Tumor Classification

"Recent studies show that AI can improve how patients are classified based on their tumor’s characteristics. For instance, researchers developed a DL model that accurately predicts EGFR and KRAS mutations—genetic alterations commonly associated with aggressive forms of the disease—in lung cancer from routine histology images, potentially eliminating the need for often costly conventional genetic sequencing (1040)Zhao Y, et al. (2025) Lancet Oncol, 26: 136..\n\nIn medulloblastoma, the most common malignant brain tumor in children, accurate molecular classification of tumors plays a critical role in guiding treatment decisions but typically requires invasive procedures and time-consuming molecular tests. In a recent study, researchers showed that ML algorithms trained on gene expression data classified medulloblastoma with high accuracy, including those with poor prognosis and high metastasis rates (1041)Hourfar H, et al. (2025) Clin Oncol (R Coll Radiol), 40: 103789..\n\nIn another study, researchers developed an AI model that used spatial transcriptomics to automate histologic and molecular diagnosis of central nervous system tumors. The model accurately predicted tissue histology and tumor subtypes with more than 89 percent accuracy, demonstrating its potential to precisely identify and diagnose subtypes, even when the tissue sample is sparse or unsuitable for conventional molecular testing (1042)Ritter M, et al. (2025) Nat Cancer, 6: 292.. Beyond a single cancer type, a pan-cancer classification model has also been developed, which differentiated more than 170 tumor types across all organ sites. Achieving over 97 percent precision across diverse sample types, this approach shows how AI can expand the use of DNA methylation–based tumor classification beyond brain tumors and support diagnosis across a broad range of cancers (1043)Yuan D, et al. (2025) Nat Cancer.."