Well differentiated follicular neoplasms of the thyroid: reproducibility and validity of a 'decision tree' classification based on nucleolar and karyometric features.

Abstract

This study was conducted on fine-needle aspirates of well differentiated follicular neoplasms of the thyroid. A 'decision tree' classification based on the percentage of nucleolated nuclei, percentage of nuclei with two or more nucleoli and mean major nuclear diameter was adopted. We observed that the reproducibility and the validity of the follicular adenoma vs follicular carcinoma discrimination are greater than in the subjective evaluation. Moreover, similar classification results were obtained when measurements were performed either with a fully automated image analysis system or with semiautomatic or manual instrumentation. As for reproducibility of the inter-instrument comparisons, the k statistic values ranged from 0.85 to 1.00 (mean value 0.90, that is, an 'almost perfect' degree of agreement); in the subjective evaluations, the inter-observer comparisons showed values ranging from 0.20 to 0.56 (mean value 0.37, that is, 'fair'). In the decision tree classification, feature value thresholds were selected in order to have specificity of 100% and the predictive value of a positive result (carcinoma) of 100%; accuracy was 87% (range 86-89%), sensitivity 74% (71-79%), the predictive value of a negative result (adenoma) 79% (78-82%). In the subjective evaluation the values were as follows: accuracy 67% (64-71%), sensitivity 57% (50-64%), specificity 77% (71-79%), predictive value of a negative result (adenoma) 64% (61-69%), predictive value of a positive result (carcinoma) 71% (67-75%). The conclusion is that, by using a routine microscope equipped with a micrometer, the preoperative diagnosis of follicular carcinoma from smears can be formulated with a high degree of certainty.