Comparison of standard and double reading and computer-aided detection (CAD) of interval cancers at prior negative screening mammograms: blind review

Open access image repositories: high-quality data to enable machine learning research

Originally motivated by the need for research reproducibility and data reuse, large-scale, open access information repositories have become key resources for training and testing of advanced machine learning applications in biomedical and clinical research. To be of value, such repositories must provide large, high-quality data sets, where quality is defined as minimising variance due to data collection protocols and data misrepresentations. Curation is the key to quality. We have constructed a large public access image repository, The Cancer Imaging Archive, dedicated to the promotion of open science to advance the global effort to diagnose and treat cancer. Drawing on this experience and our experience in applying machine learning techniques to the analysis of radiology and pathology image data, we will review the requirements placed on such information repositories by state-of-the-art machine learning applications and how these requirements can be met.

Identifying pulmonary nodules or masses on chest radiography using deep learning: external validation and strategies to improve clinical practice

AIM

To test the diagnostic performance of a deep learning-based system for the detection of clinically significant pulmonary nodules/masses on chest radiographs.

MATERIALS AND METHODS

Using a retrospective study of 100 patients (47 with clinically significant pulmonary nodules/masses and 53 control subjects without pulmonary nodules), two radiologists verified clinically significantly pulmonary nodules/masses according to chest computed tomography (CT) findings. A computer-aided diagnosis (CAD) software using a deep-learning approach was used to detect pulmonary nodules/masses to determine the diagnostic performance in four algorithms (heat map, abnormal probability, nodule probability, and mass probability).

RESULTS

A total of 100 cases were included in the analysis. Among the four algorithms, mass algorithm could achieve a 76.6% sensitivity (36/47, 11 false negative) and 88.68% specificity (47/53, six false-positive) in the detection of pulmonary nodules/masses at the optimal probability score cut-off of 0.2884. Compared to the other three algorithms, mass probability algorithm had best predictive ability for pulmonary nodule/mass detection at the optimal probability score cut-off of 0.2884 (AUC^Mass: 0.916 versus AUC^{Heat map}: 0.682, p<0.001; AUC^Mass: 0.916 versus AUC^Abnormal: 0.810, p=0.002; AUC^Mass: 0.916 versus AUC^Nodule: 0.813, p=0.014).

CONCLUSION

In conclusion, the deep-learning based computer-aided diagnosis system will likely play a vital role in the early detection and diagnosis of pulmonary nodules/masses on chest radiographs. In future applications, these algorithms could support triage workflow via double reading to improve sensitivity and specificity during the diagnostic process.

The diagnostic value of three-dimensional CT angiography for patients with acute coronary artery disease

Computed tomography angiography (CTA) is an efficient method for the diagnosis of heart disease. However, few contemporary studies have evaluated the prognostic value of three-dimensional (3D)-CTA for patients with acute coronary artery disease. The aim of the present study was to investigate the diagnostic value of 3D-CTA for patients with acute coronary artery disease. A total of 136 patients with suspected acute coronary artery disease were recruited and received conventional coronary angiography (CCA) and 3D-CTA. 3D-CTA was used to assess calcified plaques in the coronary arteries (CCTA), the ratio of calcified plaque volume to vessel circumference (RVTC) and diagnostic accuracy. The results revealed that 3D-CTA was a more effective diagnostic method for identifying calcified plaques in patients with acute coronary artery disease compared with CCA. 3D-CTA demonstrated a significantly better area under curve, sensitivity, specificity, positive predictive value and negative predictive value compared with CCA (P<0.01). In the present study, 3D-CTA was used to successfully diagnose 86 patients with acute coronary artery disease, 34 with myocardial infarction and 16 with stable angina. 3D-CTA images clearly showed global noise levels and target-to-background ratios determined by manually delineated coronary plaque lesions compared with CCA. Furthermore, 3D-CTA was significantly better for discriminating ischemia compared with CCA (P<0.01). In conclusion, the results of the present study suggest that 3D-CTA provides superior diagnostic performance compared with CCA alone in patients with acute coronary artery disease.

Interval Cancers in Breast Cancer Screening: Comparison of Stage and Biological Characteristics with Screen-Detected Cancers or Incident Cancers in the Absence of Screening

Aims and background

To analyze stage distribution and biological features of interval cancers observed in Verona mammography screening compared to screen-detected cancers and “clinical” cancers occurring in the absence of screening, as provided by the Veneto Cancer Registry.

Methods and study design

Screen-detected cancers were identified in the screening archives. Interval cancers and clinical cancers (occurring in women never screened or not yet invited) were identified through the local cancer registry. Studied variables were age, stage, pathological pT and pN category, histological grading, estrogen and progesterone receptor status, and proliferation index (Ki67).

