Endocrine Abstracts

Background: Radiomic analysis has been an emerging diagnostic tool in tumor classification and treatment prediction with a growing volume of radiomics research studying neuroendocrine tumors (NETs). Current reviews have highlighted numerous issues in radiomics research leading to the development of the radiomics quality scoring (RQS) system. Literature summarizing radiomics work in the NET space has concentrated on gastroenteropancreatic NETs. This work will provide a review of the topic of radiomics, summarize the radiomics research studying non GEP-NET subtypes, and provide an analysis of radiomics workflows using RQS metrics.

Methods: A systematic literature review was conducted by extracting articles from PubMed, Scopus, and Embase database using unique search terms and boolean operators. Inclusion criteria (1) non GEP-NET and (2) radiomics workflow were used to extract articles of interest. Once the unique articles were identified the RQS questionnaire was used to evaluate the RQS score based on the radiomics workflow used in the study.

Results: In total, there were 352 unique articles and 22 were included for analysis: 10 lung NETs, 9 pituitary NETs, 2 thyroid NETs, and 1 adrenal NET. The RQS score across all articles averaged 33.96%. The RQS average of pituitary, lung, thyroid, and adrenal studies were 33.33%, 35.19%, 30.56%, and 41.67% respectively. Most articles found highly conclusive results in determining tumor subtype and predicting treatment outcomes, while a select few focused on auxiliary objectives such as developing annotated radiomic datasets and reproducibility of radiomic feature extractions. One of the highest RQS scoring articles (47.22%) was able to successfully implement an MRI-based radiomics workflow which was able to predict treatment response in individuals with prolactinomas (Pituitary NETs), achieving high AUC values across training, validation, and external multi validation cohorts (AUCs 0.96, 0.92, and 0.92 respectively).

Conclusions: While research in radiomic analysis of non GEP-NETs has not been as extensive as those in GEP-NET, promising models have been developed. Future radiomic research in non GEP-NETs should focus on creating large multicenter prospective studies, implementing multimodal imaging strategies, using standardized segmentation/extraction techniques, and emphasizing external model validation to improve the quality of the resultant studies.

Abstract ID #33190