A Matched-Pair Analysis of Nuclear Morphologic Features Between Core Needle Biopsy and Surgical Specimen in Thyroid Tumors Using a Deep Learning Model.

Abstract

Core needle biopsy (CNB) is a method that can be used as an alternative to fine-needle aspiration for the sampling of thyroid nodules to provide a preoperative diagnosis. Nuclear atypia is of paramount importance in diagnosing thyroid tumors. We aimed to identify the differences in nuclear morphologic features between CNB and surgical specimens. Quantitative image analysis was performed on whole slide images (WSI) of paired CNB and surgical specimens from 50 follicular cell-derived tumors including adenoma, non-invasive follicular thyroid neoplasm with papillary-like nuclear features, and carcinoma. The nuclear features of tumor cells were extracted from WSI using a pre-trained deep-learning model. In matched-pair analysis, the ratios (surgical specimens-to-CNB) of the mean nuclear diameters and the areas of the tumor cells were 1:0.75-0.85 and 1:0.54-0.73 according to the tumor types, respectively. The ratio of the nuclear optical density of tumor cells was 1.68-2.44 times higher in CNB specimens than it was in their corresponding surgical specimens. The significant nuclear size reduction and optical density increase in CNB were most prominent in papillary carcinoma cells. Normal follicular cells showed nuclear size reduction in CNB but no change in nuclear optical density between CNB and surgical specimens. Nuclear vacuolar artifacts mimicking nuclear pseudoinclusions were frequently seen in CNB regardless of the tumor types. In conclusion, tumor cells in CNB showed marked nuclear shrinkage, darker nuclear staining, and nuclear vacuolar changes. Awareness of the morphological differences according to the sampling types can facilitate correct diagnosis in thyroid histopathology.