 0.07); mass marks, 35% and 17% (p value < 0.01); and calcification marks 62% and 75% (p value = 0.01). The two CAD versions did not differ regarding mean and median marks per case for all marks (2.3, 2.0 and 2.3, 2.0, p value = 0.65) or mass marks (1.6, 1.0 and 1.8, 2.0, p value = 0.15), but differed for calcification marks (0.8, 0 and 0.5, 0, p value < 0.01). Slightly higher specificity and fewer marks per case observed in dense breasts did not reach statistical significance. The reviewing radiologist classified most marks from both CAD systems (84% and 88%) as very easy/easy to dismiss. The two CAD versions had small differences in specificity and false positive marks. Differences, although not statistically significant, in specificities and false positive rates between dense and non-dense breasts warrant further research. Most false positive marks are easily dismissed and should not affect clinical performance.

Improving the performance of computer-aided detection of subtle breast masses using an adaptive cueing method

Current computer-aided detection (CAD) schemes for detecting mammographic masses have several limitations including high correlation with radiologists' detection and cueing most subtle masses only on one view. To increase CAD sensitivity in cueing more subtle masses that are likely missed and/or overlooked by radiologists without increasing false-positive rates, we investigated a new case-dependent cueing method by combining the original CAD-generated detection scores with a computed bilateral mammographic density asymmetry index. Using the new method, we adaptively raise the CAD-generated scores of the regions detected on 'high-risk' cases to cue more subtle mass regions and reduce the CAD scores of the regions detected on 'low-risk' cases to discard more false-positive regions. A testing dataset involving 78 positive and 338 negative cases was used to test this adaptive cueing method. Each positive case involves two sequential examinations in which the mass was detected in 'current' examination and missed in 'prior' examination but detected in a retrospective review by radiologists. Applying to this dataset, a pre-optimized CAD scheme yielded 75% case-based and 55% region-based sensitivity on 'current' examinations at a false-positive rate of 0.25 per image. CAD sensitivity was reduced to 42% (case based) and 27% (region based) on 'prior' examinations. Using the new cueing method, case-based and region-based sensitivity could maximally increase 9% and 33% on the 'prior' examinations, respectively. The percentages of the masses cued on two views also increased from 27% to 65%. The study demonstrated that using this adaptive cueing method enabled us to help CAD cue more subtle cancers without increasing the false-positive cueing rate.

Predicting readers' diagnostic accuracy with a new CAD algorithm

RATIONALE AND OBJECTIVES

Before computer-aided detection (CAD) algorithms can be used in clinical practice, they must be shown to improve readers' diagnostic accuracy over their unaided performance. This is usually accomplished through a large multireader, multicase (MRMC) clinical trial. It is burdensome, however, for an MRMC study to be performed with each new release of a CAD algorithm. The aim of this report is to present an approach for building models to predict readers' accuracy with a new CAD algorithm.

MATERIALS AND METHODS

A modeling approach for predicting readers' results with a new CAD algorithm is described. Multiple-variable logistic regression was used to build models for readers' sensitivity and false-positive rate, given the results of an MRMC study with an older CAD algorithm and the stand-alone performance results of a new CAD algorithm. Data from a large lung MRMC CAD trial are used to illustrate the modeling approach and test the ability of the models to predict readers' accuracy with the new CAD algorithm.

RESULTS

The model overestimated the readers' actual sensitivity with the new CAD algorithm, but this did not reach statistical significance (0.621 vs 0.603, P = .147). The observed and predicted false-positive rates also did not differ significantly (0.275 vs 0.285, P = .250).

CONCLUSIONS

Using one clinical study as a test case, it is shown that the modeling approach is feasible. More testing of the approach is needed to determine if and under what circumstances it can be used as an alternative to a full-scale MRMC study. Meanwhile, the approach can be used to determine if a new CAD algorithm is likely to improve readers' accuracy before embarking on a full-scale MRMC study.

The evolving role of new imaging methods in breast screening

The potential to avert breast cancer deaths through screening means that efforts continue to identify methods which may enhance early detection. While the role of most new imaging technologies remains in adjunct screening or in the work-up of mammography-detected abnormalities, some of the new breast imaging tests (such as MRI) have roles in screening groups of women defined by increased cancer risk. This paper highlights the evidence and the current role of new breast imaging technologies in screening, focusing on those that have broader application in population screening, including digital mammography, breast ultrasound in women with dense breasts, and computer-aided detection. It highlights that evidence on new imaging in screening comes mostly from non-randomised studies that have quantified test detection capability as adjunct to mammography, or have compared measures of screening performance for new technologies with that of conventional mammography. Two RCTs have provided high-quality evidence on the equivalence of digital and conventional mammography and on outcomes of screen-reading complemented by CAD. Many of these imaging technologies enhance cancer detection but also increase recall and false positives in screening.

