A COMPARATIVE STUDY OF AI MODELS FOR LUNG CANCER PREDICTION AND CLINICAL DECISION SUPPORT

Abstract

Artificial-intelligence (AI) research on lung cancer in 2026 shows rapid growth across lightweight convolutional networks, CNN–transformer hybrids, capsule-inspired models, segmentation-driven pipelines, radiomics, multimodal fusion and reader-assistance systems. Model selection is now more than accuracy maximisation: a high closed-set score does not automatically make a model the best choice for low-resource deployment, indeterminate-nodule risk estimation, radiologist assistance or clinically interpretable decision support. This study comparatively analyses 20 peer-reviewed studies published in 2026, examining reported predictive performance together with computational efficiency, explainability, imaging modality, task definition and direct clinical comparison. Public-dataset classifiers frequently report accuracy above 96%, with several attention-based or compact architectures reaching 98–100% on selected hold-out sets. Clinically anchored studies present different evidence: a radiologist reader study increased cancer-detection AUC from 0.652 to 0.761 with AI support and reduced interpretation time by 12.9%; PET/CT malignancy modelling achieved an AUC of 0.78, non-inferior to the Herder model (AUC 0.73); and CT radiomics reached an AUC of 0.941, statistically comparable to expert radiologists. Multimodal fusion also produced clinically useful risk estimates, including a test AUC of 0.823 for stacked clinical–radiomics–deep-learning fusion. The comparative evidence supports task-specific model selection: compact networks for constrained hardware, attention hybrids for high-dimensional image representation, radiomics or fusion for risk stratification, and reader-validated systems for workflow integration.