Artificial intelligence, complex system analysis and simulation modeling for early detection of lung cancer

Abstract

Listen

Translate article

18127 Background: Early detection (ED) of non-small cell lung cancer (LC) is the key to the cancer problem solving and practically regulates effectiveness of surgery. We examined immunologic factors associated with early LC. Methods: In trial (1987–2006) the data of consecutive and monitored 94 LC patients (LCP) after lobectomies (age=58.3±0.8 years; tumor size=2.32±0.07 cm; m=80, f=14) with pathologic stage IA (T1N0M0) (squamous=36, adenocarcinoma=53, large cell=5; G1=32, G2=38, G3=24), 282 patients with lung non-malignant pathology (NMP) (m=188, f=94; pneumonectomies=5, lobectomies=179, segmentectomies=98; non-malignant tumours=100; abscess=112; tuberculoma=70) and 120 healthy donors (HD) (m=69, f=51) were reviewed. Variables selected for ED study were input levels of 64 immunity blood parameters, sex, age. Representativeness of samplings was reached by means of randomisation based on unrepeated random selection. Logistic regression, clustering, discriminant analysis, neural networks computing, structural equation modeling, Monte Carlo and bootstrap simulation were used to determine any significant regularity. Results: Logistic regression modeling displayed that ED of LC significantly depended on: CD8, CD4, CD16, neutrophils, monocytes (P=0.002–0.041). Neural networks computing, genetic algorithm selection and bootstrap simulation discovered relationships between early LC and blood lymphocytes (rank=1), T-lymphocytes (2), CD4 (3), CD16 (4), B- lymphocytes (5), CD1 (6), NST-A2 (7), CD8 (8), NST-SP (9), NST-A1 (10), monocytes (11), eosinophils (12). Conclusions: Correct recognition of LCP with stage IA (T1N0M0) from NMP and HD was 81.7% by logistic regression (odds ratio=6.39), 84.1% by discriminant analysis and 99.2% by neural networks computing (area under ROC curve=0.99; error=0.063). No significant financial relationships to disclose.