Computer-aided detection of breast masses depicted on full-field digital mammograms: a performance assessment

OBJECTIVES

To investigate the feasibility of converting a computer-aided detection (CAD) scheme for digitised screen-film mammograms to full-field digital mammograms (FFDMs) and assessing CAD performance on a large database.

METHODS

The database included 6478 FFDM images acquired on 1120 females, with 525 cancer cases and 595 negative cases. The database was divided into five case groups: (1) cancer detected during screening, (2) interval cancers, (3) "high-risk" recommended for surgical excision, (4) recalled but negative and (5) negative (not recalled). A previously developed CAD scheme for masses depicted on digitised images was converted and re-optimised for FFDM images while keeping the same image-processing structure. CAD performance was analysed on the entire database.

RESULTS

The case-based sensitivity was 75.6% (397/525) for the current mammograms and 40.8% (42/103) for the prior mammograms deemed negative during clinical interpretation but "visible" during retrospective review. The region-based sensitivity was 58.1% (618/1064) for the current mammograms and 28.4% (57/201) for the prior mammograms. The CAD scheme marked 55.7% (221/397) and 35.7% (15/42) of the masses on both views of the current and the prior examinations, respectively. The overall CAD-cued false-positive rate was 0.32 per image, ranging from 0.29 to 0.51 for the five case groups.

CONCLUSION

This study indicated that (1) digitised image-based CAD can be converted for FFDMs while performing at a comparable, or better, level; (2) CAD detects a substantial fraction of cancers depicted on prior examinations, albeit most having been marked only on one view; and (3) CAD tends to mark more false-positive results on "difficult" negative cases that are more visually difficult for radiologists to interpret.

How widely is computer-aided detection used in screening and diagnostic mammography?

PURPOSE

The aim of this study was to determine how widely computer-aided detection (CAD) is used in screening and diagnostic mammography and to see if there are differences between hospital facilities and private offices.

METHODS

The nationwide Medicare Part B fee-for-service databases for 2004 to 2008 were used. The Current Procedural Terminology(®) codes for screening and diagnostic mammography (both digital and screen film) and the CAD add-on codes were selected. Procedure volume was compared for screening vs diagnostic mammography and for hospital facilities vs private offices.

RESULTS

From 2004 to 2008, Medicare screening mammography volume increased slightly from 5,728,419 to 5,827,326 (+2%), but the use of screening CAD increased from 2,257,434 to 4,305,595 (+91%). By 2008, CAD was used in 74% of all screening mammographic studies. During this same time period, the Medicare volume of diagnostic mammography declined slightly from 1,835,700 to 1,682,026 (-8%), but the use of diagnostic CAD increased from 360,483 to 845,461 (+135%). By 2008, CAD was used in 50% of all diagnostic mammographic studies. In hospital facilities in 2008, CAD was used in 70% of all screening mammographic studies, compared with 81% in private offices. For diagnostic mammography in 2008, CAD was used in 48% in hospitals, compared with 55% in private offices.

CONCLUSION

Despite some operational drawbacks to using CAD, radiologists have embraced it in an effort to improve cancer detection. Its use has grown rapidly, and in 2008, it was used in three-quarters of all screening mammographic studies and half of all diagnostic mammographic studies. Women undergoing either screening or diagnostic mammography are more likely to receive CAD if they go to a private office than if they go to a hospital facility, although the differences are not great.

Computer-aided detection; the effect of training databases on detection of subtle breast masses

RATIONALE AND OBJECTIVES

Lesion conspicuity is typically highly correlated with visual difficulty for lesion detection, and computer-aided detection (CAD) has been widely used as a "second reader" in mammography. Hence, increasing CAD sensitivity in detecting subtle cancers without increasing false-positive rates is important. The aim of this study was to investigate the effect of training database case selection on CAD performance in detecting low-conspicuity breast masses.

