Deep Cascade Classifiers to Detect Clusters of Microcalcifications
Recent advances in Computer-Aided Detection (CADe) for the automatic detection of clustered microcalcifications on mammograms show that cascade classifiers can compete with high-end commercial systems. In this paper, we introduce a deep cascade detector w
- PDF / 1,005,616 Bytes
- 8 Pages / 439.37 x 666.142 pts Page_size
- 17 Downloads / 171 Views
2
DIEI, University of Cassino and Southern Latium, Cassino, FR, Italy {a.bria,c.marrocco,m.molinara,tortorella}@unicas.it DIAG, Radboud University Nijmegen Medical Centre, Nijmegen, The Netherlands [email protected]
Abstract. Recent advances in Computer-Aided Detection (CADe) for the automatic detection of clustered microcalcifications on mammograms show that cascade classifiers can compete with high-end commercial systems. In this paper, we introduce a deep cascade detector where the learning algorithm of each binary pixel classifier has been redesigned in the early stopping mechanism conventionally used to avoid overfitting to the training data. In this way, we strongly increase the number of features considered in each stage of the cascade (hence the term “deep”), yet we still benefit from the cascade framework by obtaining a very fast processing of mammograms (less than one second per image). We evaluated the proposed approach on a database of full-field digital mammograms; the experiments revealed a statistically significant improvement of deep cascade with respect to the traditional cascade framework. We also obtained statistically significantly higher performance than one of the most widespread commercial CADe systems, the Hologic R2CAD ImageChecker. Specifically, at the same number of false positives per image of R2CAD (0.21), the deep cascade detected 96 % of true lesions against the 90 % of R2CAD, whereas at the same lesion sensitivity of R2CAD (90 %), we obtained 0.05 false positives per image for the deep cascade against the 0.21 of R2CAD. Keywords: Computer aided detection · Mammography microcalcifications · Cascade of classifiers
1
·
Clusters of
Introduction
The presence of clusters of microcalcifications (μCs) in mammograms is one of the earliest signs of breast cancer [1]. The small size of μCs and the low contrast of the image make particularly difficult the interpretation of screening mammograms even for an expert radiologist. To this end, the use of ComputerAided Detection (CADe) systems is recently widespread among radiologists to improve their detection performance [2]. CADe systems are usually based on supervised learning techniques that classify each pixel of a mammographic image c Springer International Publishing Switzerland 2016 A. Tingberg et al. (Eds.): IWDM 2016, LNCS 9699, pp. 415–422, 2016. DOI: 10.1007/978-3-319-41546-8 52
416
A. Bria et al.
as belonging to a μC or not [3–7]. These kinds of approaches, however, suffer from the high computational burden generated by the number of pixels to be processed and the high complexity of the classifiers to be employed. An efficient solution to these drawbacks has been proposed in [8] where a multistage system for the automatic detection of clustered μCs in Full-Field Digital Mammograms (FFDM) was introduced. The rationale of their method is to build an ensemble of ranking-based boosting classifiers with increasing complexity and specificity connected in series as in the cascade face detector proposed by Viola and Jones [9]. In this pape
Data Loading...
