AI-Enhanced Diagnosis of Challenging Lesions in Breast MRI: A Methodology and Application Primer

Computer-aided diagnosis (CAD) systems have become an important tool in the assessment of breast tumors with magnetic resonance imaging (MRI). CAD systems can be used for the detection and diagnosis of breast tumors as a "second opinion" review complementing the radiologist's review. CAD systems have many common parts, such as image preprocessing, tumor feature extraction, and data classification that are mostly based on machine-learning (ML) techniques. In this review article, we describe applications of ML-based CAD systems in MRI covering the detection of diagnostically challenging lesions of the breast such as nonmass enhancing (NME) lesions, and furthermore discuss how multiparametric MRI and radiomics can be applied to the study of NME, including prediction of response to neoadjuvant chemotherapy (NAC). Since ML has been widely used in the medical imaging community, we provide an overview about the state-of-the-art and novel techniques applied as classifiers to CAD systems. The differences in the CAD systems in MRI of the breast for several standard and novel applications for NME are explained in detail to provide important examples, illustrating: 1) CAD for detection and diagnosis, 2) CAD in multiparametric imaging, 3) CAD in NAC, and 4) breast cancer radiomics. We aim to provide a comparison between these CAD applications and to illustrate a global view on intelligent CAD systems based on machine and deep learning in MRI of the breast. LEVEL OF EVIDENCE: 2 TECHNICAL EFFICACY STAGE: 2.

The supplemental value of mammographic screening over breast MRI alone in BRCA2 mutation carriers

Purpose

BRCA2 mutation carriers are offered annual breast screening with MRI and mammography. The aim of this study was to investigate the supplemental value of mammographic screening over MRI screening alone.

Methods

In this multicenter study, proven BRCA2 mutation carriers, who developed breast cancer during screening using both digital mammography and state-of-art breast MRI, were identified. Clinical data were reviewed to classify cases in screen-detected and interval cancers. Imaging was reviewed to assess the diagnostic value of mammography and MRI, using the Breast Imaging and Data System (BI-RADS) classification allocated at the time of diagnosis.

Results

From January 2003 till March 2019, 62 invasive breast cancers and 23 ductal carcinomas in situ were diagnosed in 83 BRCA2 mutation carriers under surveillance. Overall screening sensitivity was 95.2% (81/85). Four interval cancers occurred (4.7% (4/85)). MRI detected 73 of 85 breast cancers (sensitivity 85.8%) and 42 mammography (sensitivity 49.9%) (p < 0.001).

Eight mammography-only lesions occurred. In 1 of 17 women younger than 40 years, a 6-mm grade 3 DCIS, retrospectively visible on MRI, was detected with mammography only in a 38-year-old woman. The other 7 mammography-only breast cancers were diagnosed in women aged 50 years and older, increasing sensitivity in this subgroup from 79.5% (35/44) to 95.5% (42/44) (p ≤ 0.001).

Conclusions

In BRCA2 mutation carriers younger than 40 years, the benefit of mammographic screening over MRI was very small. In carriers of 50 years and older, mammographic screening contributed significantly. Hence, we propose to postpone mammographic screening in BRCA2 mutation carriers to at least age 40.

Dynamical Graph Theory Networks Methods for the Analysis of Sparse Functional Connectivity Networks and for Determining Pinning Observability in Brain Networks

Neuroimaging in combination with graph theory has been successful in analyzing the functional connectome. However almost all analysis are performed based on static graph theory. The derived quantitative graph measures can only describe a snap shot of the disease over time. Neurodegenerative disease evolution is poorly understood and treatment strategies are consequently only of limited efficiency. Fusing modern dynamic graph network theory techniques and modeling strategies at different time scales with pinning observability of complex brain networks will lay the foundation for a transformational paradigm in neurodegnerative diseases research regarding disease evolution at the patient level, treatment response evaluation and revealing some central mechanism in a network that drives alterations in these diseases. We model and analyze brain networks as two-time scale sparse dynamic graph networks with hubs (clusters) representing the fast sub-system and the interconnections between hubs the slow sub-system. Alterations in brain function as seen in dementia can be dynamically modeled by determining the clusters in which disturbance inputs have entered and the impact they have on the large-scale dementia dynamic system. Observing a small fraction of specific nodes in dementia networks such that the others can be recovered is accomplished by the novel concept of pinning observability. In addition, how to control this complex network seems to be crucial in understanding the progressive abnormal neural circuits in many neurodegenerative diseases. Detecting the controlling regions in the networks, which serve as key nodes to control the aberrant dynamics of the networks to a desired state and thus influence the progressive abnormal behavior, will have a huge impact in understanding and developing therapeutic solutions and also will provide useful information about the trajectory of the disease. In this paper, we present the theoretical framework and derive the necessary conditions for (1) area aggregation and time-scale modeling in brain networks and for (2) pinning observability of nodes in dynamic graph networks. Simulation examples are given to illustrate the theoretical concepts.