MATERIALS AND METHODS

A full-field digital mammographic image database that included 525 cases depicting malignant masses was randomly partitioned into three subsets. A CAD scheme was applied to detect all initially suspected mass regions and compute region conspicuity. Training samples were iteratively selected from two of the subsets. Four types of training data sets-(1) one including all available true-positive mass regions in the two subsets ("all"), (2) one including 350 randomly selected mass regions ("diverse"), (3) one including 350 high-conspicuity mass regions ("easy"), and (4) one including 350 low-conspicuity mass regions ("difficult")-were assembled. In each training data set, the same number of randomly selected false-positive regions as the true-positives were also included. Two classifiers, an artificial neural network (ANN) and a k-nearest neighbor (KNN) algorithm, were trained using each of the four training data sets and tested on all suspected regions in the remaining data set. Using a threefold cross-validation method, the performance changes of the CAD schemes trained using one of the four training data sets were computed and compared.

RESULTS

CAD initially detected 1025 true-positive mass regions depicted on 507 cases (97% case-based sensitivity) and 9569 false-positive regions (3.5 per image) in the entire database. Using the all training data set, CAD achieved the highest overall performance on the entire testing database. However, CAD detected the highest number of low-conspicuity masses when the difficult training data set was used for training. Results did agree for both ANN-based and KNN-based classifiers in all tests. Compared to the use of the all training data set, the sensitivity of the schemes trained using the difficult data set decreased by 8.6% and 8.4% for the ANN and KNN algorithm on the entire database, respectively, but the detection of low-conspicuity masses increased by 7.1% and 15.1% for the ANN and KNN algorithm at a false-positive rate of 0.3 per image.

CONCLUSIONS

CAD performance depends on the size, diversity, and difficulty level of the training database. To increase CAD sensitivity in detecting subtle cancer, one should increase the fraction of difficult cases in the training database rather than simply increasing the training data set size.

Mammographic features of breast cancers at single reading with computer-aided detection and at double reading in a large multicenter prospective trial of computer-aided detection: CADET II

PURPOSE

To evaluate the mammographic features of breast cancer that favor lesion detection with single reading and computer-aided detection (CAD) or with double reading.

MATERIALS AND METHODS

The Computer Aided Detection Evaluation Trial II study was approved by the ethics committee, and all participants provided written informed consent. A total of 31,057 women were recruited from three screening centers between September 2006 and August 2007. They were randomly allocated to the double reading group, the single reading with CAD group, or the double reading and single reading with CAD group at a ratio of 1:1:28, respectively. In this study, cancers in the women whose mammograms were read with both single reading with CAD and double reading were retrospectively reviewed. The original mammograms were obtained for each case and reviewed by two of three experienced breast radiologists in consensus. The method of detection was noted. The size and predominant mammographic feature of the cancer were recorded, as was the breast density. CAD marking data were reviewed to determine if the cancer had been correctly marked.

RESULTS

A total of 227 cancers were detected in 28,204 women. A total of 170 cases were recalled with both reading regimens. Lesion types were masses (66%), microcalcifications (25%), parenchymal deformities (6%), and asymmetric densities (3%). The ability of the reading regimens to correctly prompt the reader to recall cases varied significantly by lesion type (P < .001). More parenchymal deformities were recalled with double reading, whereas more asymmetric densities were recalled with single reading with CAD. There was no difference in the ability of either reading regimen to prompt the reader to correctly recall masses or microcalcifications. CAD correctly prompted 100% of microcalcifications, 87% of mass lesions, 80% of asymmetric densities, and 50% of parenchymal deformities. CAD correctly marked 93% of spiculated masses compared with 80% of ill-defined masses (P = .054). There was a significant trend for cancers detected with double reading to occur only in women with a denser mammographic background pattern (P = .02). Size had no effect on lesion detection.

CONCLUSION

Readers using either single reading with CAD or double reading need to be aware of the strengths and weaknesses of reading regimens to avoid missing the more challenging cancer cases.

Impact of a CAD system in a screen-film mammography screening program: a prospective study

OBJECTIVE

The purpose of our study was to perform a prospective assessment of the impact of a CAD system in a screen-film mammography screening program during a period of 3 years.

MATERIALS AND METHODS

Our study was carried out on a population of 21,855 asymptomatic women (45-65 years). Mammograms were processed in a CAD system and independently interpreted by one of six radiologists. We analyzed the following parameters: sensitivity of radiologist's interpretation (without and with CAD), detection increase, recall rate and positive predictive value of biopsy, CAD's marks, radiologist's false negatives and comparative analysis of carcinomas detected and non-detected by CAD.

RESULTS

Detection rate was 4.3‰. CAD supposed an increase of 0.1‰ in detection rate and 1% in the total number of cases (p<0.005). The impact on recall rate was not significant (0.4%) and PPV of percutaneous biopsy was unchanged by CAD (20.23%). CAD's marks were 2.7 per case and 0.7 per view. Radiologist's false negatives were 13 lesions which were initially considered as CAD's false positives.

