Detection and classification of thyroid follicular lesions based on nuclear structure from histopathology images

Abstract

Follicular lesions of the thyroid are traditionally difficult and tedious challenges in diagnostic surgical pathology in part due to lack of obvious discriminatory cytological and microarchitectural features. We describe a computerized method to detect and classify follicular adenoma of the thyroid, follicular carcinoma of the thyroid, and normal thyroid based on the nuclear chromatin distribution from digital images of tissue obtained by routine histological methods. Our method is based on determining whether a set of nuclei, obtained from histological images using automated image segmentation, is most similar to sets of nuclei obtained from normal or diseased tissues. This comparison is performed utilizing numerical features, a support vector machine, and a simple voting strategy. We also describe novel methods to identify unique and defining chromatin patterns pertaining to each class. Unlike previous attempts in detecting and classifying these thyroid lesions using computational imaging, our results show that our method can automatically classify the data pertaining to 10 different human cases with 100% accuracy after blind cross validation using at most 43 nuclei randomly selected from each patient. We conclude that nuclear structure alone contains enough information to automatically classify the normal thyroid, follicular carcinoma, and follicular adenoma, as long as groups of nuclei (instead of individual ones) are used. We also conclude that the distribution of nuclear size and chromatin concentration (how tightly packed it is) seem to be discriminating features between nuclei of follicular adenoma, follicular carcinoma, and normal thyroid.

Figure 1

Examples randomly selected from hematoxylin and eosin stained sections of normal thyroid (A), follicular adenoma (B), and follicular carcinoma (C). By visual inspection, the follicular carcinoma appears to have larger nuclei overall than follicular adenoma and normal thyroid tissue, but the distinctions are subtle. (All micrographs taken at ×400). [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

Figure 2

The flow charts of our method. A The flow chart of classifying different lesions as shown. B: The flow chart of detecting and validating unique characteristic chromatin patterns. [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

Figure 3

Examples randomly selected from Feulgen stained sections of follicular adenoma (A) and follicular carcinoma (B). This stain highlights DNA only and therefore stains only the nuclei. No cytoplasmic counterstaining was done. (All micrographs taken at ×1000). [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

Figure 4

Random samples of preprocessed nuclei (after segmentation, gray level extraction, and intensity normalization).

Figure 5

Average classication accuracy groups of nuclei with voting strategy. The SVM based on RBF can reach 100% accuracy when 28 nuclei are used. The SVM based on quadratic kernel also shows a good classification accuracy, but does not reach 100% accuracy until 70 nuclei are used. The MNN classifier does not show a good tendency to reach 100% accuracy until 70 nuclei are used. [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

Figure 6

Two-dimensional representation nuclear population (NT, FTC, and FA), with axis corresponding to directions computed by multidimensional scaling technique excluding the top 5% outliers (see text). The solid shapes indicate the three unique regions corresponding to NT, FTC, and FA that have been identified by our algorithm. [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

Figure 7

Nuclei samples of the unique regions from Fig. 6 are shown in rows A, B, and C corresponding to NT, FA, and FTC, respectively. The p-values of NT nuclei shown in (A) from left to right are 0.04, 0.05, 0.05, 0.06, 0.07. The p-values of FA nuclei shown in (B) from left to right are 0.02, 0.03, 0.04, 0.04, 0.05; The p-values of FTC nuclei shown in (C) from left to right are 0.05, 0.06, 0.07, 0.08, 0.09, approximately.

Figure 8

Histogram of individual features. A: The histogram of the area feature with units of μm². B: The histogram of entropy feature with the units of bit.