Breast ultrasound (US) is an effective imaging modality for breast cancer detection and diagnosis. The structural characteristics of breast lesion play an important role in Computer-Aided Diagnosis (CAD). In this paper, a novel structure-aware triplet path networks (SATPN) was designed to integrate classification and two image reconstruction tasks to achieve accurate diagnosis on US images with small training dataset. Specifically, we enhance clinically-approved breast lesion structure characteristics though converting original breast US images to BIRADS-oriented feature maps (BFMs) with a distance-transformation coupled Gaussian filter. Then, the converted BFMs were used as the inputs of SATPN, which performed lesion classification task and two unsu-pervised stacked convolutional Auto-Encoder (SCAE) networks for benign and malignant image reconstruction tasks, independently. We trained the SATPN with an alternative learning strategy by balancing image reconstruction error and classification label prediction error. At the test stage, the lesion label was determined by the weighted voting with reconstruction error and label predic-tion error. We compared the performance of the SATPN with TPN using origi-nal image as input and our previous developed semi-supervised deep learning methods using BFMs as inputs. Experimental results on two breast US datasets showed that SATPN ranked the best among the three networks, with classifica-tion accuracy around 93.5%. These findings indicated that SATPN is promising for effective breast US lesion CAD using small datasets.
|Original language||English (US)|
|State||Published - Aug 9 2019|
- BIRADS Features
- Breast Cancer
- Computer-Aided Diagnosis
- Semi-supervised Deep Learning
ASJC Scopus subject areas