Results

We compared 95 interval cancers, 761 screen-detected cancers, and 1873 clinical cancer cases. Interval cancers had more aggressive features than screen-detected cancers, the difference being statistically significant for pT (P = 10–6), pN (P = 0.0003), grading (P = 0.007), estrogen receptors (P = 0.0006), and progesterone receptors (P = 0.00005), but not for Ki67 (P = 0.18). The features of interval cancers were not more aggressive than those of clinical cancers for pT (P = 0.84), pN (P = 0.33), grading (P = 0.61), estrogen receptors (P = 0.48), and progesterone receptors (P = 0.69), and were better for Ki67 (P = 0.02). In contrast, screen-detected cancers showed significantly better features than clinical cancers, for all studied variables: pT (P = 10–6), pN (P = 10–6), grading (P = 10–6), estrogen receptors (P = 10–5), progesterone receptors (P = 10–6), and Ki67 (P = 10–6).

Conclusions

Our findings are consistent with the length biased sampling hypothesis of interval cancers having a faster growth rate and a less favorable presentation than screen-detected cancers. Compared to clinical cancers, interval cancers had similar features, whereas screen-detected cancers had definitely more favorable features. This finding suggests, rather than a faster growth rate for interval cancers, a slower growth rate for screen-detected cancers, which, together with diagnostic anticipation, may explain a certain degree of overdiagnosis.

Changing patterns of microcalcification on screening mammography for prediction of breast cancer

BACKGROUND

The presence of microcalcification on mammography is one of the earliest signs in breast cancer detection. However, it is difficult to distinguish malignant calcifications from benign calcifications. The aim of this study is to evaluate correlation between changing patterns of microcalcification on screening mammography and malignant breast lesions.

METHODS

Medical records and diagnostic images of 67 women who had previously undergone at least two digital mammograms at least 6 months apart and underwent mammography-guided needle localization and surgical excision between 2011 and 2013 were retrospectively reviewed and analyzed.

RESULTS

Breast cancer was detected in the surgical specimens of 20 patients (29.9 %). Annual change of extent of microcalcification on mammography showed statistically significant correlation with pathologic outcome (P = 0.023). The changing pattern of new appearance or increased extent of microcalcification on mammography had positive predictive value of 54.8 % for breast cancer, and it was a statistically significant predictor for breast cancer (P = 0.012). Shape or number change of microcalcification without increased extent had less accurate predictive value for breast cancer, particularly in women younger than 50 years (P < 0.001).

CONCLUSIONS

This study showed that the pattern of increased extent of microcalcification on screening mammography was a significant predictor for breast cancer. We suggest that mammography-guided needle localization and surgical excision should be considered when increased extent of microcalcification is observed on screening mammography and closed follow-up without pathologic confirmation can be permitted if absence of extension of microcalcification was confirmed in women younger than 50 years.

Second opinion in breast pathology: policy, practice and perception

AIMS

To assess the laboratory policies, pathologists' clinical practice and perceptions about the value of second opinions for breast pathology cases among pathologists practising in the USA.

METHODS

Cross-sectional data were collected from 252 pathologists who interpret breast specimens in eight states using a web-based survey. Descriptive statistics were used to characterise findings.

RESULTS

Most participants had >10 years of experience interpreting breast specimens (64%), were not affiliated with academic centres (73%) and were not considered experts by their peers (79%). Laboratory policies mandating second opinions varied by diagnosis: invasive cancer 65%; ductal carcinoma in situ (DCIS) 56%; atypical ductal hyperplasia 36% and other benign cases 33%. 81% obtained second opinions in the absence of policies. Participants believed they improve diagnostic accuracy (96%) and protect from malpractice suits (83%), and were easy to obtain, did not take too much time and did not make them look less adequate. The most common (60%) approach to resolving differences between the first and second opinion is to ask for a third opinion, followed by reaching a consensus.

CONCLUSIONS

Laboratory-based second opinion policies vary for breast pathology but are most common for invasive cancer and DCIS cases. Pathologists have favourable attitudes towards second opinions, adhere to policies and obtain them even when policies are absent. Those without a formal policy may benefit from supportive clinical practices and systems that help obtain second opinions.

Exact calculations of average power for the Benjamini-Hochberg procedure

Exact analytic expressions are developed for the average power of the Benjamini and Hochberg false discovery control procedure. The result is based on explicit computation of the joint probability distribution of the total number of rejections and the number of false rejections, and expressed in terms of the cumulative distribution functions of the p-values of the hypotheses. An example of analytic evaluation of the average power is given. The result is confirmed by numerical experiments and applied to a meta-analysis of three clinical studies in mammography.

CAD May Not be Necessary for Microcalcifications in the Digital era, CAD May Benefit Radiologists for Masses

Objective:

The aim of this study was to evaluate the effectiveness of computer-aided detection (CAD) to mark the cancer on digital mammograms at the time of breast cancer diagnosis and also review retrospectively whether CAD marked the cancer if visible on any available prior mammograms, thus potentially identifying breast cancer at an earlier stage. We sought to determine why breast lesions may or may not be marked by CAD. In particular, we analyzed factors such as breast density, mammographic views, and lesion characteristics.

Materials and Methods:

Retrospective review from 2004 to 2008 revealed 3445 diagnosed breast cancers in both symptomatic and asymptomatic patients; 1293 of these were imaged with full field digital mammography (FFDM). After cancer diagnosis, in a retrospective review held by the radiologist staff, 43 of these cancers were found to be visible on prior-year mammograms (false-negative cases); these breast cancer cases are the basis of this analysis. All cases had CAD evaluation available at the time of cancer diagnosis and on prior mammography studies. Data collected included patient demographics, breast density, palpability, lesion type, mammographic size, CAD marks on current- and prior-year mammograms, needle biopsy method, pathology results (core needle and/or surgical), surgery type, and lesion size.

