Automatic annotation of histopathological images using a latent topic model based on non-negative matrix factorization - PubMed (original) (raw)

Automatic annotation of histopathological images using a latent topic model based on non-negative matrix factorization

Angel Cruz-Roa et al. J Pathol Inform. 2011.

Abstract

Histopathological images are an important resource for clinical diagnosis and biomedical research. From an image understanding point of view, the automatic annotation of these images is a challenging problem. This paper presents a new method for automatic histopathological image annotation based on three complementary strategies, first, a part-based image representation, called the bag of features, which takes advantage of the natural redundancy of histopathological images for capturing the fundamental patterns of biological structures, second, a latent topic model, based on non-negative matrix factorization, which captures the high-level visual patterns hidden in the image, and, third, a probabilistic annotation model that links visual appearance of morphological and architectural features associated to 10 histopathological image annotations. The method was evaluated using 1,604 annotated images of skin tissues, which included normal and pathological architectural and morphological features, obtaining a recall of 74% and a precision of 50%, which improved a baseline annotation method based on support vector machines in a 64% and 24%, respectively.

Keywords: Automatic Annotation; Bag of Features; Basal Cell Carcinoma; Histopathology Images; Non-negative Matrix Factorization; Visual Latent Semantic Analysis.

PubMed Disclaimer

Figures

Figure 1

Figure 1

Example of histopathological images globally annotated with multiple annotations (multilabeled images). These images correspond to the test data set used in this work and they have a resolution of 1024 × 768 pixels. Histopathological annotations of morphological and architectural features such as epidermis, collagen, and hair follicles appear in different images illustrating the high-visual variability for the same annotation

Figure 2

Figure 2

Overview of the proposed method for automatic annotation of histopathological images based on non-negative matrix factorization

Figure 3

Figure 3

Bag of feature setup used for representing histopathology images. In this work the local features extraction is performed using regular grid extraction and each patch of 8 × 8 pixels is represented by the first coefficients of a discrete cosine transform applied to each color component (RGB) independently, the visual codebook is built using k-means with k = 700, and finally each image is represented by a histogram of 700 bins normalized with L1 norm

Figure 4

Figure 4

Examples of training images with the corresponding histopathological annotations. These images have a resolution of 300 × 300 pixels and exhibit only one annotation per image

Figure 5

Figure 5

Performance evaluation on training mono-label images by each number of dimensions in the latent space

Figure 6

Figure 6

Example of an image from the test data set automatically annotated by the proposed method. The original multilabel image (a) is showed with the salient maps of the patches inside the image according with each one of the 10 histopathological annotations: collagen (b), sebaceous glands (c), hair follicles (d), inflammatory infiltration (e), eccrine glands (f), epidermis (g), nodular basal cell carcinoma (h), morpheiform basal cell carcinoma (i), micro-nodular basal cell carcinoma (j), cystic basal cell carcinoma (k), on the top of each salient image is the real membership of the class (v), the conditional probability estimated by the proposed method (p), and the final concept binarization value (b)

Similar articles

Cited by

References

    1. Madabhushi A. Digital pathology image analysis: Opportunities and challenges (Editorial) Imaging Med. 2009;1:7–10. - PMC - PubMed
    1. Lein ES, Hawrylycz MJ, Ao N, Ayres M, Bensinger A, Bernard A, et al. Genome-wide atlas of gene expression in the adult mouse brain. Nature. 2006;445:168–76. - PubMed
    1. Swedlow JR, Goldberg IG, Eliceiri KW. OME consortium. Bioimage informatics for experimental biology. Ann Rev Biophys. 2009;38:327–46. - PMC - PubMed
    1. Kvilekval K, Fedorov D, Obara B, Singh A, Manjunath BS. Bisque: A platform for bioimage analysis and management. Bioinformatics. 2010;26:544–52. - PubMed
    1. Peng H. Bioimage informatics: A new area of engineering biology. Bioinformatics. 2008;24:1827–36. - PMC - PubMed

LinkOut - more resources