Abstract P5-02-02: Second harmonic generation in combination with nuclear morphometry in the evaluation of DCIS

Abstract

Abstract Purpose/Objective: Collagen is a major extracellular matrix (ECM) constituent in normal breast and is extensively remodeled in breast carcinoma. Therefore, features of remodeled collagen in the stroma adjacent to ductal carcinoma in situ (DCIS) could indicate cancer progression. The major objective of this study is to identify potential tumor-associated collagen signatures unique to DCIS that will allow us to predict progression based on the collagen texture and nuclear morphology. In this present study, we develop two image analysis pipelines (SHG Texture Extraction and H&E Nuclear Morphology Extractor) to quantify 1) stromal changes, 2) collagen signatures and 3) nuclear morphology from normal breast to DCIS in order to predict local breast cancer recurrence. Method: We used second harmonic generation (SHG) images and H&E to analyze collagen features and to study nuclear morphology using a data set of 336 patients (from which 310 normal and 327 DCIS regions were imaged). The 336 patients were a subset of patients with pure DCIS taken from a case-control study. Clinical-pathologic factors were associated with risk of subsequent ipsilateral cancer (DCIS or invasive). The SHG framework consisted of collagen segmentation using 1) adaptive thresholding and 2) morphological operations. The H&E framework consisted of nuclear segmentation using adaptive thresholding and a maker-controlled watershed algorithm; and nuclear feature extractions including intensity, texture and morphology. Overall, the SHG framework segments collagen regions and computes textural features specifically at collagen regions. Furthermore, the H&E framework segments nuclei and computes nuclei morphology and textural features. These features were used in L1-regularized logistic regression to construct classification models to discriminate normal vs DCIS regions; and to distinguish regions from DCIS patients with vs. without local recurrences. Results: In first experiment, we performed L1-regularized logistic regression to construct a classification model to discriminate normal vs DCIS regions. Our results suggest that using only SHG collagen features, this logistic model selected 19 significant features to build a classification model that achieved area under curve (AUC) 90% and accuracy 83% using 5-Fold cross validation. When H&E nuclei features are used, the logistic model selected 88 significant features and achieved AUC 91% and accuracy 86%. By combined both SHG and H&E features, the model achieved classification AUC 93% and accuracy 88%. By using L1-regularized logistic model with combined significant SHG and H&E features, we achieved AUC 59% with an accuracy of 61% for DCIS and recurrent DCIS regions. Conclusions: Our study suggests that SHG and nuclear morphology features extracted from H&E can improve the classification of normal and DCIS regions. Overall, these results suggest that second harmonic generation and H&E nuclear morphology analysis could aid in the assessment of prognosis and risk of progression to invasive breast cancer. Citation Format: Park CC, Irshad H, Ziaee S, Martin-Tuite P, Habel L, Weaver VM, Schnitt SJ, Beck AH. Second harmonic generation in combination with nuclear morphometry in the evaluation of DCIS [abstract]. In: Proceedings of the 2017 San Antonio Breast Cancer Symposium; 2017 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2018;78(4 Suppl):Abstract nr P5-02-02.