Results:

On retrospective review of the mammograms by the staff radiologists, 43 cancers were discovered to be visible on prior-year mammograms. All 43 cancers were masses (mass classification included mass, mass with calcification, and mass with architectural distortion); no pure microcalcifications were identified in this cohort. Mammograms with CAD applied at the time of breast cancer diagnosis were able to detect 79% (34/43) of the cases and 56% (24/43) from mammograms with CAD applied during prior year(s). In heterogeneously dense/extremely dense tissue, CAD marked 79% (27/34) on mammograms taken at the time of diagnosis and 56% (19/34) on mammograms with CAD applied during the prior year(s). At time of diagnosis, CAD marked lesions in 32% (11/34) on the craniocaudal (CC) view, 21% (7/34) on the mediolateral oblique (MLO) view. Lesion size of those marked by CAD or not marked were similar, the average being 15 and 12 mm, respectively.

Conclusion:

CAD marked cancers on mammograms at the time of diagnosis in 79% of the cases and in 56% of the cases from the mammograms with CAD applied in the prior year(s). Our review demonstrated that CAD can mark invasive breast carcinomas in even dense breast tissue. CAD marked a significant portion on the CC view only, which may be an indicator to radiologists to be especially vigilant when a lesion is marked on this view.

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.

Computer-aided detection system performance on current and previous digital mammograms in patients with contralateral metachronous breast cancer

BACKGROUND

The computer-aided detection (CAD) system is widely used for screening mammography. The performance of the CAD system for contralateral breast cancer has not been reported for women with a history of breast cancer.

PURPOSE

To retrospectively evaluate the performance of a CAD system on current and previous mammograms in patients with contralateral metachronous breast cancer.

MATERIAL AND METHODS

During a 3-year period, 4945 postoperative patients had follow-up examinations, from whom we selected 55 women with contralateral breast cancers. Among them, 38 had visible malignant signs on the current mammograms. We analyzed the sensitivity and false-positive marks of the system on the current and previous mammograms according to lesion type and breast density.

RESULTS

The total visible lesion components on the current mammograms included 27 masses and 14 calcifications in 38 patients. The case-based sensitivity for all lesion types was 63.2% (24/38) with false-positive marks of 0.71 per patient. The lesion-based sensitivity for masses and calcifications was 59.3% (16/27) and 71.4% (10/14), respectively. The lesion-based sensitivity for masses in fatty and dense breasts was 68.8% (11/16) and 45.5% (5/11), respectively. The lesion-based sensitivity for calcifications in fatty and dense breasts was 100.0% (3/3) and 63.6% (7/11), respectively. The total visible lesion components on the previous mammograms included 13 masses and three calcifications in 16 patients, and the sensitivity for all lesion types was 31.3% (5/16) with false-positive marks of 0.81 per patient. On these mammograms, the sensitivity for masses and calcifications was 30.8% (4/13) and 33.3% (1/3), respectively. The sensitivity in fatty and dense breasts was 28.6% (2/7) and 33.3% (3/9), respectively.

CONCLUSION

In the women with a history of breast cancer, the sensitivity of the CAD system in visible contralateral breast cancer was lower than in most previous reports using the same CAD system probably due to the relatively small size, subtlety of the lesion findings, and dense parenchymal pattern.

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].

Detection of architectural distortion in prior mammograms

We present methods for the detection of sites of architectural distortion in prior mammograms of interval-cancer cases. We hypothesize that screening mammograms obtained prior to the detection of cancer could contain subtle signs of early stages of breast cancer, in particular, architectural distortion. The methods are based upon Gabor filters, phase portrait analysis, a novel method for the analysis of the angular spread of power, fractal analysis, Laws' texture energy measures derived from geometrically transformed regions of interest (ROIs), and Haralick's texture features. With Gabor filters and phase portrait analysis, 4224 ROIs were automatically obtained from 106 prior mammograms of 56 interval-cancer cases, including 301 true-positive ROIs related to architectural distortion, and from 52 mammograms of 13 normal cases. For each ROI, the fractal dimension, the entropy of the angular spread of power, 10 Laws' measures, and Haralick's 14 features were computed. The areas under the receiver operating characteristic curves obtained using the features selected by stepwise logistic regression and the leave-one-ROI-out method are 0.76 with the Bayesian classifier, 0.75 with Fisher linear discriminant analysis, and 0.78 with a single-layer feed-forward neural network. Free-response receiver operating characteristics indicated sensitivities of 0.80 and 0.90 at 5.8 and 8.1 false positives per image, respectively, with the Bayesian classifier and the leave-one-image-out method.

