Clinical values of nuclear morphometric analysis in fibroepithelial lesions.

Abstract

Fibroepithelial lesions (FELs) of the breast encompass a broad spectrum of lesions, ranging from commonly encountered fibroadenomas (FAs) to rare phyllodes tumors (PTs). Accurately diagnosing and grading these lesions is crucial for making management decisions, but it can be challenging due to their overlapping features and the subjective nature of histological assessment. Here, we evaluated the role of digital nuclear morphometric analysis in FEL diagnosis and prognosis. A digital nuclear morphometric analysis was conducted on 241 PTs and 59 FAs. Immunohistochemical staining for cytokeratin and Leukocyte common antigen (LCA) was used to exclude non-stromal components, and nuclear area, perimeters, calipers, circularity, and eccentricity in the stromal cells were quantified with QuPath software. The correlations of these features with FEL diagnosis and prognosis was assessed. All nuclear features, including area, perimeter, circularity, maximum caliper, minimum caliper and eccentricity, showed significant differences between FAs and benign PTs (p ≤ 0.002). Only nuclear area, perimeter, minimum caliper and eccentricity correlated significantly with PT grading (p ≤ 0.022). For differentiation of FAs from benign PTs, the model integrating all differential nuclear features demonstrated a specificity of 90% and sensitivity of 70%. For PT grading, the nuclear morphometric score showed a specificity of 78% and sensitivity of 96% for distinguishing benign/borderline from malignant PTs. In addition, a relationship of nuclear circularity was found with PT recurrence. The Kaplan-meier analysis, using the best cutoff determined by ROC curve, showed shorter event free survival in benign PTs with high circularity (chi-square = 4.650, p = 0.031). Our data suggested the digital nuclear morphometric analysis could have potentials to objectively differentiate different FELs and predict PT outcome. These findings could provide the evidence-based data to support the development of deep-learning based algorithm on nuclear morphometrics in FEL diagnosis.