Abstract Introduction: The progression of breast cancer (BrCa) involves nuclear morphometric changes, which are used in the pathological diagnosis of breast cancer. Although changes in nuclear morphometry (NM) contribute to histologic alterations observed in breast cancer, an accurate and autonomous quantification of NM changes have remained elusive. Here we created an image analysis macro to quantify changes in nuclear parameters, including size, shape, and DNA content and used these parameters to predict BrCa progression. Materials & Methods: Three tissue microarrays (TMAs) with 140 BrCa cases, stratified by TNM stage were used for this study. The TMAs were generated using 6 mm cores from primary tumors of a Korean cohort from Yengnam University Hospital, Daegu, South Korea. Specimens were obtained from surgical resection between January 1995 and January 2004. Each case was represented by 2 cores on the TMAs. Clinical information were provided by the pathology reports and patients’ medical records. H&E slides were first used for pathological diagnosis of BrCa after which slides were scanned with the Aperio scanner and the image of each core was separated using Aperio ImageScope software. The H&E stained nuclei from the tumor core were quantified using ImagePro® Premier 9.1 software (MediaCybernetics). For each core, data of all regions of interest were pooled and the covariance of each parameter was generated. Data were first analyzed alone and then in combination using multivariate logistic regression (MLR) to predict the aggressive cases or recurrence. For all analysis, p<0.05 was considered statistically significant. Results: Using multivariate logistic regression (MLR) to differentiate aggressive BrCa (TNM T3 & T4) from indolent BrCa (TNM T1 & T2) on the TMAs, our H&E NM model generated an receiver operating characteristic curve-area under the curve (ROC-AUC) of 0.75 with a sensitivity of 50.9% and specificity of 84.3% at the predictive probability cutoff of 0.50. Parameters of nuclei diameter, size, staining intensity, and aspect ratio contributed to the MLR model. In addition, our H&E NM model also showed power in the prediction of recurrence based on the MLR (ROC-AUC = 0.80) with a sensitivity of 25.9% and specificity of 96.3% at the probability cutoff of 0.53. Parameters of nuclei diameter, size, staining intensity, and cluster contributed to the MLR model. The predictive probability (PP) is significantly higher in the aggressive and recurrent cases based on the two MLR models, respectively. Conclusions: We discovered that our automated quantification of tissue nuclear morphometry can be used to accurately predict BrCa aggressiveness and recurrence in women who underwent surgical resection of their primary tumors. Although further confirmatory studies our needed, our data suggests that this tool could provide an accurate predictor of tumor behavior and patient prognosis that could be used to inform therapy for BrCa patients. Citation Format: Neil Carleton, Guangjing Zhu, Linda Resar, Lisa Rooper, Young Kyung Bae, Robert W. Veltri. Prediction of breast cancer progression using nuclear morphometry. [abstract]. In: Proceedings of the 107th Annual Meeting of the American Association for Cancer Research; 2016 Apr 16-20; New Orleans, LA. Philadelphia (PA): AACR; Cancer Res 2016;76(14 Suppl):Abstract nr 3922.
Read full abstract