Computer-aided detection of architectural distortion in prior mammograms of interval cancer

Architectural distortion is an important sign of breast cancer, but because of its subtlety, it is a common cause of false-negative findings on screening mammograms. This paper presents methods for the detection of architectural distortion in mammograms of interval cancer cases taken prior to the detection of breast cancer using Gabor filters, phase portrait analysis, fractal analysis, and texture analysis. The methods were used to detect initial candidates for sites of architectural distortion in prior mammograms of interval cancer and also normal control cases. A total of 4,224 regions of interest (ROIs) were automatically obtained from 106 prior mammograms of 56 interval cancer cases, including 301 ROIs related to architectural distortion, and from 52 prior mammograms of 13 normal cases. For each ROI, the fractal dimension and Haralick's texture features were computed. Feature selection was performed separately using stepwise logistic regression and stepwise regression. The best results achieved, in terms of the area under the receiver operating characteristics curve, with the features selected by stepwise logistic regression are 0.76 with the Bayesian classifier, 0.73 with Fisher linear discriminant analysis, 0.77 with an artificial neural network based on radial basis functions, and 0.77 with a support vector machine. Analysis of the performance of the methods with free-response receiver operating characteristics indicated a sensitivity of 0.80 at 7.6 false positives per image. The methods have good potential in detecting architectural distortion in mammograms of interval cancer cases.

Randomized trial of a lay health advisor and computer intervention to increase mammography screening in African American women

BACKGROUND

Low-income African American women face numerous barriers to mammography screening. We tested the efficacy of a combined interactive computer program and lay health advisor intervention to increase mammography screening.

METHODS

In this randomized, single blind study, participants were 181 African American female health center patients of ages 41 to 75 years, at < or =250% of poverty level, with no breast cancer history, and with no screening mammogram in the past 15 months. They were assigned to either (a) a low-dose comparison group consisting of a culturally appropriate mammography screening pamphlet or (b) interactive, tailored computer instruction at baseline and four monthly lay health advisor counseling sessions. Self-reported screening data were collected at baseline and 6 months and verified by medical record.

RESULTS

For intent-to-treat analysis of primary outcome (medical record-verified mammography screening, available on all but two participants), the intervention group had increased screening to 51% (45 of 89) compared with 18% (16 of 90) for the comparison group at 6 months. When adjusted for employment status, disability, first-degree relatives with breast cancer, health insurance, and previous breast biopsies, the intervention group was three times more likely (adjusted relative risk, 2.7; 95% confidence interval, 1.8-3.7; P < 0.0001) to get screened than the low-dose comparison group. Similar results were found for self-reported mammography stage of screening adoption.

CONCLUSIONS

The combined intervention was efficacious in improving mammography screening in low-income African American women, with an unadjusted effect size (relative risk, 2.84) significantly higher (P < 0.05) than that in previous studies of each intervention alone.

"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.

Evaluation of computer-aided diagnosis on a large clinical full-field digital mammographic dataset

RATIONALE AND OBJECTIVES

To convert and optimize our previously developed computerized analysis methods for use with images from full-field digital mammography (FFDM) for breast mass classification to aid in the diagnosis of breast cancer.

MATERIALS AND METHODS

An institutional review board approved protocol was obtained, with waiver of consent for retrospective use of mammograms and pathology data. Seven hundred thirty-nine FFDM images, which contained 287 biopsy-proven breast mass lesions, of which 148 lesions were malignant and 139 lesions were benign, were retrospectively collected. Lesion margins were delineated by an expert breast radiologist and were used as the truth for lesion-segmentation evaluation. Our computerized image analysis method consisted of several steps: 1) identified lesions were automatically extracted from the parenchymal background using computerized segmentation methods; 2) a set of image characteristics (mathematic descriptors) were automatically extracted from image data of the lesions and surrounding tissues; and 3) selected features were merged into an estimate of the probability of malignancy using a Bayesian artificial neural network classifier. Performance of the analyses was evaluated at various stages of the conversion using receiver-operating characteristic analysis.

RESULTS

An area under the curve value of 0.81 was obtained in the task of distinguishing between malignant and benign mass lesions in a round-robin by case evaluation on the entire FFDM dataset. We failed to show a statistically significant difference (P = .83) compared to results from our previous study in which the computerized classification was performed on digitized screen-film mammograms.

CONCLUSIONS

Our computerized analysis methods developed on digitized screen-film mammography can be converted for use with FFDM. Results show that the computerized analysis methods for the diagnosis of breast mass lesions on FFDM are promising, and can potentially be used to aid clinicians in the diagnostic interpretation of FFDM.

Blinded comparison of computer-aided detection with human second reading in screening mammography

OBJECTIVE

The purpose of this study was to compare a human second reader with computer-aided detection (CAD) for the reduction of false-negative cases by a primary radiologist. We retrospectively reviewed our clinical practice.

MATERIALS AND METHODS

We found that 6,381 consecutive screening mammograms were interpreted by a primary reader. This radiologist then reinterpreted the studies using CAD ("CAD reader"). A second human reader who was blinded to the CAD results but knowledgeable of the primary reader's findings reviewed the studies, looking for abnormalities not seen by the first reader.