CONCLUSIONS

CAD supposed a significant increase in detection, without modifications in recall rates and PPV of biopsy. However, better results could have been achieved if radiologists had considered actionable those cases marked by CAD but initially misinterpreted.

Using computer-aided detection in mammography as a decision support

Objective:

To evaluate an interactive computer-aided detection (CAD) system for reading mammograms to improve decision making.

Methods:

A dedicated mammographic workstation has been developed in which readers can probe image locations for the presence of CAD information. If present, CAD findings are displayed with the computed malignancy rating. A reader study was conducted in which four screening radiologists and five non-radiologists participated to study the effect of this system on detection performance. The participants read 120 cases of which 40 cases had a malignant mass that was missed at the original screening. The readers read each mammogram both with and without CAD in separate sessions. Each reader reported localized findings and assigned a malignancy score per finding. Mean sensitivity was computed in an interval of false-positive fractions less than 10%.

Results:

Mean sensitivity was 25.1% in the sessions without CAD and 34.8% in the CAD-assisted sessions. The increase in detection performance was significant (p = 0.012). Average reading time was 84.7 ± 61.5 s/case in the unaided sessions and was not significantly higher when interactive CAD was used (85.9 ± 57.8 s/case).

Conclusion:

Interactive use of CAD in mammography may be more effective than traditional CAD for improving mass detection without affecting reading time.

"CADEAT": considerations on the use of CAD (computer-aided diagnosis) in mammography

Computer-aided diagnosis (CAD) has been extensively reported to increase sensitivity by about 10% when added to a single reading while increasing recall rate by 12%, and its current use can be safely recommended in clinical practice. CAD has been suggested as a possible alternative to conventional double reading in screening. Uncontrolled comparison is consistent and suggests that CAD is comparable to double reading in incremental cancer detection rate (CAD +10.6%, double reading +9.1%) and possibly better in recall rate (CAD +12.5%, double reading +28.8%). However, controlled studies comparing single reading + CAD to conventional double reading are not consistent and on average suggest a lower cancer detection rate (-5.1%) and a lower recall rate (-9.8%) for CAD. Scientific evidence is not sufficient for a safe recommendation of single reading + CAD as a current alternative to conventional double reading.

Invasive breast cancers detected by screening mammography: a detailed comparison of computer-aided detection-assisted single reading and double reading

To compare double reading plus arbitration for discordance, (currently best practice, (BP)) with computer-aided-detection (CAD)-assisted single reading (CAD-R) for detection of invasive cancers detected within BreastScreen Australia. Secondarily, to examine characteristics of cancers detected/rejected using each method. Mammograms of 157 randomly selected double-read invasive cancers were mixed 1:9 with normal cancers (total 1569), all detected in a BreastScreen service. Cancers were detected by two readers or one reader (C2 and C1 cancers, ratio 70:30%) in the program. The 1569 film-screen mammograms were read by two radiologists (reader A (RA) and reader B(RB)), with findings recorded before and after CAD. Discordant findings with BP were resolved by arbitration. We compared CAD-assisted reading (CAD-RA, CAD-RB) with BP, and CAD and arbitration contribution to findings. We correlated cancer size, sensitivity and mammographic density with detection methods. BP sensitivity 90.4% compared with CAD-RA sensitivity 86.6% (P = 0.12) and CAD-RB 94.3% (P = 0.14). CAD-RB specificity was less than BP (P = 0.01). CAD sensitivity was 93%, but readers rejected most positive CAD prompts. After CAD, reader's sensitivity increased 1.9% and specificity dropped 0.2% and 0.8%. Arbitration decreased specificity 4.7%. Receiving operator curves analysis demonstrated BP accuracy better than CAD-RA, borderline significance (P = 0.07), but not CAD-RB. Secondarily, cancer size was similar for BP and CAD-R. Cancers recalled after arbitration (P = 0.01) and CAD-R (P = 0.10) were smaller. No difference in cancer size or sensitivity between reading methods was found with increasing breast density. CAD-R and BP sensitivity and cancer detection size were not significantly different. CAD-R specificity was significantly lower for one reader.

