Breast MRI lesion classification: Improved performance of human readers with a backpropagation neural network computer‐aided diagnosis (CAD) system

Abstract

Purpose:To develop and test a computer-aided diagnosis (CAD) system to improve the performance of radiologists in classifying lesions on breast MRI (BMRI). Materials and Methods:A CAD system was developed that uses a semiautomated segmentation method. After segmentation, 42 features based on lesion shape, texture, and enhancement kinetics were computed, and the 13 best features were selected and used as inputs to a backpropagation neural network (BNN). The BNN was trained and tested using the leave-one-out method on 80 BMRI lesions (37 benign, 43 malignant). Lesion histopathology was used as the reference standard. Five human readers classified the 80 lesions first without and then with CAD assistance. The performance of the computer classifier and the human readers was assessed using receiver operating characteristic curves; the performance of the human readers was also evaluated using multireader multicase (MRMC) analysis. Results:The performance of the human readers significantly improved when aided by the CAD system (P Ͻ 0.05). MRMC analysis showed that human reader performance with and without CAD system assistance can be generalized to the population of cases (P Ͻ 0.001). Conclusion:A CAD system based on lesion morphology and enhancement kinetics can improve the performance of human readers in classifying lesions on breast MRI. BREAST CANCER is the most common cancer among women (excluding nonmelanoma skin cancer) and the second leading cause of cancer deaths in women after lung cancer (1). Breast MRI (BMRI) has emerged as a promising technique for detecting, diagnosing, and staging breast cancer. BMRI boasts a high sensitivity for detecting enhancing malignant lesions, but its specificity is less favorable (2). Computer-aided diagnosis (CAD) systems offer an intriguing approach to reducing the high false-positive rate and improving the clinical utility of BMRI.Mammography CAD systems have been developed for detecting breast tumors (3,4) and microcalcifications (5). Many classification methods have been developed and applied to characterize mammographic breast masses as benign or malignant. These methods include wavelets (6), fractals (7), statistical methods (8), visionbased methods (9) and, recently, artificial neural networks (ANN) (3-5,10 -16).ANNs have been used to detect and classify microcalcifications (11,12,17) and breast masses (4,14 -16,18 -22) in mammography. However, CAD systems for BMRI are still limited. Starita et al (2) described a web-based CAD system for the analysis of suspected BMRI lesions. They evaluated the performance of their system on 20 MR data sets, and classified the suspected lesions into five different types: benign lesions, intermediate cases with no clear decision (benign vs. malignant), malignant lesions, veins (usually rejected), and unknown regions (vein or lesion). Starita et al (2) reported 11 true-positives out of 12, and five true-negatives out of seven using their CAD classification system. Penn et al (23) described the first and second revisions of a BMRI CAD sys...

Mass classification in breast DCE-MR images using an artificial neural network trained via a bee colony optimization algorithm

“…A typical breast DCE-MRI contains a great amount of heterogeneous information that depicts different tissues, vessels, ducts, chest skin, and breast edge characteristics. Texture features have been widely used in breast DCE-MRI mass classification 7,18,[32][33][34][35][36][37] . The implemented feature extraction procedure relies on the exploration of the textural characteristics of the extracted mass.…”

“…15, which is used to extract the edges of the tumour very efficiently in segmented MR images, followed by proper threshold for extraction the lesion or region of interest (ROI) from edge enhanced segmented breast MR images. After detection of the ROI, statistical analysis techniques are applied to select an optimal set of features to achieve the highest diagnostic accuracy [16][17][18] . The extracted features are fed as input to the classifier to discriminate whether the lesion is benign or malignant.…”

Detection and classification of contrast‐enhancing masses by a fully automatic computer‐assisted diagnosis system for breast MRI

“…For integrating all extracted morphologic and dynamic features, ANNs were used (21,23,24). Feed-forward selfreflexive ANNs had been trained in advance of this investigation on the list of the above-described morphologic and dynamic parameters.…”

“…Although recent developments have focused on the automatic, computerassisted analysis of morphologic MRM features, eg, by the use of a classification tool, such as an artificial neural network (ANN), clinical assessment of MRM is still based on the observer's interpretation of morphology (11,13,15,(21)(22)(23)(24). There is clinical interest to integrate observer-independent morphologic analysis into ubiquitously available CAD systems and to link the computer-extracted features to the BI-RADS descriptors.…”

Computer-Aided Diagnosis of Breast DCE-MRI Images Using Bilateral Asymmetry of Contrast Enhancement Between Two Breasts

“…The CAD schemes first extract and compute dynamic contrast enhancement features from the characteristic kinetic curves generated from a set of the selected pixels located inside the identified and segmented breast lesions, and then apply different machine learning classifiers to distinguish between the malignant and benign lesions [16][17][18][19][20]; for example, one study reported that using a leave-one-case-out testing method, a CAD scheme using a set of selected DCE-MRI features and a Bayesian artificial neural network yielded an area under a receiver operating characteristic (ROC) curve (AUC=0.78±0.04) when applying to a dataset involving 168 malignant and 45 benign breast lesions [21]. Although the previous studies focused on detection and analysis of characteristic kinetic curves computed from the pixels inside the segmented lesions, a recent study also showed that similar to a widely used image-based risk factor, namely, the mammographic density, the background parenchymal enhancement (BPE) evaluated or computed from the entire breast areas depicting on DCE-MRI images also carried useful or higher discriminatory information associated with cancer risk assessment [22] as well as the performance of cancer detection and staging [23].…”