Abstract 4315: Exploring the relationship of SSTR2 immunohistochemical and textural features to neuroendocrine tumor grade

Abstract

Abstract Due to the heterogeneous nature of neuroendocrine tumors, concrete management poses adversity. A pathologic grading system defined by the World Health Organization (WHO) scaled 1-3 offers a means to guide the course of treatment. While Ki-67 is considered a gold standard tumor grading marker, Somatostatin Receptor 2 (SSTR2) has been strongly associated with low-grade neuroendocrine tumors, as it mediates the effects of somatostatin, an anti-cell-proliferation hormone. This study aims to focus on only SSTR2 and its stratified relationship with tumor grade to perform a similar role as Ki-67, given previous association with low-grade tumors. The main purpose of this study is to extract a variety of features from tumor histology images including SSTR2 expression, as well as textural features. Textural analysis offers an exhaustive method to extract morphologic characteristics from images and explore patterns unrecognizable to the human eye. This study will be accomplished through 3 objectives: first, to explore the relationship of SSTR2 expressive features to tumor grade. Second, to create a classification task to determine if machine learning models are able to accurately predict tumor grade directly from values of SSTR2 expression. The final objective of this study is to incorporate textural features into the classification task, and determine whether adding textural features to the classification dataset will improve the ML models’ ability to predict tumor grade. TMAs were stained for SSTR2, then scanned and analyzed to determine various quantifiable values: percent SSTR2, percent 1+ cells, percent 2+ cells, and percent 3+ cells, and H-score. From the data analysis, percent SSTR2 was the only value with nonconcurrent thresholds per grade: 51-98% for grade 1, 24-47% for grade 2, and .03-20% for grade 3. Textural features local binary pattern (LBP) as well as Grey-level co-occurrence matrix (GLCM) were extracted from the tumor images as well. Four separate machine learning algorithms (logistic regression, random forest, naive bayes, and support vector machine) were run on this collected dataset in 3 separate combinations: SSTR2 alone, Textural features alone, and a combination of SSTR2 and textural features. The dataset with the highest average classification metrics was SSTR2 expression alone, with the random forest classifier yielding 100 percent precision and accuracy. Two of the machine learning models improved in precision and accuracy when textural features were added to the SSTR2 dataset. The lowest-performing dataset was the textural features dataset, due to a heavy class imbalance that impacted the performance of the ML models. The results of this study depict a highly stratified relationship between SSTR2 percentage and tumor grade, and that adding textural features to the SSTR2 dataset will increase grade-classification abilities. Citation Format: Leah Sherbansky, Patrick Savickas, Sharwari Phanse, Jack Casey, Julie Feldstein. Exploring the relationship of SSTR2 immunohistochemical and textural features to neuroendocrine tumor grade. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 4315.