A boosting framework for visuality-preserving distance metric learning and its application to medical image retrieval

Similarity measurement is a critical component in content-based image retrieval systems, and learning a good distance metric can significantly improve retrieval performance. However, despite extensive study, there are several major shortcomings with the existing approaches for distance metric learning that can significantly affect their application to medical image retrieval. In particular, "similarity" can mean very different things in image retrieval: resemblance in visual appearance (e.g., two images that look like one another) or similarity in semantic annotation (e.g., two images of tumors that look quite different yet are both malignant). Current approaches for distance metric learning typically address only one goal without consideration of the other. This is problematic for medical image retrieval where the goal is to assist doctors in decision making. In these applications, given a query image, the goal is to retrieve similar images from a reference library whose semantic annotations could provide the medical professional with greater insight into the possible interpretations of the query image. If the system were to retrieve images that did not look like the query, then users would be less likely to trust the system; on the other hand, retrieving images that appear superficially similar to the query but are semantically unrelated is undesirable because that could lead users toward an incorrect diagnosis. Hence, learning a distance metric that preserves both visual resemblance and semantic similarity is important. We emphasize that, although our study is focused on medical image retrieval, the problem addressed in this work is critical to many image retrieval systems. We present a boosting framework for distance metric learning that aims to preserve both visual and semantic similarities. The boosting framework first learns a binary representation using side information, in the form of labeled pairs, and then computes the distance as a weighted Hamming distance using the learned binary representation. A boosting algorithm is presented to efficiently learn the distance function. We evaluate the proposed algorithm on a mammographic image reference library with an Interactive Search-Assisted Decision Support (ISADS) system and on the medical image data set from ImageCLEF. Our results show that the boosting framework compares favorably to state-of-the-art approaches for distance metric learning in retrieval accuracy, with much lower computational cost. Additional evaluation with the COREL collection shows that our algorithm works well for regular image data sets.

Strategies for digital mammography interpretation in a clinical patient population

Mammography is the basic imaging modality for early detection of breast cancer. The aim of this prospective study was to evaluate the impact of different mammogram reading strategies on the diagnostic yield in a consecutive patient population referred for digital mammography to a hospital. First, the effect of using computer-aided detection (CAD) software on the performance of mammogram readers was studied. Furthermore, the impact of employing technologists as either prereaders or double readers was assessed, as compared to the conventional strategy of single reading by a radiologist. Digital mammograms of 1,048 consecutive patients were evaluated by a radiologist and 3 technologists with and without the use of CAD software. ROC analysis was used to study the effects of the different strategies. In the conventional strategy, an overall area under the curve (AUC) of 0.92 was found, corresponding to a sensitivity of 84% and specificity of 94%. When applying CAD software, the AUCs were similar before and after CAD for all readers (mean of 0.95). Employing technologists in prereading and double reading of mammograms resulted in a mean AUC of 0.91 and 0.96, respectively. In the prereading strategy, the corresponding sensitivity and specificity were 81 and 96%; in the double reading strategy they were 96 and 79%, respectively. Concluding, in this clinical population, systematic application of CAD software by either radiologist or technologists failed to improve the diagnostic yield. Furthermore, employing technologists as double readers of mammograms was the most effective strategy in improving breast cancer detection in daily clinical practice.

Matching breast masses depicted on different views a comparison of three methods

RATIONALE AND OBJECTIVES

Computerized determination of optimal search areas on mammograms for matching breast mass regions depicted on two ipsilateral views remains a challenge for developing multiview-based computer-aided detection (CAD) schemes. The purpose of this study was to compare three methods aimed at matching CAD-cued mass regions depicted on two views and the associated impact on CAD performance.

MATERIALS AND METHODS

The three search methods used (1) an annular (fan-shaped) band, (2) a straight strip perpendicular to the estimated centerline, and (3) a mixed search area bound on the chest wall side by a straight line and an annular arc on the nipple side, respectively. An image database of 200 examinations with positive results depicting the masses on two views and 200 examinations with negative results was used for testing. Two performance assessment experiments were conducted. The first investigated the maximum matching sensitivity as a function of the search area size, and the second assessed the change in CAD performance using these three search methods.

RESULTS

To include all 200 paired mass regions within the search areas, maximum widths were 28 and 68 mm for the use of the straight strip and the annular band search methods, respectively. When applying a single-image-based CAD scheme to this image database, 172 masses (86% sensitivity) and 523 false-positive (FP) regions (0.33 per image) were detected and cued. Among the positive findings, 92 were cued by the CAD system on both views, and 80 were cued on only one view. In an attempt to match as many of the 172 CAD-cued masses (true-positive [TP] regions) on two views by incrementally reducing the CAD threshold inside the different search areas, the CAD scheme generated 158 TP-TP paired matches with 14 TP-FP paired matches, 142 TP-TP paired matches with 30 TP-FP paired matches, and 146 TP-TP paired matches with 26 TP-FP paired matches, using the methods involving the straight strip, the annular band, and the mixed search areas, respectively. Using the straight strip search method, the CAD also eliminated 25% of FP regions initially cued by the single-image-based CAD scheme and generated the lowest case-based FP detection rate, namely, 15% less than that generated by the annular band method.