RESULTS

Two cancers were called back by the second human reader that were not called back by the CAD reader; however, the CAD system had marked the findings, but they were dismissed by the primary reader. Because of the small numbers, the difference between the CAD and second human reader was not statistically significant. The CAD and human second readers increased the recall rates 6.4% and 7.2% (p = 0.70), respectively, and the biopsy rates 10% and 14.7%. The positive predictive value was 0% (0/3) for the CAD reader and was 40% (2/5) for the human second reader. The relative increases in the cancer detection rate compared with the primary reader's detection rate were 0% for the CAD reader and 15.4% (2/13) for the human second reader (p = 0.50).

CONCLUSION

A human second reader or the use of a CAD system can increase the cancer detection rate, but we found no statistical difference between the two because of the small sample size. A possible benefit from a human second reader is that CAD systems can only point to possible abnormalities, whereas a human must determine the significance of the finding. Having two humans review a study may increase detection rates due to interpreter--hence, perceptual--variability and not just increased detection.

Interval breast cancers in screening: the effect of mammography review method on classification

Surveillance of interval cancers (IC) lacks standardisation of review methodologies. We investigated the extent to which 'informed' or 'blinded' review may affect IC classification. This is a retrospective study of 100 validated screening mammograms (20 IC, 80 negative screens) independently reviewed by six radiologists. Three sequenced review methods with increasing information were used: (1) blinded (no IC information, case mix), (2) partially informed, and (3) fully informed. IC 'screening error' (SE) reports averaged 24% (10-40), 33% (20-55), and 42% (35-50) for phases 1, 2, and 3, while 'minimal signs' (MS) reports averaged 6% (5-15), 10% (10-20), and 20% (15-30), respectively. Negative mammograms classification was MS in 18% (7-39) or SE in 19% (11-29), respectively. MS or SE classification was more likely for method 2 (OR=1.78, p=0.033) and method 3 (OR=3.91, p=0.000) relative to method 1, but no reader effect was evident. Inter-observer agreement in classifying at method 1 was slight (k 0.20), lowest (k 0.06) for MS, and fair (k 0.25) for negative and SE categories. More 'informed' review is more likely to yield an IC classification as MS or SE. Due to expected variability, review methods need standardisation to improve screening quality. Our data support blinded review of IC in mammography screening.

Two-modality mammography may confer an advantage over either full-field digital mammography or screen-film mammography

RATIONALE AND OBJECTIVES

We sought to compare the cancer detection rate and receiver operating characteristic (ROC) area under the curve of full-field digital mammography, screen-film mammography, and a combined technique that allowed diagnosis if a finding was suspicious on film mammography, on digital mammography, or both.

MATERIALS AND METHODS

We used the data originally analyzed by Lewin and associates in 2002. In that trial, 6,736 paired full-field and digital mammograms were performed in 4,489 women. We used parametric and nonparametric tests to compare the area under the curve for ROC scores of film-screen only, digital mammography only, and the combined test. We used McNemar's test for paired proportions to compare the cancer detection rates.

RESULTS

With the parametric test, neither the difference in the area under the curve between the film and combined nor the difference between the digital and combined ROC curves was significant at the Bonferroni-corrected 0.025 alpha level (film versus combined difference = 0.0563, P = .0712; digital versus combined difference = 0.0894, P = .0455). The nonparametric test showed that there was a significant difference between both film and combined (difference = 0.073, P = .008) and digital versus combined ROC curves (difference = 0.1164, P = .0008). The continuity-corrected McNemar's test showed a significant increase in the proportion of cancers detected by the combined modality over film (chi(2) = 7.111, df = 1, P = .0077), and over digital (chi(2) = 12.071, df = 1, P = .0005).

CONCLUSION

Using two mammograms, one film and one digital, significantly increases the detection of breast cancer.

Effect of Computer-Aided Detection on Independent Double Reading of Paired Screen-Film and Full-Field Digital Screening Mammograms

OBJECTIVE

The purpose of this study was to evaluate the performance and potential contribution of computer-aided detection (CAD) to independent double reading of paired screen-film and full-field digital screening mammograms.

MATERIALS AND METHODS

The cases of 3,683 women who underwent both screen-film mammography and full-field digital mammography (FFDM) with independent double reading for each technique were followed for 2 years to include cancers detected in the interval between screening rounds and cancers detected at the next screening round. Fifty-five biopsy-proven cancers were diagnosed. The baseline screening mammograms of the 55 cancers were defined as having positive findings if at least one of two independent readers scored it 2 or higher on a 5-point rating scale. The baseline mammograms of interval (n = 10) or secondround (n = 16) cancers were retrospectively classified as overlooked (n = 2), minimal sign actionable (n = 8), minimal sign nonactionable (n = 5), and normal (n = 11). The baseline mammograms of these cases of cancer were evaluated with a CAD system, and the CAD results were compared (McNemar's test for paired proportions) with the findings at prospective independent double reading of mammograms obtained with each technique.

RESULTS

For FFDM, CAD sensitivity was 95% (37/39) compared with 64% (25/39) for double reading (p = 0.006), and for screen-film mammography, CAD sensitivity was 85% (33/39) compared with 77% (30/39) for prospective double reading (p = 0.57) of radiographically visible lesions in baseline mammograms. CAD correctly marked five (13%) of 39 cancers on screen-film mammography and 14 (36%) of 39 cancers on FFDM not detected at prospective independent double reading.

