In:
Journal of Clinical Oncology, American Society of Clinical Oncology (ASCO), Vol. 41, No. 16_suppl ( 2023-06-01), p. e21217-e21217
Abstract:
e21217 Background: Radiomics can predict diagnosis, metastasis, actionable mutations and treatment response in NSCLC patients by analyzing the heterogeneity of tumors and its surrounding tissues from medical images. In this abstract, machine-learning models based on radiomic features, in patients with NSCLC, were established and evaluated. Methods: Patients with NSCLC and treated with ICIs were selected. Main tumor and peri-tumoral space were segmented on chest CT scans with contrast at the start of immunotherapy by four clinicians. Among 255 radiomic features were extracted using LIFEx software (IMIV/CEA, Orsay, France), the top 30 features with the highest Fleiss’ kappa coefficient were chosen. The Random Forest (RF) algorithm with mixed effects was utilized to develop models. We divided data into two groups as a training set (70%) and a validation set (30%) and conducted bootstrapping to evaluate the efficiency of the models. The performance of the models was determined by calculating the sensitivity, specificity, positive predictive values (PPV), and negative predictive values (NPV). Durable disease control, progression, clinical progression, EGFR mutations, bone metastasis was evaluated by each model. Durable disease control was defined as no progression of diseases for 24 weeks or more after initiation of ICIs according to RECIST 1.1. Progression was defined by RECIST 1.1 on the first follow-up CT. Clinical progression was defined when treatment was discontinued due to disease progression. Results: Total 102 patients were analyzed; 55 (53.9%) were female and 47 (46.1%) were male. The mean age at the start of ICI treatment was 65.6 [range: 22-89] . The multi-reader radiomics-based models predicts durable disease control, progression, clinical progression, EGFR mutation, and bone metastasis with a sensitivity of 0.700, 0.714, 0.286, 0.909, and 0.810, and specificity of 0.417, 0.444, 0.778, 0.200, and 0.222 respectively. The statistical values of the models are shown in the Table. Conclusions: The machine-learning models grounded on radiomics features has limited accuracy to prognosticate ICIs treatment outcome, EGFR mutations, and distant bone metastasis. Further studies with larger sample sizes are warranted. [Table: see text]
Type of Medium:
Online Resource
ISSN:
0732-183X
,
1527-7755
DOI:
10.1200/JCO.2023.41.16_suppl.e21217
Language:
English
Publisher:
American Society of Clinical Oncology (ASCO)
Publication Date:
2023
detail.hit.zdb_id:
2005181-5