CONCLUSIONS

This study showed that among these three search methods, the straight strip method required a smaller search area and achieved the highest level of CAD performance.

Performance of computer-aided detection applied to full-field digital mammography in detection of breast cancers

OBJECTIVE

The aim of this retrospective study was to evaluate performance of computer-aided detection (CAD) with full-field digital mammography (FFDM) in detection of breast cancers.

MATERIALS AND METHODS

CAD was retrospectively applied to standard mammographic views of 127 cases with biopsy proven breast cancers detected with FFDM (Senographe 2000, GE Medical Systems). CAD sensitivity was assessed in total group of 127 cases and for subgroups based on breast density, mammographic lesion type, mammographic lesion size, histopathology and mode of presentation.

RESULTS

Overall CAD sensitivity was 91% (115 of 127 cases). There were no statistical differences (p > 0.1) in CAD detection of cancers in dense breasts 90% (53/59) versus non-dense breasts 91% (62/68). There was statistical difference (p < 0.05) in CAD detection of cancers that appeared mammographically as microcalcifications only versus other mammographic manifestations. CAD detected 100% (44/44) of cancers manifesting as microcalcifications, 89% (47/53) as no-calcified masses or asymmetries, 88% (14/16) as masses with associated calcifications, and 71% (10/14) as architectural distortions. CAD sensitivity for cancers 1-10mm was 84% (38/45); 11-20mm 93% (55/59); and >20mm 97% (22/23).

CONCLUSION

CAD applied to FFDM showed 100% sensitivity in identifying cancers manifesting as microcalcifications only and high sensitivity 86% (71/83) for other mammographic appearances of cancer. Sensitivity is influenced by lesion size. CAD in FFDM is an adjunct helping radiologist in early detection of breast cancers.

Investigation of reading mode and relative sensitivity as factors that influence reader performance when using computer-aided detection software

RATIONALE AND OBJECTIVES

The aim of this study was to investigate the effects of relative sensitivity (reader without computer-aided detection [CAD] vs stand-alone CAD) and reading mode on reader performance when using CAD software.

MATERIALS AND METHODS

Two sets of 100 images (low-contrast and high-contrast sets) were created by adding low-contrast or high-contrast simulated masses to random locations in 100 normal mammograms. This produced a relative sensitivity, substantially less for the low-contrast set and similar for the high-contrast set. Seven readers reviewed every image in each set and specified location and probability scores using three reading modes (without CAD, second read with CAD, and concurrent read with CAD). Reader detection accuracy was analyzed using areas under free-response receiver operating characteristic curves, sensitivity, and the number of false-positive findings per image.

RESULTS

For the low-contrast set, average differences in areas under free-response receiver operating characteristic curves, sensitivity, and false-positive findings per image without CAD were 0.02, 0.12, and 0.11, respectively, compared to second read and 0.05, 0.17, and 0.09 (not statistically significant), respectively, compared to concurrent read. For the high-contrast set, average differences were 0.002 (not statistically significant), 0.04, and 0.05, respectively, compared to second read and -0.004 (not statistically significant), 0.04, and 0.08 (not statistically significant), respectively, compared to concurrent read (all differences were statistically significant except as noted). Differences were greater in the low-contrast set than the high-contrast set. Differences between second read and concurrent read were not significant.

CONCLUSIONS

Relative sensitivity is a critical factor that determines incremental improvement in reader performance when using CAD and appears to be more important than reading mode. Relative sensitivity may determine the clinical usefulness of CAD in different clinical applications and for different types of users.