CONCLUSION

CAD showed the potential to increase the cancer detection rate for FFDM and for screen-film mammography in breast cancer screening performed with independent double reading.

Does the accuracy of single reading with CAD (computer-aided detection) compare with that of double reading?: A review of the literature

AIM

To examine current evidence to determine whether the accuracy of single reading with computed-aided detection (CAD) compares with that of double reading.

METHODS

We performed a literature review to identify studies where both protocols had been investigated and compared. We identified eight studies that compared single reading with CAD against double reading, of which six reported on comparisons of both sensitivity and specificity.

RESULTS

Of the six studies identified, three showed no differences in either sensitivity or specificity. One showed single reading with CAD had a higher sensitivity at the same specificity, another that single reading with CAD had a higher specificity at the same sensitivity. However, one study, in a real-life setting, showed that single reading with CAD had a higher sensitivity but a lower specificity.

CONCLUSION

As the majority of the studies were not in a real-life setting, used test sets, lacked sufficient training in the use of CAD and simulated double reading (using a protocol of recall if one suggests), current evidence is therefore limited as to the accuracy, in terms of sensitivity and specificity, of single reading with CAD in comparison with the most common practice in the UK of double reading using a protocol of consensus or arbitration.

Testing the effect of computer-assisted detection on interpretive performance in screening mammography

OBJECTIVE

The objective of our study was to test whether the use of computer-assisted detection (CAD) improves sensitivity at no cost to specificity for the detection of breast cancer and enables more accurate assessment of fatty breast tissue compared with dense breast tissue.

MATERIALS AND METHODS

We created a stratified random sample of screening mammograms weighted with difficult cases split evenly among women with fatty breast tissue and those with dense breast tissue: 114 patients were cancer-free, 114 had cancer 1 year after screening, and 113 had cancer 13-24 months after screening. In test settings 6 months apart, 19 community radiologists interpreted 341 bilateral screening mammograms with and without CAD. We compared the sensitivity and specificity using regression models adjusting for repeated measures.

RESULTS

CAD assistance did not affect overall sensitivity (cancer by 1 year: 63.2% without CAD and 62.0% with CAD; cancer in 13-24 months: 33.5% without CAD and 32.3% with CAD), but its effect differed for visible masses that were marked by CAD compared with those that were not marked by CAD (hereafter referred to as "unmarked"). CAD was associated with improved sensitivity for marked visible cancers and decreased sensitivity for unmarked visible masses; the sensitivities without and with CAD, respectively, were as follows: marked cancer by 1 year, 82.7% versus 83.1%; marked cancer in 13-24 months, 44.2% versus 57.9%; unmarked cancer by 1 year, 37.4% versus 30.1%; unmarked cancer in 13-24 months, 29.7% versus 23.0% (p < 0.03 for both interactions between assistance and CAD marking for cancer by 1 year and cancer in 13-24 months). CAD marked 77% (70/91) of the visible cancers by 1 year and 67.3% (37/55) of the visible cancers in 13-24 months. CAD marked more visible calcified lesions (86%) than masses and asymmetric densities (67%) (p < 0.05). Overall specificity was 72% without and 75% with CAD (p < 0.02). CAD had a greater effect on both specificity (p < 0.02) and sensitivity (p < 0.03) among radiologists who interpret more than 50 mammograms per week. The results were the same for fatty breast tissue and dense breast tissue.

CONCLUSION

In this experiment, CAD increased interpretive specificity but did not affect sensitivity because visible noncalcified lesions that went unmarked by CAD were less likely to be assessed as abnormal by radiologists. Breast density did not affect CAD's performance.

Prospective Assessment of Computer-Aided Detection in Interpretation of Screening Mammography

OBJECTIVE

The purpose of this study was to prospectively assess the usefulness of computer-aided detection (CAD) in the interpretation of screening mammography and to provide the true sensitivity and specificity of this technique in a clinical setting.

SUBJECTS AND METHODS

Over a 26-month period, 5,016 screening mammograms were interpreted without, and subsequently with, the assistance of the iCAD MammoReader detection system. Data collected for actionable findings included dominant feature (calcification, mass, asymmetry, architectural distortion), detection method (radiologist only, CAD only, or both radiologist and CAD), BI-RADS assessment code, associated histopathology for those undergoing biopsy, and tumor stage for malignant lesions. The study population was cross-checked against an independent reference standard to identify false-negative cases.

RESULTS

Of the 5,016 cases, the recall rate increased from 12% to 14% with the addition of CAD. Of the 107 (2%) patients who underwent biopsy, 101 (94%) were prompted by the radiologist and six (6%) were prompted by CAD. Of the 124 biopsies performed on actionable findings in the 107 patients, findings in 79 (64%) were benign and in 45 (36%) were in situ or invasive carcinoma. Three study participants who were not recalled by the radiologist with the assistance of CAD developed cancer within 1 year of the screening mammogram and were considered to be false-negative cases. The radiologist detected 43 (90%) of the 48 total malignancies and 45 (94%) of the 48 malignancies with the assistance of CAD. CAD missed eight cancers that were detected by the radiologist, which presented as architectural distortions (n = 3), irregular masses (n = 4), and a circumscribed mass (n = 1). CAD detected two in situ cancers as a faint cluster of calcifications that had not been perceived by the radiologist and one mass that was dismissed by the radiologist, accounting for at least a 4.7% increase in cancer detection rate. Sensitivity of screening mammography with the use of CAD (94%) represented an absolute and relative 4% increase over the sensitivity of the radiologist alone (90%). Specificity of screening mammography with and without the use of CAD was 99%.

