Radiomics in early-stage lung adenocarcinoma: a prediction tool for tumour immune microenvironments

• Published in: Canadian Journal of Surgery 2021-12, Vol.64, p.S118-S118
• Main Authors: Kaafarani, M, Huynh, C, Chouiali, F, Muthukrishnan, N, Maleki, F, Ovens, K, Gold, M, Sorin, M, Falutz, R, Rayes, R, ghani, R, Spicer, J
• Format: Article
• Language: English
• Subjects: Artificial intelligence; Deep learning; Lung cancer; Radiomics
• ISSN: 0008-428X; 1488-2310

Background: A reliable noninvasive biomarker of response to immunotherapy in patients with lung cancer remains elusive and represents a major unmet need. Tumoural and peritumoural immune infiltration is known to associate with response to immune checkpoint inhibitors. Radiomic analysis of pretreatment medical imaging converts medical images into high-dimensional quantitative data amenable to deep learning artificial intelligence analytical algorithms. This study aims to create a radiomic-based signature that predicts immune infiltration patterns of early-stage lung cancer to assist with clinical decision-making and tailor cancer-directed therapy accordingly. Supervised machine learning (ML) and several deep learning (DL) architectures will be employed to create a predictive model of the lung and tumour immune microenvironments (TIME) of corresponding regions of interest. Methods: A cohort of 110 patients who underwent surgical resection for lung adenocarcinoma (LUAD) between 2014 and 2020 was identified. The cohort consisted of 60% female patients, diagnosed at a median age of 67 (range 61-74) years. At the time of diagnosis, 73% of patients had stage I, 19% stage II and 8% stage III disease. Preoperative computed tomographic scans were collected, deidentified and contoured for tumour core, tumour-lung interface and normal adjacent lung volumes from the same lobe. Immune infiltrates from corresponding regions of interest from these same patients were assessed by multiplex immunofluorescence microscopy on a tissue microarray constructed for purpose. Antibodies against CD8, CD4, FOxP3, CD68, H3Cit, NE and DAPI were employed, and multiplexing was accomplished with the OPAL system. Results: Radiomic signatures have promise to provide a reliable noninvasive way to predict the tumour immune microenvironment in lung cancer. If successful, these strategies may improve treatment assignment and help improve clinical outcomes.