Pathologists from nine centers were trained to use digital image analysis (QuPath) to quantify cellular components in NSCLC whole slide images and assess tumor cell percentage. Across two ring trials, interobserver reliability remained poor despite feedback and training, with ICC values improving only slightly from 0.09 to 0.24. Digital analysis tended to yield higher tumor percentage estimates in cases close to the 20% threshold commonly used for molecular testing, while human errors decreased from 5.6% to 1.25% after feedback. Most variability was attributed to subjective tasks such as image annotation, indicating that future artificial intelligence–based tools may help improve reproducibility and standardization.

Carretero-Barrio, I., Pijuan, L., Illarramendi, A., Curto, D., López-Ríos, F., Estébanez-Gallo, Á., Castellvi, J., Granados-Aparici, S., Compañ-Quilis, D., Noguera, R., Esteban-Rodríguez, I., Sánchez-Güerri, I., Ramos-Guerra, A. D., Ortuño, J. E., Garrido, P., Ledesma-Carbayo, M. J., Benito, A., & Palacios, J. (2024). Concordance in the estimation of tumor percentage in non-small cell lung cancer using digital pathology. Scientific reports, 14(1), 24163.