Joint Kernel-Based Supervised Hashing for Scalable Histopathological Image Analysis
Histopathology is crucial to diagnosis of cancer, yet its interpretation is tedious and challenging. To facilitate this procedure, content-based image retrieval methods have been developed as case-based reasoning tools. Recently, with the rapid growth of
- PDF / 1,820,494 Bytes
- 8 Pages / 439.363 x 666.131 pts Page_size
- 72 Downloads / 226 Views
Department of Computer Science, Rutgers University, Piscataway, NJ, USA Department of Computer Science, UNC Charlotte, Charlotte, NC, USA 3 Department of Computer Science and Engineering, UT Arlington, Arlington, TX, USA 4 Department of Biomedical Engineering, University of Florida, Gainesville, FL, USA 2
Abstract. Histopathology is crucial to diagnosis of cancer, yet its interpretation is tedious and challenging. To facilitate this procedure, content-based image retrieval methods have been developed as case-based reasoning tools. Recently, with the rapid growth of histopathological images, hashing-based retrieval approaches are gaining popularity due to their exceptional scalability. In this paper, we exploit a joint kernel-based supervised hashing (JKSH) framework for fusion of complementary features. Specifically, hashing functions are designed based on linearly combined kernel functions associated with individual features, and supervised information is incorporated to bridge the semantic gap between low-level features and high-level diagnosis. An alternating optimization method is utilized to learn the kernel combination and hashing functions. The obtained hashing functions compress high-dimensional features into tens of binary bits, enabling fast retrieval from a large database. Our approach is extensively validated on thousands of breast-tissue histopathological images by distinguishing between actionable and benign cases. It achieves 88.1% retrieval precision and 91.2% classification accuracy within 14.0 ms query time, comparing favorably with traditional methods.
1 Introduction For years, histopathology has played a key role in the early diagnosis of breast cancer, which is the second leading cause of cancer-related death among women. Unfortunately, examination of histopathological images is very tedious and error-prone due to their large size, inter- and intra-observer variability among pathologists, and several other factors [11]. To facilitate this procedure, many content-based image retrieval (CBIR) methods have been proposed as computer-aided diagnosis (CAD) tools [1, 3, 13, 14]. These approaches compare a query histopathological image with previously diagnosed cases stored in a database, and return the most similar cases along with the likelihood of abnormality of the query. Compared with classifier-based CAD methods [2, 5], CBIR approaches could provide more clinical evidence to assist the diagnosis. In addition, they can also contribute to digital slide archiving, pathologist training, and various
Corresponding author.
c Springer International Publishing Switzerland 2015 N. Navab et al. (Eds.): MICCAI 2015, Part III, LNCS 9351, pp. 366–373, 2015. DOI: 10.1007/978-3-319-24574-4_44
Joint Kernel-Based Supervised Hashing Database Images
Database Features
Kernel Space
Hashing Functions 0
N
N
1
0
M
N
Offline
¦
M m 1
m m
P N
00
1 10 h1(x)
367
Database Hash Codes
1 hP(x) 01
11
001 … 010 011 … 100 101 … 110
Online Actionable / Benign
101 … 010
Query Image
Query Fea
Data Loading...