The effect of Premium View post-processing software on digital mammographic reporting

The aim of this study was to identify the effect of the installation of Premium View post-processing software on our mammographic reporting performance, in particular the effects on our recall rate, biopsy rate and cancer detection rate. The case notes and imaging of all patients discussed at the weekly indeterminate imaging multidisciplinary team meeting were reviewed retrospectively before, immediately after and at a delayed interval following the installation of Premium View post-processing software. Factors recorded included the mammographic abnormality, further investigations and final histology. The indeterminate mammogram rate increased significantly from a baseline of 5.7% (before Premium View) to 8.7% in the time period immediately after the installation of Premium View (p=0.002). The stereotactic biopsy rate also increased from 0.8% to 2.4% (p=0.001), with a significant increase in the overall cancer detection rate from 3.4% to 4.4% (p=0.02). In the follow-up period several months after the installation of Premium View, the indeterminate mammogram rate returned to a level similar to that before Premium View (6%; p=0.7). The stereotactic biopsy rate remained significantly higher at 1.6% (p=0.07), as did the overall cancer detection rate of 5.0% (p=0.003). In conclusion, the use of Premium View may lead to higher cancer detection rates, at the expense of an initial increase in recall rate. Although prospective studies are suggested, this result is of interest in light of the proposed installation of digital mammography across the NHS Breast Screening Programme.

Computer-Aided Diagnosis in Mammography Using Content-based Image Retrieval Approaches: Current Status and Future Perspectives

As the rapid advance of digital imaging technologies, the content-based image retrieval (CBIR) has became one of the most vivid research areas in computer vision. In the last several years, developing computer-aided detection and/or diagnosis (CAD) schemes that use CBIR to search for the clinically relevant and visually similar medical images (or regions) depicting suspicious lesions has also been attracting research interest. CBIR-based CAD schemes have potential to provide radiologists with "visual aid" and increase their confidence in accepting CAD-cued results in the decision making. The CAD performance and reliability depends on a number of factors including the optimization of lesion segmentation, feature selection, reference database size, computational efficiency, and relationship between the clinical relevance and visual similarity of the CAD results. By presenting and comparing a number of approaches commonly used in previous studies, this article identifies and discusses the optimal approaches in developing CBIR-based CAD schemes and assessing their performance. Although preliminary studies have suggested that using CBIR-based CAD schemes might improve radiologists' performance and/or increase their confidence in the decision making, this technology is still in the early development stage. Much research work is needed before the CBIR-based CAD schemes can be accepted in the clinical practice.

Computer-assisted diagnosis (CAD) in mammography: comparison of diagnostic accuracy of a new algorithm (Cyclopus, Medicad) with two commercial systems

PURPOSE

The study compares the diagnostic accuracy (correct identification of cancer) of a new computer-assisted diagnosis (CAD) system (Cyclopus) with two other commercial systems (R2 and CADx).

MATERIALS AND METHODS

Cyclopus was tested on a set of 120 mammograms on which the two compared commercial systems had been previously tested. The set consisted of mammograms reported as negative, preceding 31 interval cancers reviewed as screening error or minimal sign, and of 89 verified negative controls randomly selected from the same screening database.

RESULTS

Cyclopus sensitivity was 74.1% (R2=54.8%; CADx=41.9%) and was higher for interval cancers reviewed as screening error (90.9%; R2=54.5%; CADx=81.8%) compared with those reviewed as minimal sign (65.0%; R2=55.0%; CADx=20.0%). Specificity was 15.7% (R2=29.2%; CADx=17.9%). Overall accuracy was 30.8% (R2=35.8%; CADx=24.1%). The positive predictive value of a case with CAD marks [regions of interest (ROI)] was 23.4% (23/98; R2=16.0%; CADx=15.1%). Average ROI number per view among negative controls was 1.13 (R2=0.93; CADx=0.99). Cyclopus was more sensitive for masses compared with isolated microcalcifications (208 vs 62 ROI; R2=90 vs 213; CADx=192 vs 130).

CONCLUSIONS

Compared with two other commercial systems, Cyclopus was more sensitive (R2 p=0.14; CADx p=0.02) and less specific (R2 p=0.02; CADx p=0.64).

True detection versus "accidental" detection of small lung cancer by a computer-aided detection (CAD) program on chest radiographs

To evaluate the number of actual detections versus "accidental" detections by a computer-aided detection (CAD) system for small nodular lung cancers (<or=30 mm) on chest radiographs, using two different criteria for measuring performance. A Food-and-Drug-Administration-approved CAD program (version 1.0; Riverain Medical) was applied to 34 chest radiographs with a "radiologist-missed" nodular cancer and 36 radiographs with a radiologist-mentioned nodule (a newer version 3.0 was also applied to the 36-case database). The marks applied by this CAD system consisted of 5-cm-diameter circles. A strict "nodule-in-center" criterion and a generous "nodule-in-circle" criterion were compared as methods for the calculation of CAD sensitivity. The increased sensitivities by the nodule-in-circle criterion were considered as nodules detected by chance. The number of false-positive (FP) marks was also analyzed. For the 34 radiologist-missed cancers, the nodule-in-circle criterion caused eight more cancers (24%) to be detected by chance, as compared to the nodule-in-center criterion, when using the version 1.0 results. For the 36 radiologist-mentioned nodules, the nodule-in-circle criterion caused seven more lesions (19%) to be detected by chance, as compared to the nodule-in-center criterion, when using the version 1.0 results, and three more lesions (8%) to be detected by chance when using the version 3.0 results. Version 1.0 yielded a mean of six FP marks per image, while version 3.0 yielded only three FP marks per image. The specific criteria used to define true- and false-positive CAD detections can substantially influence the apparent accuracy of a CAD system.

