What it is
The I3LUNG study enrolled 2,396 patients to test whether AI can guide immunotherapy selection in non-small cell lung cancer, integrating clinical and blood data, CT images, digital pathology, and genomics into early-fusion and intermediate-fusion models. Machine-learning and deep-learning models using clinical-and-blood data alone reached an area under the curve of up to 0.77 in the test set and significantly surpassed PD-L1, ECOG performance status, neutrophil-to-lymphocyte ratio, LDH, and the Lung Immune Prognostic Index. In external validation the AUC dropped to a range of 0.55 to 0.72, which the authors attribute to population differences, and in a usability study both expert and nonexpert physicians improved their predictions using the explainable AI tool.
Why it matters
Immunotherapy selection in NSCLC still rests on imperfect PD-L1 and clinical scores, so a tool that outperforms those markers and, importantly, measurably improves physicians' own predictions addresses a real decision-making gap. The authors report that multimodal integration did not clearly beat the clinical-and-blood-only model, a sober finding that keeps the practical claim tied to the simpler, more deployable data.
Underlined numbers link to their source. Every metric and quoted figure is listed under Sources and data below.
Filed underRadiomics and Machine Learning in Medical Imaging, Cancer Immunotherapy and Biomarkers, Lung Cancer Diagnosis and Treatment