Contrast-enhanced dual-energy mammography: a promising new imaging tool in breast cancer detection

Contrast-enhanced dual-energy mammography (CEDM) is a promising new breast imaging tool for breast cancer detection. In CEDM, an iodine-based contrast agent is intravenously administered and subsequently, dual-energy mammography is performed. This results in a set of images containing both a regular mammogram and an image that contains contrast enhancement information. Preliminary studies have indicated that CEDM is superior to conventional mammography and might even match the diagnostic performance of breast MRI. In this review, the imaging technique, protocol and patient handling of CEDM is presented. Furthermore, an overview of current results on CEDM and potential future indications are outlined.

Malignant lesions on mammography: accuracy of two different computer-aided detection systems

We retrospectively compared the accuracy of two computer-aided detection (CAD) systems for the detection of malignant breast lesions on full-field digital mammograms. Mammograms of 326 patients were analyzed (117 patients with breast cancer, 209 negative cases), and each set of cases was read by two CAD systems (Second Look versus AccuDetect Galileo). True-positive fractions per image and case for soft densities, microcalcifications, and total cancers were assessed. Study results showed better overall performance of AccuDetect Galileo (when compared to Second Look) in detecting masses, microcalcifications, and all cancer types, especially in extremely dense breast parenchyma.

Abstract P3-02-08: Is repetition of the contralateral mammogram of patients referred from breast cancer screening for unilateral findings necessary?

Introduction - The Netherlands started a nationwide breast cancer screening program in 1989, including women from the age of 50 until 75. Screening mammograms are performed two yearly in a mobile unit and consist of a bilateral two-view mammogram of the breast (mediolateral oblique and craniocaudal views).

If indicated, patients are referred to a breast clinic for diagnostic analysis. This work-up consist of among others repeating the bilateral two-view mammogram. Since the screening is digitalized, repeating of at least the mammogram of the non suspicious side might be unnecessary.

The aim of this study is to determine the additional value of repeating the contralateral mammogram in patients referred for a suspicious unilateral lesion.

Material and methods – 395 patients were referred from breast screening program to our institution for unilateral findings between October 2009 and August 2011. In all patients a bilateral mammogram was repeated and analyzed by an experienced breast radiologist. In the case of breast cancer a breast magnetic resonance imaging (MRI) was performed for preoperative staging. Anonymised data concerning the date of registration of the screening mammogram, the referred side (left/right or bilateral), age, screening's BI-RADS classification, breast density, biopsy results and breast MRI results were collected.

Results - Of the 395 patients referred for a suspicious unilateral finding, a malignancy on the referred side was observed in 144 patients (36.5%). In five patients bilateral breast cancer was detected. In one patient no malignancy was detected on the referred side, though on the contralateral side. Three of these six contralateral malignancies were directly mammographically detected. All six malignancies were detected with preoperative breast MRI.

Conclusion - Repetition of the two-view mammogram of the contralateral side in patients referred with a unilateral suspicious finding seems unnecessary, since all contralateral malignancies were depicted on the preoperative staging breast MRI recommended according to the EUSOBI-guidelines. Omission of the contralateral mammogram could lead to reduction of associated health care costs, radiation exposure, and patient discomfort caused by the exam.

Citation Information: Cancer Res 2012;72(24 Suppl):Abstract nr P3-02-08.

Noninvasive diagnosis of ruptured peripheral atherosclerotic lesions and myocardial infarction by antibody profiling

Novel biomarkers, such as circulating (auto)antibody signatures, may improve early detection and treatment of ruptured atherosclerotic lesions and accompanying cardiovascular events, such as myocardial infarction. Using a phage-display library derived from cDNAs preferentially expressed in ruptured peripheral human atherosclerotic plaques, we performed serological antigen selection to isolate displayed cDNA products specifically interacting with antibodies in sera from patients with proven ruptured peripheral atherosclerotic lesions. Two cDNA products were subsequently evaluated on a validation series of patients with peripheral atherosclerotic lesions, healthy controls, and patients with coronary artery disease at different stages. Our biomarker set was able to discriminate between patients with peripheral ruptured lesions and patients with peripheral stable plaques with 100% specificity and 76% sensitivity. Furthermore, 93% of patients with an acute myocardial infarction (AMI) tested positive for our biomarkers, whereas all patients with stable angina pectoris tested negative. Moreover, 90% of AMI patients who initially tested negative for troponin T, for which a positive result is known to indicate myocardial infarction, tested positive for our biomarkers upon hospital admission. In conclusion, antibody profiling constitutes a promising approach for noninvasive diagnosis of atherosclerotic lesions, because a positive serum response against a set of 2 cDNA products showed a strong association with the presence of ruptured peripheral atherosclerotic lesions and myocardial infarction.