CONCLUSION

Routine use of CAD while interpreting screening mammograms significantly increases recall rates, has no significant effect on positive predictive value for biopsy, and can increase cancer detection rate by at least 4.7% and sensitivity by at least 4%. This study provides "true" values for sensitivity and specificity for use of CAD in interpretation of screening mammography as measured prospectively in the context of a working clinical setting.

Ductal Carcinoma in Situ: Computer-aided Detection in Screening Mammography

PURPOSE

To retrospectively evaluate the sensitivity of computer-aided detection (CAD) in depicting ductal carcinoma in situ (DCIS) on screening mammograms by using biopsy proved lesion location as the reference standard.

MATERIALS AND METHODS

Institutional review board approval was obtained, with a waiver of patient informed consent for this HIPAA-compliant study. Findings of all image-guided biopsies with a pathologic diagnosis of DCIS during a 1-year period were reviewed. Fifty-eight lesions in 55 women (average age, 61.41 years +/- 12.89 [standard deviation]) were available for review. The screening mammogram of the affected breast and, if available, the prior screening mammogram were digitized by the CAD system. An assessment was then made as to whether the CAD system marked the area of DCIS on the current and prior mammograms. Patient age, location and mammographic size of the lesion, type of lesion, and breast density were recorded and were analyzed by using chi2, Fisher exact, or Cochran-Mantel-Haenzel tests, where applicable.

RESULTS

CAD identified DCIS in 53 (91%) of 58 lesions on craniocaudal (CC) and mediolateral oblique (MLO) views of screening mammograms obtained in the year of the diagnosis. On screening mammograms obtained prior to the year of the diagnosis (34 patients), no radiologically or CAD-detected lesion was present on 11 (32%) of 34 mammograms. CAD identified DCIS in 16 (70%) of 23 lesions on one of the two views. Seven (30%) of 23 lesions had mammographic findings at retrospective review that were not identified with CAD.

CONCLUSION

CAD had a high sensitivity in the depiction of DCIS.

Computer-aided detection (CAD) of cancers detected on double reading by one reader only

We evaluated the role of computer-aided detection (CAD) in cancers undergoing double reading and detected by one reader only. A series of 33 cancers, originally missed by the first reader and detected by the second reader, and 75 negative controls were processed to assess CAD sensitivity, and was read by the six radiologists who originally missed the cancers with the help of CAD printouts. CAD case-based sensitivity, specificity and positive predictive value were 51.5%, 18.6% and 21.7%, respectively. Average sensitivity of all radiologists in all cancers in the series was 74.7%, being higher for CAD+ (86.2%) than for CAD- (62.5%) cancers (P<0.01). When reading cancer cases that they had originally missed, radiologists had a sensitivity of 75.8%, which was higher for CAD+ (100.0%) than for CAD- (58.3%) cancers. The average recall rate was 14.2%, the majority of recalls (45 out of 64) occurring for lesions marked by CAD. CAD may help in detecting at most half of cancers missed at a single reading but detected by a second reader.

Liver Metastases on Serial Contrast-enhanced Multidetector Computed Tomography Examinations

OBJECTIVE

To verify the earliest detectability of liver metastases in patients who underwent serial multidetector computed tomography (MDCT) examinations.

METHODS

We selected 12 patients with known primary cancer who underwent 4 or more contrast-enhanced, 4-detector MDCTs. When metastases had been reported, an evaluation of the preceding MDCT was done to define whether the lesion was detectable, detectable only by minimal signs, undetectable, or detected but misdiagnosed as a benign lesion (MBL).

RESULTS

Eighty-eight lesions were analyzed. Evaluating the preceding examination, we defined detectable (n=8), detectable only by minimal signs (n=5), undetectable (n=74), and MBL (n=1). The group with minimal signs was composed of 4 small hypodense foci and 1 calcification. The MBL was a non-Hodgkin lesion first misdiagnosed as a hemangioma.

CONCLUSION

Approximately 15% of liver metastases were prospectively missed, 9% of them being retrospectively detectable, 6% being retrospectively visible as minimal signs, whereas only 1% of liver metastases were misdiagnosed as a benign lesion.

Prediction of breast cancer using volatile biomarkers in the breath

We evaluated a breath test for volatile organic compounds (VOCs) as a predictor of breast cancer. Breath VOCs were assayed in 51 asymptomatic women with abnormal mammograms and biopsy-proven breast cancer, and 42 age-matched healthy women. A fuzzy logic model predicted breast cancer with accuracy superior to previously reported findings. Following random assignment to a training set (64) or a prediction set (29), a model was constructed in the training set employing five breath VOCs that predicted breast cancer in the prediction set with 93.8% sensitivity and 84.6% specificity. The same model predicted no breast cancer in 16/50 (32.0%) women with abnormal mammograms and no cancer on biopsy. A two-minute breath test could potentially provide a safe, accurate and painless screening test for breast cancer, but prospective validation studies are required.