Computer-aided detection in full-field digital mammography in a clinical population: performance of radiologist and technologists

The purpose of the study was to evaluate the impact of a computer-aided detection (CAD) system on the performance of mammogram readers in interpreting digital mammograms in a clinical population. Furthermore, the ability of a CAD system to detect breast cancer in digital mammography was studied in comparison to the performance of radiologists and technologists as mammogram readers. Digital mammograms of 1,048 consecutive patients were evaluated by a radiologist and three technologists. Abnormalities were recorded and an imaging conclusion was given as a BI-RADS score before and after CAD analysis. Pathology results during 12 months follow up were used as a reference standard for breast cancer. Fifty-one malignancies were found in 50 patients. Sensitivity and specificity were computed before and after CAD analysis and provided with 95% CIs. In order to assess the detection rate of malignancies by CAD and the observers, the pathological locations of these 51 breast cancers were matched with the locations of the CAD marks and the mammographic locations that were considered to be suspicious by the observers. For all observers, the sensitivity rates did not change after application of CAD. A mean sensitivity of 92% was found for all technologists and 84% for the radiologist. For two technologists, the specificity decreased (from 84 to 83% and from 77 to 75%). For the radiologist and one technologist, the application of CAD did not have any impact on the specificity rates (95 and 83%, respectively). CAD detected 78% of all malignancies. Five malignancies were indicated by CAD without being noticed as suspicious by the observers. In conclusion, the results show that systematic application of CAD in a clinical patient population failed to improve the overall sensitivity of mammogram interpretation by the readers and was associated with an increase in false-positive results. However, CAD marked five malignancies that were missed by the different readers.

Improving performance of computer-aided detection scheme by combining results from two machine learning classifiers

RATIONALE AND OBJECTIVES

Global data-based and local instance-based machine-learning methods and classifiers have been widely used to optimize computer-aided detection and diagnosis (CAD) schemes to classify between true-positive and false-positive detections. In this study, the correlation between these two types of classifiers was investigated using a new independent testing data set, and the potential improvement of a CAD scheme's performance by combining the results of the two classifiers in detecting breast masses was assessed.

MATERIALS AND METHODS

The CAD scheme first used image filtering and a multilayer topographic region growth algorithm to detect and segment suspicious mass regions. The scheme then used an image feature-based classifier to classify these regions into true-positive and false-positive regions. Two classifiers were used in this study. One was a global data-based machine-learning classifier, an artificial neural network (ANN), and the other was a local instance-based machine-learning classifier, a k-nearest neighbor (KNN) algorithm. An independent image database including 400 mammographic examinations was used in this study. Of these, 200 were cancer cases and 200 were negative cases. The preoptimized CAD scheme was applied twice to the database using the two different classifiers. The correlation between the two sets of classification results was analyzed. Three sets of CAD performance results using the ANN, KNN, and average detection scores from both classifiers were assessed and compared using the free-response receiver-operating characteristic method.

RESULTS

The results showed that the ANN achieved higher performance than the KNN algorithm, with a normalized area under the performance curve (AUC) of 0.891 versus 0.845. The correlation coefficients between the detection scores generated by the two classifiers were 0.436 and 0.161 for the true-positive and false-positive detections, respectively. The average detection scores of the two classifiers improved CAD performance and reliability by increasing the AUC to 0.912 and reducing the standard error of the estimated AUC by 14.4%. The detection sensitivity was also increased from 75.8% (ANN) and 65.9% (KNN) to 80.3% at a false-positive detection rate of 0.3 per image.

CONCLUSIONS

This study demonstrates that two global data-based and local data-based machine-learning classifiers (ANN and KNN) generated low correlated detection results and that combining the detection scores of these two classifiers significantly improved overall CAD performance (P < .01) and reduced standard error in CAD performance assessment.