Abstract PO-006: Artificial intelligence aided interpretation of ALK fluorescent in situ hybridization for lung cancer: An algorithm development based on 10-year-annotated quality control files in central laboratory

Abstract

Abstract Molecular testing became the most important part of lung cancer diagnostics. Among them, an interpretation of fluorescent in situ hybridization (FISH) is challenging for the routine clinical samples with various amounts and quality. In this study, we trained a variant of DenseNet, a convolutional neural network model with the highest performance in image classification and suitable for quality assessment. Train dataset is composed of the arbitrarily classified dataset made and assessed by a professional pathologist during routine molecular test using whole-slide tissue images and molecular profiles including DNA density and the signal intensity of FISH obtained from 15,000 tests from lung cancer biopsies and resections, which were sent to the central laboratory for molecular studies from 45 hospitals nationwide. The performance of our platform is tested and compared to the quality assessment dataset made from each a general pathologist and a technician. The performance of our method for classifying DNA concentration and the signal intensity of FISH is comparable to that of a general pathologist and technicians. we are under improving its average area under the curve (AUC) over 0.85. This result suggests that a deep learning model based on a convolutional neural network can assist pathologists and pathology technician in rigorous assessing the quality of tissue morphology and signal of FISH for tissue samples with various molecular quality before or during the molecular test and can be applied to quality control of lung cancer molecular testing in a central laboratory. Citation Format: Tae-Jung Kim, Ho Chul Kang, Yonggeun Lee, Seokman Seo, Donghwan Kim, Sungjin Cho. Artificial intelligence aided interpretation of ALK fluorescent in situ hybridization for lung cancer: An algorithm development based on 10-year-annotated quality control files in central laboratory [abstract]. In: Proceedings of the AACR Virtual Special Conference on Artificial Intelligence, Diagnosis, and Imaging; 2021 Jan 13-14. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(5_Suppl):Abstract nr PO-006.