CAD for mammography: the technique, results, current role and further developments

CAD systems, developed to assist the radiologist in the detection of suspicious lesions on mammograms, are currently controversially discussed. The highly sensitive detection of malignant structures including priors by CAD is linked with a low specific performance and a high rate of falsely positive markings. This causes controversial results regarding the effect of CAD systems for the diagnosing radiologist. This review aims to give an overview of the current literature, to state the currently discussed controversial results of CAD and to give an outlook on the next developments, which are not limited to senology, but include many other applications of CAD systems in radiology.

Multimodality computerized diagnosis of breast lesions using mammography and sonography

RATIONALE AND OBJECTIVES

The purpose of this study is to investigate the use of computer-extracted features of lesions imaged by means of two modalities, mammography and breast ultrasound, in the computerized classification of breast lesions.

MATERIAL AND METHODS

We performed computerized analysis on a database of 97 patients with a total of 100 lesions (40 malignant, 40 benign solid, and 20 cystic lesions). Mammograms and ultrasound images were available for these breast lesions. There was an average of three mammographic images and two ultrasound images per lesion. Based on seed points indicated by a radiologist, the computer automatically segmented lesions from the parenchymal background and automatically extracted a set of characteristic features for each lesion. For each feature, its value averaged over all images pertaining to a given lesion was input to a Bayesian neural network for classification. We also investigated different approaches to combine image-based features into this by-lesion analysis. In that analysis, mean, maximum, and minimum feature values were considered for all images representing a lesion. We considered performance by using a leave-one-lesion-out approach, based on image features from mammography alone (two to five features), ultrasound alone (three to four features), and a combination of features from both modalities (three to five features total).

RESULTS

For the classification task of distinguishing cancer from other abnormalities in a lesion-based analysis by using a single modality, areas under the receiver operating characteristic curves (A(z) values) increased significantly when the computer selected the manner (mean, minimum, or maximum) in which image-based features were combined into lesion-based features. The highest performance was found for lesion-based analysis and automated feature selection from mean, maximum, and minimum values of features from both modalities (resulting in a total of four features being used). That A(z) value for the task of distinguishing cancer was 0.92, showing a statistically significant increase over that achieved with features from either mammography or ultrasound alone.

CONCLUSION

Computerized classification of cancer significantly improved when lesion features from both modalities were combined. Classification performance depended on specific methods for combining features from multiple images per lesion. These results are encouraging and warrant further exploration of computerized methods for multimodality imaging.

Categorizing breast mammographic density: intra- and interobserver reproducibility of BI-RADS density categories

The inter- and intraobserver agreement (kappa-statistic) in reporting according to Breast Imaging Reporting and Data System (BI-RADS((R))) breast density categories was tested in 12 dedicated breast radiologists reading a digitized set of 100 two-view mammograms. Average intraobserver agreement was substantial (kappa=0.71, range 0.32-0.88) on a four-grade scale (D1/D2/D3/D4) and almost perfect (kappa=0.81, range 0.62-1.00) on a two-grade scale (D1-2/D3-4). Average interobserver agreement was moderate (kappa=0.54, range 0.02-0.77) on a four-grade scale and substantial (kappa=0.71, range 0.31-0.88) on a two-grade scale. Major disagreement was found for intermediate categories (D2=0.25, D3=0.28). Categorization of breast density according to BI-RADS is feasible and consistency is good within readers and reasonable between readers. Interobserver inconsistency does occur, and checking the adoption of proper criteria through a proficiency test and appropriate training might be useful. As inconsistency is probably due to erroneous perception of classification criteria, standard sets of reference images should be made available for training.

Detection of pulmonary nodules at multirow-detector CT: effectiveness of double reading to improve sensitivity at standard-dose and low-dose chest CT

The purpose of this study was to assess the effectiveness of double reading to increase the sensitivity of lung nodule detection at standard-dose (SDCT) and low-dose multirow-detector CT (LDCT). SDCT (100 mAs effective tube current) and LDCT (20 mAs) of nine patients with pulmonary metastases were obtained within 5 min using four-row detector CT. Softcopy images reconstructed with 5-mm slice thickness were read by three radiologists independently. Images with 1.25-mm slice thickness served as the gold standard. Sensitivity was assessed for single readers and combinations. The effectiveness of double reading was expressed as the increase of sensitivity. Average sensitivity for detection of 390 nodules (size 3.9+/-3.2 mm) for single readers was 0.63 (SDCT) and 0.64 (LDCT). Double reading significantly increased sensitivity to 0.74 and 0.79, respectively. No significant difference between sensitivity at SDCT and LDCT was observed. The percentage of nodules detected by all three readers concordantly was 52% for SDCT and 47% for LDCT. Although double reading increased the detection rate of pulmonary nodules from 63% to 74-79%, a considerable proportion of nodules remained undetected. No difference between sensitivities at LDCT and SDCT for detection of small nodules was observed.